M5(design): generic modbus_device + JSON readings; YAML profiles; Accordion config; manual CLI/MQTT testing

Rework the M5 design doc data model and add manual-testing surface:

- Two-layer model: rename energy_meters/energy_readings -> generic
  modbus_device + modbus_reading; readings become a JSON payload blob
  instead of a wide fixed-column table (aggregation via SQLite
  json_extract, generated-column index as a future lever).
- Protocol knowledge moves to read-only YAML profiles (data+functions,
  no OOP inheritance); deployment/config (friendly_name, unit_id, host)
  stays on the device row; multiple devices share one profile.
- UUID = stable internal identity, also anchors HA discovery unique_id
  (Z2M rename model); device metadata for HA derived from the profile.
- Naming scheme C: modbus_* / /api/modbus/devices at the data+API layer,
  'Energy' as the first domain view in the frontend.
- New task M5-T01B: ConfigPage Accordion sectioning (avoids a second
  in-page side nav conflicting with the T01 app sidebar).
- Manual testing: read-only modbus_cli 'probe' subcommand (FC03/04 only,
  no register writes); mqtt/test now publishes a test message viewable
  in MQTT Explorer; no MQTT CLI (manual verification via UI + HA).

CLAUDE.md

@@ -36,14 +36,29 @@
 
 - **默认逐步**：给一个 milestone 文档，按其中原子任务**一步一步**实现。
 - **(a) 只实现一步**：用户说"只实现一步 / 这一个任务"时，**只做那一个任务卡**，跑完校验闸门后停下，等用户确认，不要顺手往下做。
-- **(b) 完成整个 milestone**：仅当用户在提示词里**显式要求启用 sub-agent 并指定模型**时，才用指定模型起 implementer sub-agent，按任务依赖顺序跑完整条链。
+- **(b) 完成整个 milestone**：仅当用户在提示词里**显式要求启用 sub-agent**时，才起 implementer / reviewer / fixer sub-agent（模型用下方**『默认模型档位』**，用户人工指定则覆盖），按任务依赖顺序跑完整条链。
-- **Sub-agent 纪律**：只在用户显式要求时才 spawn sub-agent；单步/小改动在主线内联完成。起 sub-agent 时用用户**指定的模型**（Agent 工具的 `model` 覆盖）。
+- **Sub-agent 纪律**：只在用户显式要求时才 spawn sub-agent；单步/小改动在主线内联完成。起 sub-agent 时按下方**『默认模型档位』**选择模型（用户人工指定则以人工指定为准），用 Agent 工具的 `model` 字段落实。
 
-### 角色（Orchestrator → Implementer → Reviewer）
+### 默认模型档位（实现模式 sub-agent；可被人工指定覆盖）
+
+起 implementer / reviewer / fixer sub-agent 时，**默认**用下列模型档位，无需用户每次人工指定：
+
+| 角色 | 默认模型 | 推理档位 |
+| --- | --- | --- |
+| **Implementer** | **Sonnet** | high reasoning effort |
+| **Fixer**（返工） | **Sonnet** | high reasoning effort |
+| **Reviewer** | **Opus** | extra-high reasoning effort |
+
+- **人工指定覆盖**：若用户在提示词里**显式指定了其他模型**（针对任一角色），则**以用户人工指定为准**，覆盖上述默认。
+- 用 Agent 工具的 `model` 字段落实模型选择；该字段当前仅支持 `sonnet` / `opus` / `haiku` / `fable`。
+- **推理档位说明**：Agent 工具未暴露独立的 reasoning-effort 旋钮，"high / extra-high reasoning" 通过 spawn prompt 里的显式指令传达（要求 implementer/fixer 动手前充分推理边界条件；要求 reviewer 以对抗性外部审计心态最高强度复核）。若所在 harness 提供真正的 effort 设置，则一并按此档位设置。
+
+### 角色（Orchestrator → Implementer → Reviewer → Fixer）
 
 - 我（主线）= **Orchestrator**：挑依赖已满足的下一个任务、派发、转述结果、维护任务 `Status`。
-- **Implementer**（便宜模型，用户指定）：一次一个任务，严格按任务卡，不扩范围。
+- **Implementer**（默认 **Sonnet**，high reasoning；见上方『默认模型档位』）：一次一个任务，严格按任务卡，不扩范围。
-- **Reviewer**（强模型，用户指定）：实现完成后起 Reviewer sub-agent，按任务卡 `Acceptance criteria` + `Reviewer checklist` 复核、**独立重跑校验闸门**，驱动 implementer 返工直到本轮 PASS。
+- **Reviewer**（默认 **Opus**，extra-high reasoning）：实现完成后起 Reviewer sub-agent，按任务卡 `Acceptance criteria` + `Reviewer checklist` 复核、**独立重跑校验闸门**，驱动返工直到本轮 PASS。
+- **Fixer**（默认 **Sonnet**，high reasoning）：按 reviewer 的编号返工清单返工；**每轮返工起一个干净的 Fixer**（与首次实现的 Implementer 分开冷启动），先读对应 `review-notes/<task>-review-<n>.md` 再改。
 
 #### Reviewer 盲审纪律（M1 教训）
 
@@ -101,7 +116,9 @@ python scripts/export_openapi.py && git diff --exit-code openapi/   # 改了路
 ## Commit 规范（重点）
 
 ### 分支
-- 每个 milestone/feature 一个分支（如 `feature/m1-db-consolidation`），**不在 `main` 上直接提交**。
+- **本仓库是个人单用户项目：默认直接在 `main` 上开发**，不强制 feature 分支，**直接提交并 push 到 `main` 是允许的（无需开 PR）**。
+- 仍保持**每个任务一个干净 commit**（message 前缀任务/里程碑 ID）。改动较大想隔离时可临时开分支，用完**快进合并**回 `main`（保持线性历史），非必需。
+- 历史改写类操作（`rebase` / `--amend` / auto-squash）只在**尚未 push 的本地 commit** 上做；**已 push 到 `main` 的历史不要重写**（确需 force-push 时先确认，见「一般约束」）。
 
 ### 一轮实现完成（用户确认「实现完成」后）
 - 准备好**这一轮的 commit message** 并提交，作为本轮的 **base commit**。
@@ -130,7 +147,8 @@ python scripts/export_openapi.py && git diff --exit-code openapi/   # 改了路
   - autosquash **改写历史**：仅在 push / 开 PR **之前**做。若该分支已 push，需要 force-push——属对外操作，**先取得用户确认再做**。
 
 ### 一般约束
-- commit / push 只在用户要求时进行；push、force-push、开/改 PR 等对外操作先确认。
+- **个人单用户仓库：直接 commit 并 push 到 `main` 已获授权**——在校验闸门全绿、一轮工作完成时即可提交 / 推送，无需逐次征求同意。
+- 仍需**先取得用户确认**的操作：**force-push / 改写已推送历史**，以及**打 tag**（会触发镜像 CI / 对外发布；且打 tag 前须按下方「发版前置走查」真跑一次 `docker build`）。
 
 ## 发版前置走查（打 tag 前必做）
 

README.md

@@ -6,7 +6,7 @@
 
 - FastAPI Web 应用（React SPA 前端 + JSON API）
 - SQLite + SQLAlchemy + Alembic 的单库结构
-- username/password + server-side session 鉴权
+- username/password + server-side session 鉴权（含登录加固，见下文）
 - runtime config 页面与 app DB 持久化
 - public IPv4 monitor、历史持久化与定时检查
 - SMTP 配置、测试发信与 public IPv4 changed 邮件通知
@@ -228,6 +228,59 @@ React SPA 主要页面路由（客户端路由，均由 FastAPI fallback 到 `in
 
 无论是本地 `host:port` 还是反向代理后的域名访问，登录成功后进入 SPA 首页（`/`）。
 
+## M4 登录加固
+
+M4 在基础鉴权之上叠加了三层防御，详细说明见 [`docs/auth.md`](./docs/auth.md)。
+
+### 防爆破 / 指数退避
+
+登录失败超过 3 次后进入指数退避（`wait = min(900s, 1s × 2^(failures-3))`），期间请求返回 `429 Too Many Requests`（含 `Retry-After` 响应头）；成功登录后自动清零。退避按 **client IP** 与 **username** 双键取较大值，不会因此永久锁定账号（只是延迟，不是封号）。
+
+- 全局开关：`AUTH_LOGIN_THROTTLE_ENABLED`（CONFIG_FIELDS，默认 `true`）
+- 反代后需要 `AUTH_TRUST_FORWARDED_FOR=true` 才会读 `X-Forwarded-For`（`.env` 部署级配置，默认 `false`）
+
+### CLI 逃生通道
+
+拿到服务器 CLI 权限时，可以在**不依赖任何已存凭据**（无需密码、恢复码）的情况下重置密码、解锁退避、关停 TOTP：
+
+```bash
+# 重置密码（不加 --password 则交互式输入，不回显）
+python -m scripts.admin_cli reset-password admin
+
+# 解锁退避（被 429 挡住时使用）
+python -m scripts.admin_cli unlock --all            # 清所有退避行
+python -m scripts.admin_cli unlock --ip 1.2.3.4    # 按 IP 清
+python -m scripts.admin_cli unlock --username admin # 按 username 清
+
+# TOTP 相关（需要先启用，见下文）
+python -m scripts.admin_cli disable-totp admin     # 关停 TOTP（零凭据，逃生口）
+python -m scripts.admin_cli reissue-totp admin     # 重新发放 TOTP secret（打印新 URI）
+
+# 查看用户列表
+python -m scripts.admin_cli list-admin
+```
+
+在 Docker 容器内执行时：
+
+```bash
+docker compose exec app python -m scripts.admin_cli <command>
+```
+
+### 可选 TOTP 二次验证
+
+admin 可在 React SPA 设置页（`/config`）自选启用 RFC 6238 TOTP：
+
+1. 设置页点「启用 TOTP」→ 后端生成 `otpauth://` URI，前端渲染二维码（`qrcode.react`）
+2. 用 Authenticator App（如 Google Authenticator、Authy）扫码
+3. 输入当前 6 位动态码确认 → TOTP 启用
+4. 妥善保存一次性展示的 10 个恢复码（格式 `xxxx-xxxx`）
+
+启用后，登录需要两步：密码 → 6 位动态码（或恢复码，一次性）。不启用则维持纯密码登录，行为不变。
+
+恢复码丢失时，可用 CLI 逃生：`python -m scripts.admin_cli disable-totp admin`，随后即可纯密码登录。
+
+TOTP issuer 标签（显示在 Authenticator 里）通过 `AUTH_TOTP_ISSUER` 环境变量配置（`.env` 部署级），默认回退 `app_name`。
+
 ## Config 持久化
 
 当前 config 页面不会把修改写回 `.env`。

alembic_app/env.py

@@ -6,7 +6,8 @@ from sqlalchemy import engine_from_config, pool
 from app.config import get_settings
 from app.db import Base
 from app.models.config import AppConfigEntry  # noqa: F401
-from app.models.auth import AuthSession, AuthUser  # noqa: F401
+from app.models.auth import AuthSession, AuthUser, RecoveryCode  # noqa: F401
+from app.models.auth_throttle import LoginThrottle  # noqa: F401
 from app.models.public_ip import PublicIPHistory, PublicIPState  # noqa: F401
 from app.models.location import Location  # noqa: F401
 from app.models.poo import PooRecord  # noqa: F401

alembic_app/versions/20260621_07_auth_login_throttle.py

@@ -0,0 +1,43 @@
+"""add auth_login_throttle table for exponential back-off throttling
+
+Revision ID: 20260621_07_auth_login_throttle
+Revises: 20260611_06_merge_location_poo_tables
+Create Date: 2026-06-21 00:00:00.000000
+"""
+
+from typing import Sequence, Union
+
+from alembic import op
+import sqlalchemy as sa
+
+
+revision: str = "20260621_07_auth_login_throttle"
+down_revision: Union[str, None] = "20260611_06_merge_location_poo_tables"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    op.create_table(
+        "auth_login_throttle",
+        sa.Column("id", sa.Integer(), autoincrement=True, nullable=False),
+        sa.Column("key", sa.String(length=255), nullable=False),
+        sa.Column("scope", sa.String(length=16), nullable=False),
+        sa.Column("failures", sa.Integer(), nullable=False),
+        sa.Column("first_failed_at", sa.DateTime(timezone=True), nullable=False),
+        sa.Column("last_failed_at", sa.DateTime(timezone=True), nullable=False),
+        sa.Column("next_allowed_at", sa.DateTime(timezone=True), nullable=True),
+        sa.PrimaryKeyConstraint("id"),
+        sa.UniqueConstraint("scope", "key", name="uq_auth_login_throttle_scope_key"),
+    )
+    op.create_index(
+        "ix_auth_login_throttle_scope_key",
+        "auth_login_throttle",
+        ["scope", "key"],
+        unique=False,
+    )
+
+
+def downgrade() -> None:
+    op.drop_index("ix_auth_login_throttle_scope_key", table_name="auth_login_throttle")
+    op.drop_table("auth_login_throttle")

alembic_app/versions/20260621_08_totp.py

@@ -0,0 +1,61 @@
+"""add TOTP fields to auth_users and create auth_recovery_code table
+
+Revision ID: 20260621_08_totp
+Revises: 20260621_07_auth_login_throttle
+Create Date: 2026-06-21 00:00:00.000000
+"""
+
+from typing import Sequence, Union
+
+import sqlalchemy as sa
+from alembic import op
+
+
+revision: str = "20260621_08_totp"
+down_revision: Union[str, None] = "20260621_07_auth_login_throttle"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    # Add totp_secret (nullable) to auth_users — existing rows get NULL, which is correct.
+    op.add_column("auth_users", sa.Column("totp_secret", sa.String(length=64), nullable=True))
+    # Add totp_enabled (NOT NULL) with server_default="0" so existing rows default to false.
+    op.add_column(
+        "auth_users",
+        sa.Column(
+            "totp_enabled",
+            sa.Boolean(),
+            nullable=False,
+            server_default="0",
+        ),
+    )
+
+    # Create auth_recovery_code table for one-time TOTP recovery codes.
+    op.create_table(
+        "auth_recovery_code",
+        sa.Column("id", sa.Integer(), autoincrement=True, nullable=False),
+        sa.Column("user_id", sa.Integer(), nullable=False),
+        sa.Column("code_hash", sa.String(length=255), nullable=False),
+        sa.Column("used_at", sa.DateTime(timezone=True), nullable=True),
+        sa.ForeignKeyConstraint(["user_id"], ["auth_users.id"]),
+        sa.PrimaryKeyConstraint("id"),
+    )
+    op.create_index(
+        op.f("ix_auth_recovery_code_user_id"),
+        "auth_recovery_code",
+        ["user_id"],
+        unique=False,
+    )
+
+
+def downgrade() -> None:
+    # Drop auth_recovery_code table first (has FK to auth_users).
+    op.drop_index(op.f("ix_auth_recovery_code_user_id"), table_name="auth_recovery_code")
+    op.drop_table("auth_recovery_code")
+
+    # Drop the two TOTP columns from auth_users.
+    # Use batch_alter_table for SQLite compatibility (alembic's portable column-drop path).
+    with op.batch_alter_table("auth_users") as batch_op:
+        batch_op.drop_column("totp_enabled")
+        batch_op.drop_column("totp_secret")

app/api/routes/api/session.py

@@ -2,7 +2,7 @@ from __future__ import annotations
 
 import logging
 
-from fastapi import APIRouter, Depends, HTTPException, Response, status
+from fastapi import APIRouter, Depends, HTTPException, Request, Response, status
 from sqlalchemy.orm import Session
 
 from app.api.routes.api.deps import require_csrf, require_session
@@ -14,6 +14,12 @@ from app.schemas.session import (
     SessionResponse,
     SessionUser,
 )
+from app.schemas.totp import (
+    TotpDisableRequest,
+    TotpEnableRequest,
+    TotpSetupResponse,
+    TotpStatusResponse,
+)
 from app.services.auth import (
     AuthPasswordChangeError,
     AuthenticatedSession,
@@ -22,6 +28,8 @@ from app.services.auth import (
     create_session,
     revoke_session,
 )
+from app.services import login_throttle
+from app.services import totp as totp_service
 
 logger = logging.getLogger(__name__)
 
@@ -46,9 +54,26 @@ def get_session(
     return _build_session_response(auth)
 
 
+def _get_client_ip(request: Request, *, trust_forwarded_for: bool) -> str:
+    """Extract the client IP address from the request.
+
+    When ``trust_forwarded_for`` is True (reverse-proxy deployments) the
+    left-most value from the ``X-Forwarded-For`` header is used.  Otherwise the
+    direct socket IP (``request.client.host``) is used.
+    """
+    if trust_forwarded_for:
+        xff = request.headers.get("X-Forwarded-For", "")
+        if xff:
+            return xff.split(",")[0].strip()
+    if request.client is not None:
+        return request.client.host
+    return "unknown"
+
+
 @router.post("/auth/login", response_model=SessionResponse)
 def post_login(
     body: LoginRequest,
+    request: Request,
     response: Response,
     db: Session = Depends(get_db),
     settings: Settings = Depends(get_app_settings),
@@ -58,15 +83,67 @@ def post_login(
 
     On success, sets an HttpOnly session cookie and returns the session user + CSRF token.
     On failure, returns 401 with no cookie set.
+    Repeated failures trigger exponential back-off (429 + Retry-After).
     No X-CSRF-Token required (unauthenticated endpoint).
     """
+    client_ip = _get_client_ip(request, trust_forwarded_for=settings.auth_trust_forwarded_for)
+
+    # --- Throttle check (before any password verification) ---
+    if settings.auth_login_throttle_enabled:
+        wait_seconds = login_throttle.check_and_get_wait(
+            db, ip=client_ip, username=body.username
+        )
+        if wait_seconds > 0:
+            raise HTTPException(
+                status_code=status.HTTP_429_TOO_MANY_REQUESTS,
+                detail="too many failed login attempts; please try again later",
+                headers={"Retry-After": str(wait_seconds)},
+            )
+
+    # --- Password verification ---
     user = authenticate_user(db, username=body.username, password=body.password)
     if user is None:
+        if settings.auth_login_throttle_enabled:
+            login_throttle.register_failure(db, ip=client_ip, username=body.username)
         raise HTTPException(
             status_code=status.HTTP_401_UNAUTHORIZED,
             detail="invalid username or password",
         )
 
+    # --- TOTP second-factor check (only when TOTP is enabled for the user) ---
+    if user.totp_enabled:
+        if not body.totp_code:
+            # Password correct but no TOTP code supplied: signal the front-end to
+            # prompt for the second factor.  Do NOT issue a session.
+            # Deliberate: this is a normal two-step protocol step from a legitimate
+            # user — we do NOT register a throttle failure here.  The attacker must
+            # already have the correct password to reach this branch, and counting
+            # each normal first-step as a failure would risk locking out the real
+            # admin on every login attempt.
+            raise HTTPException(
+                status_code=status.HTTP_401_UNAUTHORIZED,
+                detail={"totp_required": True},
+            )
+
+        # TOTP code supplied — verify against time-based code or a recovery code.
+        totp_ok = totp_service.verify_totp_code(user, body.totp_code)
+        if not totp_ok:
+            totp_ok = totp_service.verify_recovery_code(db, user=user, code=body.totp_code)
+
+        if not totp_ok:
+            # Second-factor failure is an active attack signal — register failure
+            # so that repeated wrong TOTP/recovery-code guesses trigger back-off.
+            if settings.auth_login_throttle_enabled:
+                login_throttle.register_failure(db, ip=client_ip, username=body.username)
+            raise HTTPException(
+                status_code=status.HTTP_401_UNAUTHORIZED,
+                detail="invalid username or password",
+            )
+
+    # --- Success: clear back-off state and issue session ---
+    if settings.auth_login_throttle_enabled:
+        login_throttle.clear(db, ip=client_ip, username=body.username)
+
     auth_session, raw_token = create_session(db, user=user, settings=settings)
     logger.info("Created API authenticated session for user '%s'", user.username)
 
@@ -139,3 +216,109 @@ def post_change_password(
 
     logger.info("Password updated for user '%s'", auth.user.username)
     return Response(status_code=status.HTTP_204_NO_CONTENT)
+
+
+# ---------------------------------------------------------------------------
+# TOTP endpoints (M4-T05)
+# ---------------------------------------------------------------------------
+
+
+@router.post("/auth/totp/setup", response_model=TotpSetupResponse)
+def post_totp_setup(
+    db: Session = Depends(get_db),
+    settings: Settings = Depends(get_app_settings),
+    auth: AuthenticatedSession = Depends(require_session),
+    _csrf: None = Depends(require_csrf),
+) -> TotpSetupResponse:
+    """
+    Generate a new pending TOTP secret, otpauth URI, and one-time recovery codes.
+
+    The secret is stored in the DB but TOTP is NOT yet enabled (totp_enabled stays
+    False until the user confirms with POST /api/auth/totp/enable).
+
+    Recovery codes are returned here as plaintext exactly once; their Argon2 hashes
+    are persisted immediately so enable only needs to flip the enabled flag.
+
+    Repeating this call replaces any prior pending secret and regenerates codes.
+
+    Requires: session cookie + X-CSRF-Token.
+    """
+    secret, otpauth_uri, recovery_codes = totp_service.setup(
+        db,
+        user=auth.user,
+        issuer=settings.effective_totp_issuer,
+    )
+    return TotpSetupResponse(
+        secret=secret,
+        otpauth_uri=otpauth_uri,
+        recovery_codes=recovery_codes,
+    )
+
+
+@router.post("/auth/totp/enable", status_code=status.HTTP_204_NO_CONTENT)
+def post_totp_enable(
+    body: TotpEnableRequest,
+    db: Session = Depends(get_db),
+    auth: AuthenticatedSession = Depends(require_session),
+    _csrf: None = Depends(require_csrf),
+) -> Response:
+    """
+    Enable TOTP by confirming with the current 6-digit code from the authenticator app.
+
+    Requires a prior call to POST /api/auth/totp/setup (so that a pending secret
+    exists).  On success, totp_enabled becomes True.
+
+    Returns 400 if the code is wrong or there is no pending secret.
+    Requires: session cookie + X-CSRF-Token.
+    """
+    ok = totp_service.enable(db, user=auth.user, code=body.code)
+    if not ok:
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail="invalid TOTP code or no pending setup",
+        )
+    return Response(status_code=status.HTTP_204_NO_CONTENT)
+
+
+@router.post("/auth/totp/disable", status_code=status.HTTP_204_NO_CONTENT)
+def post_totp_disable(
+    body: TotpDisableRequest,
+    db: Session = Depends(get_db),
+    auth: AuthenticatedSession = Depends(require_session),
+    _csrf: None = Depends(require_csrf),
+) -> Response:
+    """
+    Disable TOTP.  The caller must provide exactly one of:
+    - ``password``: the user's current login password, OR
+    - ``code``: the current 6-digit TOTP code.
+
+    On success: totp_enabled=False, totp_secret cleared, all recovery codes deleted.
+    Returns 400 if neither credential matches or neither is provided.
+    Requires: session cookie + X-CSRF-Token.
+    """
+    ok = totp_service.disable(
+        db,
+        user=auth.user,
+        password=body.password,
+        code=body.code,
+    )
+    if not ok:
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail="invalid credential; provide a valid password or TOTP code",
+        )
+    return Response(status_code=status.HTTP_204_NO_CONTENT)
+
+
+@router.get("/auth/totp", response_model=TotpStatusResponse)
+def get_totp_status(
+    auth: AuthenticatedSession = Depends(require_session),
+) -> TotpStatusResponse:
+    """
+    Return the current TOTP status for the authenticated user.
+
+    Response contains only ``{"enabled": bool}``.
+    Secret and recovery codes are NEVER returned here.
+    Requires: session cookie only (no CSRF — read-only).
+    """
+    return TotpStatusResponse(enabled=auth.user.totp_enabled)

app/config.py

@@ -37,6 +37,9 @@ class Settings(BaseSettings):
     auth_session_cookie_name: str = "home_automation_session"
     auth_session_ttl_hours: int = 12
     auth_cookie_secure_override: bool | None = True
+    auth_login_throttle_enabled: bool = True
+    auth_trust_forwarded_for: bool = False
+    auth_totp_issuer: str = ""  # defaults to app_name when empty
 
     model_config = SettingsConfigDict(
         env_file=".env",
@@ -86,6 +89,12 @@ class Settings(BaseSettings):
             return self.auth_cookie_secure_override
         return not self.is_development
 
+    @computed_field
+    @property
+    def effective_totp_issuer(self) -> str:
+        """The issuer label shown in Authenticator apps. Falls back to app_name."""
+        return self.auth_totp_issuer.strip() or self.app_name
+
 
 @lru_cache
 def get_settings() -> Settings:

app/models/auth.py

@@ -15,8 +15,12 @@ class AuthUser(Base):
     is_active: Mapped[bool] = mapped_column(Boolean, nullable=False, default=True)
     force_password_change: Mapped[bool] = mapped_column(Boolean, nullable=False, default=True)
     created_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), nullable=False)
+    # TOTP fields (Phase B, M4-T04) — nullable/false by default so existing users are unaffected
+    totp_secret: Mapped[str | None] = mapped_column(String(64), nullable=True)
+    totp_enabled: Mapped[bool] = mapped_column(Boolean, nullable=False, default=False)
 
     sessions: Mapped[list["AuthSession"]] = relationship(back_populates="user")
+    recovery_codes: Mapped[list["RecoveryCode"]] = relationship(back_populates="user")
 
 
 class AuthSession(Base):
@@ -31,3 +35,21 @@ class AuthSession(Base):
     revoked_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), nullable=True)
 
     user: Mapped[AuthUser] = relationship(back_populates="sessions")
+
+
+class RecoveryCode(Base):
+    """One-time TOTP recovery codes for AuthUser.
+
+    ``code_hash`` stores the Argon2 hash of the plaintext code (plaintext is
+    returned only at setup time and never stored).  ``used_at`` is NULL while
+    the code is still valid; set to the consumption timestamp when consumed.
+    """
+
+    __tablename__ = "auth_recovery_code"
+
+    id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
+    user_id: Mapped[int] = mapped_column(ForeignKey("auth_users.id"), nullable=False, index=True)
+    code_hash: Mapped[str] = mapped_column(String(255), nullable=False)
+    used_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), nullable=True)
+
+    user: Mapped[AuthUser] = relationship(back_populates="recovery_codes")

app/models/auth_throttle.py

@@ -0,0 +1,31 @@
+from datetime import datetime
+
+from sqlalchemy import DateTime, Index, Integer, String, UniqueConstraint
+from sqlalchemy.orm import Mapped, mapped_column
+
+from app.db import Base
+
+
+class LoginThrottle(Base):
+    """Tracks per-key login failure state for exponential back-off throttling.
+
+    ``scope`` is either ``'ip'`` or ``'user'``; ``key`` is the IP address or
+    username string respectively.  The pair ``(scope, key)`` is unique — one
+    row per tracked entity.  A row is deleted on successful login (clear).
+    """
+
+    __tablename__ = "auth_login_throttle"
+    __table_args__ = (
+        UniqueConstraint("scope", "key", name="uq_auth_login_throttle_scope_key"),
+        Index("ix_auth_login_throttle_scope_key", "scope", "key"),
+    )
+
+    id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
+    key: Mapped[str] = mapped_column(String(255), nullable=False)
+    scope: Mapped[str] = mapped_column(String(16), nullable=False)  # 'ip' | 'user'
+    failures: Mapped[int] = mapped_column(Integer, nullable=False, default=0)
+    first_failed_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), nullable=False)
+    last_failed_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), nullable=False)
+    next_allowed_at: Mapped[datetime | None] = mapped_column(
+        DateTime(timezone=True), nullable=True
+    )

app/schemas/session.py

@@ -16,6 +16,7 @@ class SessionResponse(BaseModel):
 class LoginRequest(BaseModel):
     username: str
     password: str
+    totp_code: str | None = None
 
 
 class PasswordChangeRequest(BaseModel):

app/schemas/totp.py

@@ -0,0 +1,59 @@
+"""Pydantic schemas for TOTP setup / enable / disable / status endpoints (M4-T05)."""
+from __future__ import annotations
+
+from pydantic import BaseModel
+
+
+# ---------------------------------------------------------------------------
+# Setup (POST /api/auth/totp/setup) — response
+# ---------------------------------------------------------------------------
+
+
+class TotpSetupResponse(BaseModel):
+    """Returned once after a setup call.
+
+    ``secret`` and ``recovery_codes`` are **one-time plaintext values**.
+    They are never returned again by any subsequent API call.
+    The frontend must display and instruct the user to save them before confirming.
+    """
+
+    secret: str
+    otpauth_uri: str
+    recovery_codes: list[str]
+
+
+# ---------------------------------------------------------------------------
+# Enable (POST /api/auth/totp/enable) — request
+# ---------------------------------------------------------------------------
+
+
+class TotpEnableRequest(BaseModel):
+    """The user confirms setup by providing the 6-digit TOTP code."""
+
+    code: str
+
+
+# ---------------------------------------------------------------------------
+# Disable (POST /api/auth/totp/disable) — request
+# ---------------------------------------------------------------------------
+
+
+class TotpDisableRequest(BaseModel):
+    """Disable TOTP by proving identity.
+
+    Exactly one of ``password`` or ``code`` must be provided.
+    """
+
+    password: str | None = None
+    code: str | None = None
+
+
+# ---------------------------------------------------------------------------
+# Status (GET /api/auth/totp) — response
+# ---------------------------------------------------------------------------
+
+
+class TotpStatusResponse(BaseModel):
+    """Minimal status response — never exposes secret or recovery codes."""
+
+    enabled: bool

app/services/config_page.py

@@ -49,6 +49,12 @@ CONFIG_FIELDS: tuple[ConfigField, ...] = (
         "auth_cookie_secure_override",
         "Cookie Secure Override",
     ),
+    ConfigField(
+        "Authentication",
+        "AUTH_LOGIN_THROTTLE_ENABLED",
+        "auth_login_throttle_enabled",
+        "Login Throttle Enabled",
+    ),
     ConfigField("Poo", "POO_WEBHOOK_ID", "poo_webhook_id", "Poo Webhook ID", secret=True),
     ConfigField(
         "Poo",
@@ -284,4 +290,6 @@ def _settings_payload(settings: Settings) -> dict[str, Any]:
         "auth_session_cookie_name": settings.auth_session_cookie_name,
         "auth_session_ttl_hours": settings.auth_session_ttl_hours,
         "auth_cookie_secure_override": settings.auth_cookie_secure_override,
+        "auth_login_throttle_enabled": settings.auth_login_throttle_enabled,
+        "auth_trust_forwarded_for": settings.auth_trust_forwarded_for,
     }

app/services/login_throttle.py

@@ -0,0 +1,179 @@
+"""Exponential back-off throttle service for the login endpoint.
+
+Design
+------
+Failures are tracked per (scope, key) in the ``auth_login_throttle`` table:
+
+* scope ``'ip'``   → key is the client IP address
+* scope ``'user'`` → key is the username
+
+On each login attempt the endpoint checks **both** keys and takes the **maximum**
+remaining wait (in seconds).  A wait > 0 means the caller is still inside a
+back-off window and should receive 429.
+
+Exponential formula
+-------------------
+The first ``N_FREE`` failures are free (no delay).  After that::
+
+    wait = min(BACKOFF_CAP_S, BACKOFF_BASE_S * 2 ** (failures - N_FREE))
+
+Constants (module-level so they can be tuned without touching the algorithm):
+
+* ``N_FREE``         = 3   — failures before any delay starts
+* ``BACKOFF_BASE_S`` = 1   — base wait in seconds (1st delayed attempt → 2 s)
+* ``BACKOFF_CAP_S``  = 900 — hard cap (15 min)
+
+Example delay schedule:
+  failures 1, 2, 3 → 0 s (free)
+  failures 4       → 2 s
+  failures 5       → 4 s
+  failures 6       → 8 s
+  failures 7       → 16 s
+  …
+  failures 13+     → 900 s (capped)
+
+Timezone notes
+--------------
+The model uses ``DateTime(timezone=True)``.  SQLite stores timestamps without
+timezone info, so when rows are read back the column value may be timezone-naive.
+We normalise every stored value to UTC-aware via ``_as_utc`` before comparing
+against ``datetime.now(UTC)``.
+"""
+
+from __future__ import annotations
+
+import math
+from datetime import UTC, datetime, timedelta
+
+from sqlalchemy import delete, select
+from sqlalchemy.orm import Session
+
+from app.models.auth_throttle import LoginThrottle
+
+# ---------------------------------------------------------------------------
+# Tunable constants
+# ---------------------------------------------------------------------------
+
+N_FREE: int = 3          # failures before any backoff starts
+BACKOFF_BASE_S: int = 1  # seconds — wait after the first delayed failure
+BACKOFF_CAP_S: int = 900  # seconds — maximum wait (15 min)
+
+# Internal scope labels (keep in sync with model constraint)
+_SCOPE_IP = "ip"
+_SCOPE_USER = "user"
+
+
+# ---------------------------------------------------------------------------
+# Public API
+# ---------------------------------------------------------------------------
+
+
+def check_and_get_wait(session: Session, *, ip: str, username: str) -> int:
+    """Return the number of seconds the caller must still wait, or 0 if allowed.
+
+    Checks both the IP row and the username row; returns the *larger* remaining
+    wait so that both keys must have their back-off window expire before the
+    caller is permitted to try again.
+    """
+    now = datetime.now(UTC)
+    wait_ip = _remaining_wait(session, scope=_SCOPE_IP, key=ip, now=now)
+    wait_user = _remaining_wait(session, scope=_SCOPE_USER, key=username, now=now)
+    return max(wait_ip, wait_user)
+
+
+def register_failure(session: Session, *, ip: str, username: str) -> None:
+    """Record one login failure for both the IP and the username keys.
+
+    Updates (or creates) the throttle row for each key, increments the failure
+    counter, and recomputes ``next_allowed_at`` using the exponential formula.
+    """
+    now = datetime.now(UTC)
+    _upsert_failure(session, scope=_SCOPE_IP, key=ip, now=now)
+    _upsert_failure(session, scope=_SCOPE_USER, key=username, now=now)
+    session.commit()
+
+
+def clear(session: Session, *, ip: str, username: str) -> None:
+    """Delete the throttle rows for the given IP and username (successful login).
+
+    It is safe to call this even if no rows exist for the given keys.
+    """
+    session.execute(
+        delete(LoginThrottle).where(
+            (LoginThrottle.scope == _SCOPE_IP) & (LoginThrottle.key == ip)
+            | (LoginThrottle.scope == _SCOPE_USER) & (LoginThrottle.key == username)
+        )
+    )
+    session.commit()
+
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+
+def _remaining_wait(session: Session, *, scope: str, key: str, now: datetime) -> int:
+    """Return the number of seconds that ``key`` in ``scope`` must still wait.
+
+    Returns 0 if the key has no throttle row or if the back-off window has
+    already expired.
+    """
+    row = session.scalar(
+        select(LoginThrottle)
+        .where(LoginThrottle.scope == scope, LoginThrottle.key == key)
+        .limit(1)
+    )
+    if row is None or row.next_allowed_at is None:
+        return 0
+
+    next_allowed = _as_utc(row.next_allowed_at)
+    if next_allowed <= now:
+        return 0
+
+    remaining = (next_allowed - now).total_seconds()
+    # Round up so callers are never told "0 seconds" while still blocked.
+    return math.ceil(remaining)
+
+
+def _compute_next_allowed_at(failures: int, now: datetime) -> datetime | None:
+    """Compute the earliest time the key should be allowed to try again.
+
+    Returns ``None`` for the first ``N_FREE`` failures (no delay).
+    """
+    if failures <= N_FREE:
+        return None
+    wait_s = min(BACKOFF_CAP_S, BACKOFF_BASE_S * (2 ** (failures - N_FREE)))
+    return now + timedelta(seconds=wait_s)
+
+
+def _upsert_failure(session: Session, *, scope: str, key: str, now: datetime) -> None:
+    """Create or update the throttle row for (scope, key) with one more failure."""
+    row = session.scalar(
+        select(LoginThrottle)
+        .where(LoginThrottle.scope == scope, LoginThrottle.key == key)
+        .limit(1)
+    )
+    if row is None:
+        new_failures = 1
+        row = LoginThrottle(
+            scope=scope,
+            key=key,
+            failures=new_failures,
+            first_failed_at=now,
+            last_failed_at=now,
+            next_allowed_at=_compute_next_allowed_at(new_failures, now),
+        )
+        session.add(row)
+    else:
+        row.failures += 1
+        row.last_failed_at = now
+        row.next_allowed_at = _compute_next_allowed_at(row.failures, now)
+
+
+def _as_utc(value: datetime | None) -> datetime | None:
+    """Normalise a possibly-naive datetime to UTC-aware."""
+    if value is None:
+        return None
+    if value.tzinfo is None:
+        return value.replace(tzinfo=UTC)
+    return value.astimezone(UTC)

app/services/totp.py

@@ -0,0 +1,234 @@
+"""TOTP service: secret management, recovery-code lifecycle, setup/enable/disable (M4-T05).
+
+State-machine summary
+---------------------
+DISABLED (default)
+  totp_secret=None, totp_enabled=False, no recovery codes
+
+PENDING (after setup, before enable)
+  totp_secret=<base32>, totp_enabled=False, recovery codes stored as hashes
+  - Re-calling setup replaces the pending secret and regenerates recovery codes.
+
+ENABLED (after enable)
+  totp_secret=<base32>, totp_enabled=True, recovery codes stored as hashes
+
+After disable:
+  totp_secret=None, totp_enabled=False, recovery codes deleted  → back to DISABLED
+
+
+Recovery-code timing
+--------------------
+1. setup  → generate 10 plaintext codes, persist their Argon2 hashes immediately,
+            return plaintext to caller (ONLY time).
+2. enable → verify 6-digit TOTP code; if ok, flip totp_enabled=True.
+            Recovery codes are already in the DB from step 1.
+3. disable → clear secret, clear enabled flag, delete all recovery codes.
+
+Security note: ``totp_secret`` is stored as plaintext (same posture as other
+secrets in this project: protected by file-system permissions on the SQLite
+database).  Recovery codes are stored as Argon2 hashes only.
+"""
+from __future__ import annotations
+
+import logging
+import secrets
+import string
+
+import pyotp
+from sqlalchemy import delete, select
+from sqlalchemy.orm import Session
+
+from app.models.auth import AuthUser, RecoveryCode
+from app.services.auth import hash_password, verify_password
+
+logger = logging.getLogger(__name__)
+
+# Number of recovery codes to generate per setup
+_RECOVERY_CODE_COUNT = 10
+# Alphabet for each 4-character segment: lowercase letters + digits, no ambiguous chars
+_CODE_ALPHABET = string.ascii_lowercase + string.digits
+
+
+# ---------------------------------------------------------------------------
+# Internal helpers
+# ---------------------------------------------------------------------------
+
+
+def _generate_recovery_code() -> str:
+    """Return a single recovery code in ``xxxx-xxxx`` format."""
+    segment = lambda: "".join(secrets.choice(_CODE_ALPHABET) for _ in range(4))  # noqa: E731
+    return f"{segment()}-{segment()}"
+
+
+def _verify_totp_code(secret: str, code: str) -> bool:
+    """Verify a 6-digit TOTP code against the given base-32 secret (±1 window)."""
+    return pyotp.TOTP(secret).verify(code, valid_window=1)
+
+
+def _delete_recovery_codes(db: Session, *, user_id: int) -> None:
+    """Delete ALL recovery codes for a user (called on setup re-run and disable)."""
+    db.execute(delete(RecoveryCode).where(RecoveryCode.user_id == user_id))
+
+
+# ---------------------------------------------------------------------------
+# Public API
+# ---------------------------------------------------------------------------
+
+
+def setup(
+    db: Session,
+    *,
+    user: AuthUser,
+    issuer: str,
+) -> tuple[str, str, list[str]]:
+    """Generate a new pending TOTP secret and recovery codes.
+
+    The secret is stored in ``user.totp_secret`` (``totp_enabled`` stays
+    ``False``).  Any pre-existing pending secret and recovery codes are
+    replaced atomically.
+
+    Returns
+    -------
+    (secret, otpauth_uri, plaintext_recovery_codes)
+
+    ``plaintext_recovery_codes`` are returned **once** here and MUST NOT be
+    stored or returned elsewhere.
+    """
+    # Generate new TOTP secret
+    new_secret = pyotp.random_base32()
+    otpauth_uri = pyotp.TOTP(new_secret).provisioning_uri(
+        name=user.username,
+        issuer_name=issuer,
+    )
+
+    # Generate plaintext recovery codes
+    plaintext_codes = [_generate_recovery_code() for _ in range(_RECOVERY_CODE_COUNT)]
+
+    # --- Persist atomically ---
+    # 1. Delete any old pending recovery codes (idempotent on re-setup)
+    assert user.id is not None  # mypy guard; always set after DB insertion
+    _delete_recovery_codes(db, user_id=user.id)
+
+    # 2. Update the secret (pending — totp_enabled stays False)
+    user.totp_secret = new_secret
+
+    # 3. Persist new recovery codes as Argon2 hashes
+    for plaintext in plaintext_codes:
+        db.add(RecoveryCode(user_id=user.id, code_hash=hash_password(plaintext)))
+
+    db.commit()
+    db.refresh(user)
+
+    logger.info("TOTP setup (pending) for user '%s'; %d recovery codes generated.", user.username, _RECOVERY_CODE_COUNT)
+    return new_secret, otpauth_uri, plaintext_codes
+
+
+def enable(db: Session, *, user: AuthUser, code: str) -> bool:
+    """Enable TOTP after the user proves they can generate the correct code.
+
+    Parameters
+    ----------
+    code:
+        The current 6-digit TOTP code from the user's authenticator app.
+
+    Returns
+    -------
+    True on success; False if the code is invalid or no secret is pending.
+    """
+    if not user.totp_secret:
+        logger.info("TOTP enable rejected for '%s': no pending secret.", user.username)
+        return False
+
+    if not _verify_totp_code(user.totp_secret, code):
+        logger.info("TOTP enable rejected for '%s': wrong code.", user.username)
+        return False
+
+    user.totp_enabled = True
+    db.commit()
+    db.refresh(user)
+    logger.info("TOTP enabled for user '%s'.", user.username)
+    return True
+
+
+def disable(
+    db: Session,
+    *,
+    user: AuthUser,
+    password: str | None = None,
+    code: str | None = None,
+) -> bool:
+    """Disable TOTP.  Caller must supply exactly one of ``password`` or ``code``.
+
+    On success: ``totp_enabled=False``, ``totp_secret=None``, all recovery
+    codes deleted.
+
+    Returns
+    -------
+    True on success; False if neither credential is valid.
+    """
+    if not password and not code:
+        logger.info("TOTP disable rejected for '%s': no credential provided.", user.username)
+        return False
+
+    if password:
+        if not verify_password(password, user.password_hash):
+            logger.info("TOTP disable rejected for '%s': wrong password.", user.username)
+            return False
+    elif code:
+        if not user.totp_secret or not _verify_totp_code(user.totp_secret, code):
+            logger.info("TOTP disable rejected for '%s': wrong TOTP code.", user.username)
+            return False
+
+    # Clear TOTP state
+    assert user.id is not None
+    _delete_recovery_codes(db, user_id=user.id)
+    user.totp_enabled = False
+    user.totp_secret = None
+    db.commit()
+    db.refresh(user)
+    logger.info("TOTP disabled for user '%s'.", user.username)
+    return True
+
+
+def verify_totp_code(user: AuthUser, code: str) -> bool:
+    """Verify a 6-digit TOTP code for a user.
+
+    Returns True if the code is valid (±1 time window), False otherwise.
+    The user must have a ``totp_secret`` set; returns False if not.
+
+    This is the public entry-point for T06 login two-factor verification.
+    """
+    if not user.totp_secret:
+        return False
+    return _verify_totp_code(user.totp_secret, code)
+
+
+def verify_recovery_code(db: Session, *, user: AuthUser, code: str) -> bool:
+    """Verify and consume a one-time recovery code.
+
+    Finds the first unused recovery code whose hash matches ``code``, marks it
+    as consumed (sets ``used_at``), and returns ``True``.  Returns ``False`` if
+    no matching unused code exists.
+
+    This function is provided for T06 (login two-factor) but lives here so the
+    TOTP service owns all recovery-code logic.
+    """
+    from datetime import UTC, datetime
+
+    unused = db.execute(
+        select(RecoveryCode).where(
+            RecoveryCode.user_id == user.id,
+            RecoveryCode.used_at.is_(None),
+        )
+    ).scalars().all()
+
+    for rc in unused:
+        if verify_password(code, rc.code_hash):
+            rc.used_at = datetime.now(UTC)
+            db.commit()
+            logger.info(
+                "Recovery code consumed for user '%s' (id=%d).", user.username, rc.id
+            )
+            return True
+
+    return False

dev-requirements.txt

@@ -78,6 +78,8 @@ pydantic-settings==2.13.1
     # via -r requirements.in
 pygments==2.20.0
     # via pytest
+pyotp==2.10.0
+    # via -r requirements.in
 pyproject-hooks==1.2.0
     # via
     #   build

docker-compose.dev.yml

@@ -20,9 +20,9 @@ services:
     build: .
     image: home-automation:dev
     container_name: home-automation-app-dev
-    # Publish on 8001 for dev. `!override` REPLACES the base ports list instead of
+    # Publish on 8002 for dev. `!override` REPLACES the base ports list instead of
     # appending to it, so the dev stack does NOT also bind the production 8881.
     ports: !override
-      - "127.0.0.1:8001:8000"
+      - "127.0.0.1:8002:8000"
     environment:
       APP_DATABASE_URL: "sqlite:////app/data/app.db"

docs/architecture-overview.md

@@ -69,12 +69,30 @@ React SPA 前端（M2 引入）。Vite + React + TypeScript + Mantine，由 Fast
 
 ### `scripts/`
 
-辅助脚本目录。当前包含 OpenAPI 导出脚本（`export_openapi.py`）与数据层辅助脚本。
+辅助脚本目录。当前包含：
+
+- `export_openapi.py`：导出 OpenAPI schema 静态产物
+- `run_migrations.py`：运行 Alembic migration
+- `app_db_adopt.py`：App DB 接管 / 初始化
+- `migrate_legacy_data.py`：一次性历史数据搬迁脚本
+- `admin_cli.py`：Admin CLI 逃生通道（M4），见下方"登录加固"说明
 
 ### `openapi/`
 
 OpenAPI schema 静态产物（`openapi.json` / `openapi.yaml`），由 `python scripts/export_openapi.py` 生成，纳入版本控制。前端 codegen 以此为契约源。
 
+## 登录加固（M4）
+
+M4 在基础 Argon2 + server-side session 鉴权之上叠加了三层防御：
+
+**防爆破 / 指数退避**：`app/services/login_throttle.py` 按 client IP 与 username 双键记失败计数，失败超过 3 次后指数增长等待时间（最长 15 分钟），`POST /api/auth/login` 在退避窗口内直接返回 `429 + Retry-After`，不执行 Argon2 验证。成功登录后清零。退避是延迟而非永久封号；全局开关 `AUTH_LOGIN_THROTTLE_ENABLED`（CONFIG_FIELDS，默认开）；反代后需设 `AUTH_TRUST_FORWARDED_FOR=true`（`.env` 部署级，默认 false）。
+
+**CLI 逃生通道**：`scripts/admin_cli.py`（入口 `python -m scripts.admin_cli`）直连本地 DB，**无需 HTTP 服务运行、无需任何已存凭据**，支持：重置密码（`reset-password`）、解锁退避（`unlock`）、关停 TOTP（`disable-totp`，零凭据最终逃生）、重新发放 TOTP secret（`reissue-totp`）、查看用户列表（`list-admin`）。CLI 只动 auth 行，不触碰用户数据表。
+
+**可选 TOTP 二次验证**：admin 可在设置页自选启用 RFC 6238 TOTP。启用后登录为两步（密码 → 6 位动态码或一次性恢复码）；不启用维持纯密码。TOTP secret 明文存库（与其他 secret 一致，靠文件权限保护）；恢复码以 Argon2 哈希存储，使用后消费（一次性）。后端使用 `pyotp`，二维码由前端 `qrcode.react` 渲染。issuer 标签由 `AUTH_TOTP_ISSUER`（`.env` 部署级）配置，默认回退 `app_name`。
+
+详细说明：[`docs/auth.md`](./auth.md)
+
 ## 当前约束
 
 - 当前数据库继续使用 SQLite

docs/auth.md

@@ -1,120 +1,229 @@
-# 基础鉴权说明
+# 鉴权说明
 
-本文档说明当前 Python 重构项目里已经落地的第一版鉴权基座。
+本文档说明当前已落地的鉴权基座（基础 session 鉴权 + M4 登录加固）。
-
-这一轮只解决：
-
-- 登录页
-- 登录 / 登出流程
-- server-side session
-- 一个最小受保护页面
-
-这一轮明确不解决：
-
-- 完整 config persistence
-- 完整 config CRUD
-- 多用户权限系统
-- OAuth / SSO / RBAC
 
 ## 当前 auth 模型
 
-- 认证方式：`username/password`
+- 认证方式：`username/password`（可选启用 TOTP 二次验证）
 - 会话方式：server-side session
 - 客户端凭据：session cookie
-- 页面形态：Jinja server-side template
 
-## 当前持久化
+## 持久化
 
-当前新增一个共享 App DB：
+所有 auth 相关数据存放在单一 App DB（`APP_DATABASE_URL`，默认 `sqlite:///./data/app.db`）中：
 
-- `APP_DATABASE_URL`
+- `auth_users`：用户表（含 TOTP 字段 `totp_secret` / `totp_enabled`）
-- 默认值：`sqlite:///./data/app.db`
+- `auth_sessions`：session token 哈希与过期时间
-
+- `auth_login_throttle`：登录失败退避状态（按 IP / username 双键）
-当前 auth 相关数据存放在这个 DB 中：
+- `auth_recovery_code`：TOTP 恢复码哈希（一次性）
-
+- `app_config`：runtime 配置持久化
-- `auth_users`
-- `auth_sessions`
-- `app_config`
-
-当前没有把 auth 数据和 `location` / `poo` DB 混放。
-
-当前这部分现在也走 Alembic 管理：
-
-- Alembic 环境：`alembic_app.ini` + `alembic_app/`
-- 初始化脚本：`python scripts/app_db_adopt.py`
-
-当前没有 legacy app DB，所以这一版脚本只负责初始化新库，不负责 legacy adoption。
-
-`app_config` 现在承接运行时配置持久化。
-
-其中：
-
-- `.env` 负责 bootstrap / fallback
-- `app_config` 表负责运行时配置覆盖
-- 登录密码仍然属于认证数据，使用 Argon2 哈希，不存进 `app_config`
 
 ## 首次启动与 bootstrap
 
-如果 auth DB 中还没有任何用户，应用启动时会要求：
+如果 auth DB 中还没有任何用户，应用启动时会使用：
 
 - `AUTH_BOOTSTRAP_USERNAME`
 - `AUTH_BOOTSTRAP_PASSWORD`
 
-并创建首个 admin 用户。
+创建首个 admin 用户。当前默认 bootstrap 值为：
-
-当前默认 bootstrap 值就是：
 
 - username: `admin`
 - password: `admin`
 
 首次登录后，系统会强制要求修改密码。
 
-如果你希望在首次启动前就覆盖默认值，可以直接设置环境变量：
+## 基础安全设计
 
-- `AUTH_BOOTSTRAP_USERNAME`
+当前这版已经落实的安全点：
-- `AUTH_BOOTSTRAP_PASSWORD`
-
-建议流程是：
-
-1. 配好 `.env`
-2. 运行 `python scripts/app_db_adopt.py`
-3. 启动应用
-4. 用 `admin / admin` 首次登录
-5. 立即修改密码
-
-## 安全设计
-
-当前这版已经落实的基础安全点：
 
 - 密码不明文存储，使用 Argon2 哈希
 - session cookie 为 `HttpOnly`
 - cookie 使用 `SameSite=Lax`
 - `Secure` cookie 在非 `development` 环境默认开启
-- 登录表单与登出表单都有基础 CSRF 校验
+- 写请求（POST/PUT/PATCH/DELETE）需携带 `X-CSRF-Token` header（SameSite=Lax + 自定义 header 纵深防御，无需 per-session token 值比对）
 - session token 为随机生成，服务端只持久化 token hash
-- session 有过期时间与显式失效机制
+- session 有过期时间（默认 12 小时）与显式失效机制
 
-## 当前受保护范围
+---
 
-当前这轮只保护了页面入口：
+## M4 登录加固
 
-- `GET /config`
+M4 在基础鉴权之上叠加了三层防御：防爆破/指数退避、CLI 逃生通道、可选 TOTP 二次验证。
-- `POST /config`
-- `POST /config/change-password`
-- `POST /logout`
 
-相关流程：
+### 1. 防爆破 / 指数退避
 
-- `GET /login`
+#### 机制
-- `POST /login`
 
-未登录访问 `/config` 时会被重定向到 `/login`。
+登录失败按指数增长延迟，目标是拖垮暴力枚举，同时不因此永久锁定账号。
 
-## 下一步不在本轮内
+**退避是延迟（429 + Retry-After），不是永久封号**——单 admin 场景下永久锁会被攻击者反向用来故意打锁，所以退避只增加等待时间，CLI 是最终逃生口。
 
-后续可以在这个基座上继续做：
+#### 双键计算
 
-- 配置页面接入
+每次登录请求同时按 **client IP** 和 **username** 各记一套失败计数，本次需等待时间 = 两者退避的**较大值**。
-- config persistence
+
-- 更细的受保护路由范围
+- 按 IP：堵单点暴力（攻击 IP 越敲越慢，合法用户换 IP 不受影响）
-- 用户初始化 / 密码轮换的更正式 runbook
+- 按 username：全局兜底（即便攻击者分布式换 IP 也有一层保护）
+
+#### 指数公式
+
+```
+N_FREE  = 3       — 前 3 次失败不延迟（免费次数）
+BASE    = 1  秒
+CAP     = 900 秒（15 分钟）
+
+wait = min(CAP, BASE × 2^(failures - N_FREE))   （failures > N_FREE 时生效）
+```
+
+延迟示意：
+
+| 累计失败次数 | 等待时间 |
+| --- | --- |
+| 1–3 | 0 秒（免费） |
+| 4 | 2 秒 |
+| 5 | 4 秒 |
+| 6 | 8 秒 |
+| 7 | 16 秒 |
+| … | … |
+| 13+ | 900 秒（封顶） |
+
+成功登录后，该 IP 和 username 的退避状态**立即清零**。
+
+#### 登录端点行为（`POST /api/auth/login`）
+
+1. 先查退避（在窗口内直接 `429`，**不验密码**，节省 Argon2 计算并防止枚举）
+2. 验密码失败 → 记一次失败（IP + username 各记），返回 `401`
+3. 密码正确但 TOTP 启用且缺少 `totp_code` → 返回 `401 {totp_required: true}`，**不记失败**（这是正常两步流程的第一步，不是攻击信号）
+4. 密码正确但 TOTP 验证失败 → 记一次失败，返回 `401`
+5. 全部通过 → 清零退避状态，发 session cookie
+
+#### 配置项
+
+| 配置项 | 类型 | 默认 | 说明 |
+| --- | --- | --- | --- |
+| `AUTH_LOGIN_THROTTLE_ENABLED` | bool（CONFIG_FIELDS） | `true` | 全局开关，关闭后整个退避机制 no-op |
+| `AUTH_TRUST_FORWARDED_FOR` | bool（`.env` 部署级） | `false` | `true` 时取 `X-Forwarded-For` 最左作为 client IP；**反代后必须显式开启**，否则反代 IP 全视为同一 IP，按 IP 退避失效 |
+
+> **反代部署注意**：`AUTH_TRUST_FORWARDED_FOR` 默认 `false`（直接用 socket IP）。部署在 nginx 等反代后面时，应设 `AUTH_TRUST_FORWARDED_FOR=true`，并确保反代正确设置 `X-Forwarded-For`。
+
+---
+
+### 2. CLI 逃生通道
+
+#### 设计原则
+
+CLI 通过 **直连本地 DB**（`get_session_local()`）工作，**无需 HTTP 服务器运行、无需任何已存凭据**（密码、恢复码均不需要）。拿到服务器 CLI 权限本身就意味着对系统有完全控制，因此这是可接受且必须存在的最终逃生口。
+
+CLI 只动 auth 相关行（`auth_users` 的密码/TOTP 字段、`auth_login_throttle`、`auth_recovery_code`），**绝不**触碰用户数据表（`location`、`poo_records`、`public_ip_state` 等）。
+
+入口：`python -m scripts.admin_cli <command>`
+
+#### 子命令
+
+| 命令 | 用途 | 逃生场景 |
+| --- | --- | --- |
+| `reset-password <username> [--password <pwd>]` | 重置密码（不带 `--password` 则交互式输入，不回显） | 忘记密码 |
+| `unlock [--all \| --ip <ip> \| --username <u>]` | 清 `auth_login_throttle` 行 | 被退避锁住（429）时解锁 |
+| `disable-totp <username>` | 关停 TOTP（清 secret + 删全部恢复码），**零凭据可执行** | 恢复码全丢也能进 |
+| `reissue-totp <username>` | 生成新 TOTP secret 并打印 `otpauth://` URI | 设备丢失、需重新注册 Authenticator |
+| `list-admin` | 列出所有用户与状态列 | 排障用 |
+
+#### 使用示例
+
+```bash
+# 重置密码（不带 --password 时交互式 prompt，不回显）
+python -m scripts.admin_cli reset-password admin
+
+# 重置密码（非交互，脚本里用）
+python -m scripts.admin_cli reset-password admin --password "newpassword"
+
+# 解锁所有退避
+python -m scripts.admin_cli unlock --all
+
+# 解锁特定 IP
+python -m scripts.admin_cli unlock --ip 1.2.3.4
+
+# 解锁特定 username
+python -m scripts.admin_cli unlock --username admin
+
+# 关停 TOTP（零凭据，最终逃生口）
+python -m scripts.admin_cli disable-totp admin
+
+# 重新发放 TOTP secret（打印新 otpauth:// URI，扫码重新注册）
+python -m scripts.admin_cli reissue-totp admin
+
+# 查看用户列表
+python -m scripts.admin_cli list-admin
+```
+
+在 Docker 容器内执行：
+
+```bash
+docker compose exec app python -m scripts.admin_cli <command>
+```
+
+#### `reissue-totp` 语义说明
+
+- 对**已启用** TOTP 的用户：新 secret **立即在登录时生效**，旧 Authenticator 生成的码立即失效；**无需**再走 web `enable` 步骤。
+- 现有恢复码**不被删除**——恢复码是独立的随机哈希值，与 TOTP secret 无密码学绑定，reissue 后恢复码依然有效（仍可用于登录）。
+- 如果需要完整清理（secret + 所有恢复码），使用 `disable-totp`。
+
+---
+
+### 3. 可选 TOTP 二次验证
+
+#### 设计
+
+- TOTP 遵循 RFC 6238，使用 `pyotp` 库。
+- 二维码在**前端**由 `qrcode.react` 渲染（后端只返回 `otpauth://` URI，不引图像依赖）。
+- `totp_secret` 明文存库（与项目其他 secret 处理一致，靠数据库文件权限保护）。
+- 恢复码以 Argon2 哈希存库，一次性（使用后标记 `used_at`）。
+
+#### 启用流程
+
+1. **setup**（`POST /api/auth/totp/setup`）：生成 pending secret，返回 secret、`otpauth://` URI、10 个明文恢复码（**仅此一次**）；此时 `totp_enabled` 仍为 `false`。
+2. **扫码**：用 Authenticator App 扫前端渲染的二维码（或手动输入 secret）。
+3. **enable**（`POST /api/auth/totp/enable`）：输入当前 6 位码确认 → `totp_enabled` 变为 `true`。
+4. 妥善保存恢复码（不会再次展示）。
+
+#### 停用流程
+
+**web 停用**（`POST /api/auth/totp/disable`）：需提供当前密码或当前 6 位 TOTP 码。成功后清 secret、删全部恢复码，恢复纯密码登录。
+
+**CLI 停用**（逃生口，恢复码全丢时）：`python -m scripts.admin_cli disable-totp admin`，零凭据，立即生效。
+
+#### 登录二步流程
+
+1. 提交 `POST /api/auth/login {username, password}` → 密码正确但 TOTP 已启用 → `401 {totp_required: true}`（不发 session）
+2. 前端切到第二屏，提交 `POST /api/auth/login {username, password, totp_code}` → 通过 → 发 session cookie
+
+`totp_code` 可以是 6 位 TOTP 动态码，也可以是 `xxxx-xxxx` 格式恢复码（命中即消费，不可复用）。
+
+#### API 端点
+
+| 端点 | 用途 |
+| --- | --- |
+| `POST /api/auth/totp/setup` | 生成 pending secret + URI + 恢复码（一次性明文返回） |
+| `POST /api/auth/totp/enable` | 带当前 6 位码确认启用 |
+| `POST /api/auth/totp/disable` | 带密码或当前码停用 |
+| `GET /api/auth/totp` | 返回当前 TOTP 状态（`{enabled: bool}`），不返回 secret/恢复码 |
+
+全部端点需要 session cookie（`GET /api/auth/totp` 不需 CSRF；其余写端点需 `X-CSRF-Token`）。
+
+#### 配置项
+
+| 配置项 | 类型 | 默认 | 说明 |
+| --- | --- | --- | --- |
+| `AUTH_TOTP_ISSUER` | str（`.env` 部署级） | 空（回退 `app_name`） | 显示在 Authenticator App 里的 issuer 标签 |
+
+---
+
+## 受保护范围
+
+当前 JSON API 端点（`/api/*`）需要 session cookie；写端点需额外携带 `X-CSRF-Token` header。
+
+裸 ingestion 端点（`/location/record`、`/poo/record` 等设备调用端点）暂未收口到 session 保护（M3 计划引入 token 鉴权）。
+
+## 下一步（不在当前范围）
+
+- M3：token 鉴权（供脚本 / 设备 / 移动端调用 API），ingestion 端点收口。

docs/design/README.md

@@ -5,6 +5,8 @@
 - [`m1-db-consolidation.md`](./m1-db-consolidation.md) — 单库化地基
 - [`m2-frontend-v2.md`](./m2-frontend-v2.md) — React SPA 前端 v2
 - [`m3-token-mobile.md`](./m3-token-mobile.md) — token 鉴权与移动端（远期）
+- [`m4-login-hardening.md`](./m4-login-hardening.md) — 登录加固（防爆破/指数退避 + CLI 逃生 + 可选 TOTP）**先做**
+- [`m5-iot-energy.md`](./m5-iot-energy.md) — IoT 集成与能耗采集（Modbus/Energy + MQTT/HA Discovery + 前端侧边栏）
 
 本文件定义**所有任务共用的格式与协作规则**，三个里程碑文档不再重复这些约定。
 

docs/design/m4-login-hardening.md

@@ -0,0 +1,309 @@
+# M4 — 登录加固（Login Hardening）
+
+> 阅读前提：先读 [`README.md`](./README.md)（协作模型、任务卡格式、校验闸门、数据安全红线）。
+> **排期：本里程碑排在 [M5](./m5-iot-energy.md) 之前。** 系统现为公网 + 仅密码 + 无限重试，是现成风险；M4 与 M5 零文件重叠，先堵这个洞再做 IoT。
+
+## 1. 目标
+
+给暴露在公网的单 admin 登录做纵深防御：
+
+1. **防爆破 / 指数退避**：失败登录按指数增长延迟，拖垮暴力枚举；成功即清零。
+2. **CLI 逃生通道**：纯命令行重置密码 / 解锁 / 关停 TOTP，**不依赖任何已存恢复凭据**——拿到服务器 CLI 权限的人本来就无所不能，所以这是可接受、且必须存在的最终逃生口（消除"锁死就彻底进不去"的死角）。
+3. **TOTP 二次验证（可选配）**：admin 可**自选启用** RFC 6238 TOTP；启用后登录需密码 + 6 位动态码；提供一次性恢复码；CLI 能在恢复码全丢时关掉 TOTP。**非强制**，不启用则维持纯密码登录。
+
+> 三段顺序：A（防爆破，紧急、可独立先发版）→ B（TOTP，可选配）→ 收尾。Phase A 自成可用增量，Phase B 失败不影响 A。
+
+## 2. 现状（实现者可据此工作，不必通读全仓库）
+
+**单 admin + Argon2**
+- `app/models/auth.py`：`AuthUser`（`id / username unique / password_hash / is_active / force_password_change / created_at`）、`AuthSession`（`token_hash / csrf_token / expires_at / revoked_at`）。单库 `Base`。
+- `app/services/auth.py`：
+  - `authenticate_user(session, *, username, password) -> AuthUser | None`：查用户→`verify_password`（Argon2）；失败只 `logger.info`，**无任何节流**。
+  - `create_session(session, *, user, settings) -> (AuthSession, raw_token)`：`secrets.token_urlsafe`，SHA256 存 `token_hash`，TTL `auth_session_ttl_hours`（默认 12h）。
+  - `change_password(...)`、`get_authenticated_session(...)`、`initialize_auth_schema(session, settings)`（**仅当无任何用户时**用 `AUTH_BOOTSTRAP_USERNAME/PASSWORD` 建初始 admin）。
+- `app/api/routes/api/session.py`：
+  - `POST /api/auth/login`（body `{username, password}`）：`authenticate_user` → None 则 401；否则 `create_session` + `set_cookie`(HttpOnly/SameSite=Lax/`auth_cookie_secure`) → 返回 `{user, csrf_token}`。
+  - `GET /api/session`（401 或 user+csrf）、`POST /api/auth/logout`（CSRF）、`POST /api/auth/password`（CSRF + 校验当前密码）。
+
+**配置**
+- `app/config.py`：`auth_bootstrap_username/password`、`auth_session_cookie_name`、`auth_session_ttl_hours`、`auth_cookie_secure_override`（computed `auth_cookie_secure`）。
+- 配置系统：扁平 KV（`CONFIG_FIELDS` 注册表 + `Settings` + `_settings_payload`），新增标量配置项前端自动渲染（见 M5 文档 §2）。
+
+**无现成 CLI**
+- `scripts/`：`run_migrations.py` / `app_db_adopt.py` / `migrate_legacy_data.py`(argparse) / `export_openapi.py`。**没有**任何 admin/密码/auth 的 CLI。
+- `pyproject.toml`：**无 `console_scripts`**。CLI 入口沿用 `python -m scripts.<mod>` 风格（与 `python -m scripts.run_migrations` 一致）。
+
+**前端登录（单步）**
+- `frontend/src/pages/LoginPage.tsx`：`POST /api/auth/login {username,password}` → 401 显示通用错误；成功存 csrf、跳转。
+- `frontend/src/auth/SessionProvider.tsx`：`GET /api/session` 引导。
+- 无"锁定/稍后再试"提示、无二步流程。
+
+**测试**
+- `tests/test_api_session.py`（凭据/cookie/登出/改密/强制改密）；`tests/conftest.py` 用 env 设 bootstrap 凭据。
+
+## 3. 目标架构
+
+### 3.1 防爆破 / 指数退避
+
+- **状态表 `auth_login_throttle`**：按 key 记失败窗口（`key / scope / failures / first_failed_at / last_failed_at / next_allowed_at`），成功即删该 key 的行。
+- **双键**：同时按 **client IP** 与 **username** 计；本次请求需等待 = 两者退避的**较大值**。
+  - 按 IP：堵单点暴力（attacker IP 越敲越慢，合法用户换 IP 不受影响）。
+  - 按 username：单 admin 的全局兜底（即便分布式换 IP 也有一层）。
+  - **DoS 权衡**：单 admin 下纯按 username 硬锁会被攻击者故意失败把你锁死，所以退避是**延迟（429 + Retry-After）而非永久锁定**，且 CLI 是最终逃生口。
+- **client IP 来源**：默认用 socket IP；`AUTH_TRUST_FORWARDED_FOR=true` 时取 `X-Forwarded-For` 最左（部署在反代后才开）。**这点必须显式配置**，否则反代后所有请求同一 IP，按 IP 退避失效。
+- **指数公式（默认，常量可后调）**：前 `N_free=3` 次不延迟；之后 `wait = min(cap, base * 2^(failures - N_free))`，`base=1s`、`cap=900s`。
+- **接入点**：`POST /api/auth/login` 开头先查退避 → 在窗口内直接 429（带 `Retry-After`），**不验密码**（省 Argon2、防枚举）；验密码失败 → 记一次失败；成功 → 清该 IP + username 的退避行。
+- **全局开关** `AUTH_LOGIN_THROTTLE_ENABLED`（CONFIG_FIELDS，默认 true）。
+
+### 3.2 CLI 逃生通道
+
+`scripts/admin_cli.py`（`python -m scripts.admin_cli <cmd>`，argparse 子命令；直接连本地 DB，无网络，复用 `get_session_local()` + 模型 + Argon2 hasher）：
+
+| 命令 | 作用 |
+| --- | --- |
+| `reset-password <username> [--password ...]` | 重置密码（不给则交互式 prompt，不回显）；逃生：忘密码 |
+| `unlock [--all | --ip <ip> | --username <u>]` | 清 `auth_login_throttle` 行；逃生：被退避锁住 |
+| `disable-totp <username>` | 关 TOTP（清 secret + 恢复码）；逃生：**恢复码全丢也能进**（Phase B 才有意义）|
+| `reissue-totp <username>` | 重新发放 TOTP secret（可选）|
+| `list-admin` | 列用户/状态（排障）|
+
+- **数据安全**：CLI 只动 auth 状态行（人工执行的 admin 动作），**绝不**触碰用户数据表（location/poo/energy 等），不 drop 有数据的表。
+
+### 3.3 TOTP 二次验证（可选配）
+
+- **库**：`pyotp`（Python 侧）；二维码在**前端**用 JS 库（`qrcode.react`）从 `otpauth://` URI 渲染——后端不引图像依赖。
+- **存储**：
+  - `AuthUser` 增列 `totp_secret`(nullable)、`totp_enabled`(bool, default false)。secret 落库明文（与现有 secret 处理一致，靠文件权限保护；文档注明）。
+  - `auth_recovery_code` 表（`user_id / code_hash / used_at`）：恢复码**哈希**存（复用 Argon2 hasher），一次性。
+- **启用流程（opt-in）**：
+  1. `POST /api/auth/totp/setup`（已登录）→ 生成 pending secret + `otpauth://` URI + 一组恢复码（**仅此一次明文返回**），尚未启用。
+  2. `POST /api/auth/totp/enable`（带一个当前 6 位码确认）→ `totp_enabled=true`，持久化恢复码哈希。
+  3. `POST /api/auth/totp/disable`（带密码或当前码）→ 关闭、清 secret/恢复码。
+- **登录二因子（单端点、无独立 challenge token，保持无状态）**：
+  - `POST /api/auth/login` body 增可选 `totp_code`。
+  - 密码通过且 `totp_enabled`：无 `totp_code` → 返回 401 + `{totp_required: true}`（**不发 session**）；有 `totp_code` → 校验 TOTP **或**恢复码（命中恢复码则消费它）→ 通过才发 session。
+  - 未启用 TOTP：维持现状单步。
+  - 退避（§3.1）对两步都生效。
+- **issuer 标签**：`AUTH_TOTP_ISSUER`（默认 `app_name`），显示在 Authenticator app 里。
+
+### 3.4 前端
+
+- **登录页两步**：`LoginPage` 在收到 `401 + totp_required` 时切到第二屏（6 位码输入，也接受恢复码），再次 `POST /api/auth/login` 带 `totp_code`。被退避（429）显示"稍后再试"（用 `Retry-After`）。
+- **设置页 TOTP 区**：未启用→「启用」走 setup（展示二维码 + 恢复码一次性、要求输入码确认）；已启用→「停用」。恢复码只在生成时展示一次，提示妥善保存。
+
+## 4. API 契约（M4 要落地的端点）
+
+> 全部 `/api`、写端点 session + CSRF 保护、JSON 进出；schema 经 `export_openapi.py` 固化。
+
+| 分组 | 端点 | 用途 |
+| --- | --- | --- |
+| 登录 | `POST /api/auth/login` | 加退避（429+Retry-After）；body 增可选 `totp_code`；启用 TOTP 且缺码→401 `{totp_required:true}` |
+| TOTP | `POST /api/auth/totp/setup` | 生成 pending secret + otpauth URI + 恢复码（一次性）|
+| TOTP | `POST /api/auth/totp/enable` | 带当前码确认启用 |
+| TOTP | `POST /api/auth/totp/disable` | 关闭 TOTP |
+| TOTP | `GET /api/auth/totp` | 返回当前 TOTP 状态（enabled 与否）|
+
+> CLI 不是 HTTP 端点。throttle 开关与 issuer 走现有 `GET/PUT /api/config`（仅加 CONFIG_FIELDS）。
+
+## 5. 已锁定决策（讨论后拍板）
+
+1. **M4 独立里程碑、排在 M5 之前**。
+2. **范围**：防爆破 + 指数退避 + CLI 重置（密码/解锁/关 TOTP）+ TOTP（**可选配，非强制**）。
+3. **退避双键（IP + username）、延迟非永久锁定、成功清零、CLI 兜底**；反代后需开 `AUTH_TRUST_FORWARDED_FOR`。
+4. **CLI 逃生通道不依赖任何已存恢复凭据**；只动 auth 行，不碰用户数据。
+5. **TOTP 单端点二因子**（login 加可选 `totp_code`，无独立 challenge token）；恢复码哈希存、一次性；secret 明文存（与现有一致）。
+6. **二维码前端渲染**（`qrcode.react`），后端只给 `otpauth://` URI + secret；Python 仅加 `pyotp`。
+7. **CLI 入口 = `python -m scripts.admin_cli`**（无 console_scripts）。
+
+> 项目定位：个人自用、单 admin——可按单用户简化，不为多租户/找回邮件等过度设计。
+
+## 6. 任务依赖图
+
+```
+Phase A（防爆破，紧急，自成可发版增量）
+  M4-T01 [schema] auth_login_throttle 表 + 模型
+   └─► M4-T02 退避 service + 接入 login（429/Retry-After/成功清零）
+   └─► M4-T03 CLI 骨架 + reset-password + unlock
+
+Phase B（TOTP，可选配）
+  M4-T04 [schema] AuthUser TOTP 字段 + auth_recovery_code 表
+   ├─► M4-T05 TOTP service + setup/enable/disable/status API
+   ├─► M4-T06 登录二因子（login 加 totp_code；依赖 T02 已改过 login）
+   └─► M4-T07 CLI disable-totp / reissue-totp
+        M4-T08 前端：两步登录 + 设置页 TOTP（依赖 T05+T06；引入 qrcode.react）
+
+收尾
+  M4-T09 文档 + OpenAPI + roadmap（依赖全部）
+```
+
+`T01` 可先开。Phase A（T01–T03）跑完即可发一版只含防爆破 + CLI 的安全增量。
+
+---
+
+## 7. 原子任务（任务卡）
+
+> 后端沿用校验闸门（`pytest`/`ruff`/改路由或 schema 则 `export_openapi` 重导出入库）。前端闸门见 §8。新增依赖（`pyotp`、`qrcode.react`）须同步 `requirements.in/.txt` 与 `frontend/package.json` 重新锁定。
+
+### M4-T01 — `auth_login_throttle` 表 + 模型 `[schema]`
+- **Status**: `done` · **Depends**: none
+- **Context**: 存按 IP / username 的失败退避状态。仅建 schema。
+- **Files**: `create app/models/auth_throttle.py`（或并入 `app/models/auth.py`）；`create alembic_app/versions/<date>_NN_auth_login_throttle.py`；`modify alembic_app/env.py`、`scripts/app_db_adopt.py`（baseline 常量）；`create tests/test_auth_throttle_model.py`
+- **Steps**: 模型 `LoginThrottle`（`id / key str / scope str('ip'|'user') / failures int / first_failed_at / last_failed_at / next_allowed_at`，`(scope,key)` unique 索引）；新 revision 建表 + 索引；更新 baseline 常量。
+- **Out of scope / 不要碰**: 不写退避逻辑（T02）；不动登录端点。
+- **Acceptance criteria**:
+  - [ ] 临时库 upgrade 到 head 含该表与唯一索引；`downgrade -1` 干净。
+  - [ ] `APP_BASELINE_REVISION` 更新；`env.py` 已 import。
+  - [ ] 校验闸门全绿。
+- **Reviewer checklist**: `(scope,key)` 唯一；链上单 head；env.py 注册模型。
+
+### M4-T02 — 退避 service + 接入登录
+- **Status**: `done` · **Depends**: M4-T01
+- **Context**: 指数退避核心 + 接 `POST /api/auth/login`。
+- **Files**: `create app/services/login_throttle.py`；`modify app/api/routes/api/session.py`、`app/services/config_page.py`(+`CONFIG_FIELDS` `AUTH_LOGIN_THROTTLE_ENABLED`)、`app/config.py`(+`auth_login_throttle_enabled`、`auth_trust_forwarded_for`)；`modify tests/test_api_session.py`、`create tests/test_login_throttle.py`
+- **Steps**:
+  1. `check_and_get_wait(session, *, ip, username) -> int`（秒；>0 表示仍在窗口）；`register_failure(...)`（更新双键 failures/next_allowed_at，按 §3.1 公式）；`clear(session, *, ip, username)`（成功清零）。
+  2. 登录端点：取 client IP（按 `auth_trust_forwarded_for` 决定是否信 XFF 最左）；**先**查退避，>0 → 429 + `Retry-After`，不验密码；验密码失败 → `register_failure` 后 401；成功 → `clear` 再发 session。
+  3. 开关关闭时整段 no-op。
+- **Out of scope / 不要碰**: 不做 TOTP（Phase B）；不改 logout/password 端点。
+- **Acceptance criteria**:
+  - [ ] 单测：连续失败后 `wait` 按指数增长且封顶；窗口内请求 429 带 `Retry-After` 且不验密码。
+  - [ ] 单测：成功登录清零；下次无延迟。
+  - [ ] 401 仍为通用文案（不泄露用户是否存在）；现有登录测试仍绿。
+  - [ ] 开关 false 时无节流。
+  - [ ] 校验闸门全绿（OpenAPI 若变重导出）。
+- **Reviewer checklist**: 退避在验密码**之前**生效；双键较大值；XFF 仅在配置开启时信任；无把密码写进日志。
+
+### M4-T03 — CLI 骨架 + reset-password + unlock
+- **Status**: `done` · **Depends**: M4-T01
+- **Context**: 逃生通道第一批命令。
+- **Files**: `create scripts/admin_cli.py`；`create tests/test_admin_cli.py`
+- **Steps**: argparse 子命令；`reset-password <username> [--password|prompt(getpass)]` 用 Argon2 hasher 重置 `password_hash`；`unlock [--all|--ip|--username]` 删 `auth_login_throttle` 行；`list-admin`；复用 `get_session_local()` + 模型；无网络。
+- **Out of scope / 不要碰**: 不动 HTTP 端点；不碰用户数据表；TOTP 命令在 T07。
+- **Acceptance criteria**:
+  - [ ] 单测：`reset-password` 后新密码可 `verify_password` 通过、旧密码失败。
+  - [ ] 单测：`unlock` 清掉指定/全部退避行。
+  - [ ] 不存在的 user 友好报错、非零退出。
+  - [ ] `grep` 确认 CLI 不含对用户数据表的删除/drop。
+  - [ ] 校验闸门全绿。
+- **Reviewer checklist**: 密码可交互输入且不回显；只动 auth 行；幂等可重跑。
+
+### M4-T04 — AuthUser TOTP 字段 + `auth_recovery_code` 表 `[schema]`
+- **Status**: `done` · **Depends**: none（与 Phase A 并行的 schema，但 T06 依赖 T02）
+- **Files**: `modify app/models/auth.py`（`AuthUser.totp_secret`/`totp_enabled`）；`create` 模型 `RecoveryCode`；`create alembic_app/versions/<date>_NN_totp.py`；`modify alembic_app/env.py`、`scripts/app_db_adopt.py`；`create tests/test_totp_models.py`
+- **Steps**: `totp_secret` nullable、`totp_enabled` bool default false；`auth_recovery_code(user_id FK, code_hash, used_at nullable)`；revision 建列 + 表；更新 baseline。
+- **Out of scope / 不要碰**: 不写 TOTP 逻辑（T05）。
+- **Acceptance criteria**:
+  - [ ] upgrade/downgrade 干净；默认 `totp_enabled=false`、`totp_secret=null`。
+  - [ ] baseline 常量更新；env.py import。
+  - [ ] 校验闸门全绿。
+- **Reviewer checklist**: 现有用户迁移后默认未启用 TOTP（不破坏现有登录）。
+
+### M4-T05 — TOTP service + setup/enable/disable/status API
+- **Status**: `done` · **Depends**: M4-T04
+- **Files**: `create app/services/totp.py`、`app/schemas/totp.py`；`modify app/api/routes/api/session.py`（或新 `auth_totp.py` 路由）、`app/config.py`(+`auth_totp_issuer`)；`modify requirements.in/.txt`(+`pyotp`)；`create tests/test_api_totp.py`
+- **Steps**: pyotp 生成 secret + `otpauth://` URI（issuer=`auth_totp_issuer`）；恢复码生成（N=10、`xxxx-xxxx`，Argon2 哈希存、明文仅 setup 返回一次）；`setup`（pending，不启用）/`enable`（带当前码确认）/`disable`（带密码或当前码）/`GET status`；全部 session+CSRF。
+- **Out of scope / 不要碰**: 不改 login 校验（T06）。
+- **Acceptance criteria**:
+  - [ ] `setup` 返回 secret+URI+恢复码（一次性）；未确认前 `totp_enabled` 仍 false。
+  - [ ] `enable` 用错码失败、对码成功；恢复码以哈希持久化。
+  - [ ] `disable` 关闭并清 secret/恢复码。
+  - [ ] schema 经 OpenAPI 固化；secret/恢复码不回显于 `GET status`。
+  - [ ] 校验闸门全绿。
+- **Reviewer checklist**: 恢复码只存哈希、明文仅一次；secret 不进 OpenAPI 示例/日志；`requirements.txt` 锁定 `pyotp`。
+
+### M4-T06 — 登录二因子（login 加 `totp_code`）
+- **Status**: `done` · **Depends**: M4-T05, M4-T02
+- **Files**: `modify app/api/routes/api/session.py`、`app/schemas/session.py`（`LoginRequest.totp_code` 可选）、`app/services/totp.py`（校验 + 消费恢复码）；`modify tests/test_api_session.py`、`tests/test_api_totp.py`
+- **Steps**: 密码通过后：若 `totp_enabled` 且无 `totp_code` → 401 `{totp_required:true}`（不发 session、退避照算）；有 `totp_code` → 校验 TOTP 或恢复码（命中恢复码置 `used_at`）→ 通过发 session；未启用 → 现状。
+- **Out of scope / 不要碰**: 不改 setup/enable（T05）；不动退避公式（T02）。
+- **Acceptance criteria**:
+  - [ ] 未启用 TOTP 的用户登录与现状完全一致。
+  - [ ] 启用后：缺码→401 totp_required；错码→401；对码→发 session；恢复码可用且一次性。
+  - [ ] 退避对二步同样生效。
+  - [ ] OpenAPI 重导出入库；校验闸门全绿。
+- **Reviewer checklist**: `totp_required` 时确未发 cookie；恢复码消费后不可复用；通用错误不泄露细节。
+
+### M4-T07 — CLI disable-totp / reissue-totp
+- **Status**: `done` · **Depends**: M4-T04（逻辑上配合 T03 的 CLI 骨架）
+- **Files**: `modify scripts/admin_cli.py`；`modify tests/test_admin_cli.py`
+- **Steps**: `disable-totp <username>`：`totp_enabled=false` + 清 secret + 删恢复码（**不需任何恢复码即可执行**）；`reissue-totp <username>`：生成新 secret（可选打印新 URI）。
+- **Out of scope / 不要碰**: 不碰用户数据表。
+- **Acceptance criteria**:
+  - [ ] 单测：启用 TOTP 的用户经 `disable-totp` 后可纯密码登录（结合 T06 行为）。
+  - [ ] 不依赖任何恢复码即可关停。
+  - [ ] 校验闸门全绿。
+- **Reviewer checklist**: 逃生不依赖已存凭据；只动 auth 行。
+
+### M4-T08 — 前端：两步登录 + 设置页 TOTP
+- **Status**: `done` · **Depends**: M4-T05, M4-T06
+- **Files**: `modify frontend/src/pages/LoginPage.tsx`；`create frontend/src/pages/.../TotpSettings.tsx`（或并入 ConfigPage）、`frontend/src/auth/totp.ts`；`modify frontend/package.json`(+`qrcode.react`，lock 同步)；`create` 对应测试
+- **Steps**: 登录页：401+`totp_required`→第二屏 6 位码（也接受恢复码）；429→"稍后再试"（读 `Retry-After`）。设置页：未启用→启用流程（二维码由 `otpauth` URI 渲染 + 恢复码一次性展示 + 输码确认）；已启用→停用。
+- **Out of scope / 不要碰**: 不改后端；不碰防爆破逻辑。
+- **Acceptance criteria**:
+  - [ ] 未启用 TOTP 的登录体验不变；启用后两步可走通；恢复码可登录。
+  - [ ] 二维码可被 Authenticator 扫描；恢复码仅展示一次。
+  - [ ] 429 有"稍后再试"提示。
+  - [ ] 前端闸门全绿。
+- **Reviewer checklist**: 全走类型化 client；secret/恢复码不落 localStorage/日志；空/错/加载态有处理。
+
+### M4-T09 — 文档 + OpenAPI + roadmap 收尾
+- **Status**: `done` · **Depends**: 全部
+- **Files**: `modify README.md`、`docs/auth.md`（防爆破 + CLI + TOTP 说明、CLI 用法）、`docs/architecture-overview.md`；`modify docs/roadmap.md`（把"TOTP 二次验证"与新"登录防爆破"从「下一阶段」毕业、加 M4 行、注明 M4 先于 M5）；`modify docs/design/README.md`（已含 m4）；`run python scripts/export_openapi.py` 提交 `openapi/`
+- **Acceptance criteria**:
+  - [ ] 文档含 CLI 逃生用法与 TOTP 启停说明；`git diff --exit-code openapi/` 无差异。
+  - [ ] 校验闸门全绿。
+- **Reviewer checklist**: roadmap 反映 M4 已落地且先于 M5；无残留旧描述。
+
+---
+
+## 8. 前端校验闸门（前端任务每次结束都要全绿）
+
+在 `frontend/` 下：`npm ci && npm run lint && npm run typecheck && npm run test && npm run build`。新增依赖 `qrcode.react` 须提交 `package.json` + `package-lock.json`。后端同任务改路由/schema 仍需根目录 `export_openapi.py` 并提交 `openapi/`。
+
+## 9. 构建上下文 / 依赖完整性
+
+- 新增 Python 依赖 `pyotp`、前端 `qrcode.react`：同步重新锁定 `requirements.txt` / `package-lock.json`，否则镜像缺包。
+- 新文件均在既有 `app/`、`scripts/`、`frontend/` 的 COPY 范围内；`scripts/admin_cli.py` 须在镜像里可 `python -m scripts.admin_cli` 调用（确认 `scripts/` 已进镜像）。
+- `tests/test_deployment.py::test_dockerfile_copy_sources_exist` 仍应通过。
+
+## 10. 人工验收 walkthrough（实现完成后）
+
+> 重点：连续输错 → 系统自动退避锁住 → CLI 解锁后恢复登录。可用 `docker compose` 起服务，CLI 在容器内或容器外跑均可。
+
+**准备**：起服务（任选其一）
+- Docker：`docker compose up -d`（应用容器名见 `docker-compose.yml`，下文记为 `<app>`）。
+- 或本地：`source .venv/bin/activate && uvicorn app.main:app --port 8000`。
+
+**1) 触发退避锁定** —— 用**错误密码**连续登录同一账号：
+```bash
+for i in $(seq 1 8); do \
+  code=$(curl -s -o /dev/null -w "%{http_code}" -X POST http://localhost:8000/api/auth/login \
+    -H 'Content-Type: application/json' -d '{"username":"admin","password":"wrong"}'); \
+  echo "尝试 $i -> HTTP $code"; done
+```
+- 预期：前几次 `401`，超过免费次数后变 `429`（被锁/退避）。单独跑一次带 `-i` 可看到 `Retry-After` 头逐次增大（指数退避）。
+- 在退避窗口内，**即使输正确密码**也应被 `429` 挡住（验证锁定确实生效）。
+
+**2) CLI 解锁**
+- 容器内：`docker compose exec <app> python -m scripts.admin_cli unlock --all`
+- 或容器外：`python -m scripts.admin_cli unlock --all`
+- 预期：打印清除的退避条目数。
+
+**3) 验证恢复** —— 立即用**正确密码**登录：
+```bash
+curl -i -X POST http://localhost:8000/api/auth/login \
+  -H 'Content-Type: application/json' -d '{"username":"admin","password":"<正确密码>"}'
+```
+- 预期：`200` 且 `Set-Cookie` 下发 session → 解锁成功。
+
+**4)（可选）其它逃生命令**
+- 重置密码：`python -m scripts.admin_cli reset-password admin`（交互输入新密码、不回显）。
+- 关停 TOTP（若已启用）：`python -m scripts.admin_cli disable-totp admin`，随后纯密码即可登录。
+
+## 11. 里程碑完成定义（DoD）
+
+- 登录失败按指数退避（429 + Retry-After），成功清零；可经 `AUTH_LOGIN_THROTTLE_ENABLED` 开关。
+- `python -m scripts.admin_cli` 能在**无任何 Web 访问、无恢复码**的情况下重置密码 / 解锁 / 关停 TOTP。
+- TOTP 可由 admin 自选启用：启用后两步登录、恢复码可用且一次性；不启用维持纯密码。
+- 后端 `pytest`/`ruff`/`export_openapi` + 前端 `lint/typecheck/test/build` 全绿且 `openapi/` 入库。
+- README/auth/architecture/roadmap 反映 M4 现实与"先于 M5"的排期。

docs/design/m5-iot-energy.md

@@ -0,0 +1,533 @@
+# M5 — IoT 集成与能耗采集（Modbus 设备采集 + MQTT/HA Discovery + 前端侧边栏）
+
+> 阅读前提：先读 [`README.md`](./README.md)（协作模型、任务卡格式、校验闸门、数据安全红线）。本里程碑建立在 M1 单库 + M2 React SPA 之上。
+> 配套参考：电表协议见 [`../references/SDM120-Modbus-Protocol.md`](../references/SDM120-Modbus-Protocol.md)。
+
+## 1. 目标
+
+给后端接入家庭 IoT 生态，并新增一条通用的 Modbus 设备采集链路（首个落地的领域是能耗）：
+
+1. **Modbus 设备采集（通用管道）**：通过 Modbus-TCP 网关周期读取挂在网关后面的 Modbus slave 设备（首个 profile 为 SDM120 单相电表），按设备的 **YAML profile** 解码为工程量，存入单库的**通用读数表**。后台静默轮询，支持多设备、多 profile。
+2. **MQTT + Home Assistant Discovery**：后端作为 MQTT 发布方，按"**可勾选暴露**"的方式把数据以 HA Discovery 自动注册成 device/entity（不止 sensor）。
+3. **前端侧边栏 + Energy 视图**：把现有顶栏改成侧边导航，承载首个领域视图 **Energy**（设备管理 + 最新读数 + 走势图）。
+
+> **命名分层（本里程碑的核心决策）**：**存储 / 采集 / API 是通用的 `modbus_*`**（不锁死在"电表"上，以后接别的 Modbus 设备无需改表）；**面向用户的呈现是领域特定的**——本里程碑只落第一个领域视图 **Energy**，消费通用 device 数据、用电表视角展示。
+
+> 三段有依赖关系，按 §6 的 `Depends` 顺序推进：A（侧边栏，独立）→ B（Modbus 采集后端 + Energy 前端）→ C（MQTT/Discovery，消费 B 的数据）。
+
+## 2. 现状（实现者可据此工作，不必通读全仓库）
+
+**单库数据层**（M1 完成态）
+- `app/db.py`：`class Base(DeclarativeBase)`，绑 `settings.app_database_url` 的 cached engine（WAL 已开），`get_engine` / `get_session_local` / `reset_db_caches` / `get_db_session`。
+- 模型都继承同一 `Base`：`app/models/{auth,config,public_ip,location,poo}.py`。
+- 单 Alembic 链 `alembic_app/`，head = `20260611_06_merge_location_poo_tables`（见 `alembic_app/versions/`）；`alembic_app/env.py` 逐个 import 所有模型。
+- 迁移命名惯例：`YYYYMMDD_NN_<desc>.py`，`revision` / `down_revision` 串链。
+- `scripts/app_db_adopt.py` 常量 `APP_BASELINE_REVISION` 指向当前 head；`scripts/run_migrations.py` 负责把 app 库升到 head。
+
+**配置系统（扁平 KV，自动渲染）**
+- `app/config.py`：`class Settings(BaseSettings)`，每个配置项一个带类型的字段 + 默认值。
+- `app/services/config_page.py`：`CONFIG_FIELDS: tuple[ConfigField, ...]` 是**注册表**（`section / env_name / setting_attr / label / secret / input_type`）；`build_config_sections`（读，secret 回空串）、`save_config_updates`（写，空 secret 保留旧值）、`build_runtime_settings`（DB override 合并进 Settings）、`_settings_payload`（把 Settings 摊平成 dict，新字段要在此补一行）。
+- `app/api/routes/api/config.py`：`GET/PUT /api/config`，session + CSRF 保护；非法值 422 且不写库。
+- 前端 `frontend/src/pages/ConfigPage.tsx`：**通用渲染**——按 section 分组、按 `input_type`/`secret` 渲染输入框。**新增标量配置项零前端改动**（追加 `CONFIG_FIELDS` + `Settings` 字段 + `_settings_payload` 一行即可）。
+- ⚠️ 扁平 KV **装不下"设备列表"和"逐实体勾选"**——这两者走专用表 + 专用 API + 自定义 UI（见 §3.2 / §3.4）。
+
+**后台调度（APScheduler）**
+- `app/main.py` lifespan：`BackgroundScheduler(timezone="UTC")`，`scheduler.add_job(_run_scheduled_public_ip_check, IntervalTrigger(hours=4), id=..., max_instances=1, coalesce=True)`，`scheduler.start()`；`yield` 后 `scheduler.shutdown(wait=False)`。
+- 周期任务惯例：一个**同步** wrapper 自己开/关 session：`session = get_session_local()(); try: service(session, ...) finally: session.close()`，service 内部不抛崩溃。
+
+**Home Assistant 现状（REST，不碰 MQTT）**
+- `app/integrations/homeassistant.py`：`HomeAssistantClient.publish_sensor()`（`POST /api/states/{entity}`）、`trigger_webhook()`。
+- 入站 webhook `app/api/routes/homeassistant.py`：`POST /homeassistant/publish`（envelope `target/action/content`）。
+- 新 MQTT Discovery 与此**并行、不冲突**，是第二条独立通道。
+
+**前端（M2）**
+- React + react-router v6 + Mantine + TanStack Query + `openapi-fetch` 生成的类型化 client（`frontend/src/api/client.ts` + `schema.d.ts`）。
+- `frontend/src/App.tsx`：`AppLayout`（当前是**顶栏**），包住所有受保护页；路由 `/`(HomePage 地图)、`/config`、`/records`；`/login`、`/change-password` 不带 layout。
+- 数据请求惯例：`useQuery`/`useMutation` + `apiClient.GET/POST/...`（见 `frontend/src/records/hooks.ts`）。
+- **无图表库**（只有 Leaflet 地图）；走势图需新引入 **Recharts**。
+
+## 3. 目标架构
+
+### 3.0 两层数据模型（本里程碑的地基决策）
+
+把"设备是什么、怎么读"与"读到了什么"彻底分成两层，**协议知识与部署信息分离**：
+
+```
+┌─────────────────────────────────────────────────────────────┐
+│ 协议层（固定，随代码走）         部署层（可配置，落 DB）       │
+│ ── YAML profile ──────────       ── modbus_device 行 ──────   │
+│ • 读哪些寄存器（FC/地址/块）     • friendly_name（可改）       │
+│ • 每个量：key / 类型 / 解码      • host / port（网关）          │
+│ • 每个量：unit / device_class    • unit_id（=电表 Meter ID，   │
+│   / ha_component                   设备上可设，故落 DB）        │
+│ • 唯一真相源，喂给"读/存/HA注册" • profile 名（选哪个 YAML）    │
+│                                  • uuid / poll_interval / enabled│
+└─────────────────────────────────────────────────────────────┘
+        ▲ 多个 device 可共享同一 profile         │
+        └─────────────────────────────────────────┘
+                          │ 采集
+                          ▼
+            ── modbus_reading 行（通用遥测）──
+            device_id · recorded_at · payload(JSON blob)
+```
+
+- **profile = 仓库内只读 YAML**（如 `app/integrations/modbus/profiles/sdm120.yaml`），随镜像打包、纳入版本控制，**只描述固定协议知识**：读哪些寄存器、怎么解码、每个量的 `key`/`unit`/`device_class`/`ha_component`。**绝不**放 friendly_name / unit_id 这类每设备各异的可配置项。
+- **可配置项落 `modbus_device` 行**，由前端 UI 设置：friendly_name、host、port、**unit_id（Modbus 从机地址，设备面板上可改）**、选用的 profile 名、poll_interval、enabled。
+- **多设备共享一个 profile**：例如两块 SDM120 都 `profile="sdm120"`，一块 friendly_name="SDM120 空调"、unit_id=1，另一块 friendly_name="SDM120 服务器"、unit_id=2——共用同一份解码规则，部署参数各自独立。
+- **读数存通用 JSON `payload`**，不再用固定列。profile 是解释器：知道 payload 里有哪些 key、各自单位与 device_class。
+  - 聚合不必搬到后端硬算：**SQLite 用 `json_extract` 在 SQL 端做 `AVG`/`MAX`/`GROUP BY`**（走势图都是带时间窗的查询，被 `(device_id, recorded_at)` 索引圈住，只扫窗内行）。
+  - 真有某个量需要高频聚合 → 后续给该量加 **generated column + 表达式索引**（非破坏性，要哪个补哪个），无需一开始把表锁成固定列。
+
+### 3.1 Modbus 采集
+
+- **传输：仅 Modbus-TCP**（用户的 Waveshare RTU↔TCP 网关，RJ45 以太网；服务器无串口）。用 **pymodbus** 的 `ModbusTcpClient`，由它处理封帧 / CRC / 超时重试 / float 解码。
+  - 角色厘清：**我们的后端 = Modbus client/master**；**网关 = TCP server**（在 "Modbus TCP" 模式下终结 TCP 再转 RTU）；**电表 = 网关后面的 RTU slave**，由 `unit_id` 寻址。`modbus_device` 一行同时装下网关地址（host:port）与 slave 地址（unit_id）。
+  - 网关若开"Modbus TCP"协议转换 → pymodbus 默认 framer 直连。
+  - 网关若是透传（RTU-over-TCP，裸 RTU 帧含 CRC）→ pymodbus 用 RTU framer over TCP。
+  - **二者实现时对一次即可确定**（连上读 Voltage 寄存器验证），不影响表结构与上层。
+- **协议知识在 YAML profile，部署信息在 DB**（见 §3.0）：
+  - `app/integrations/modbus/driver.py`：薄封装 pymodbus 的连接 + 块读 + 大端 float32 解码（word/byte 都大端，高寄存器在前）。
+  - `app/integrations/modbus/profiles.py`：YAML profile 的**加载 + 校验 + 解码 + 实体枚举**——纯"数据 + 函数"，**不做 OOP 抽象基类/继承**。`load_profile(name) -> ModbusProfile`（pydantic 模型，启动期校验）、`decode(profile, registers) -> dict[key -> value]`、`enumerate_entities(device, profile) -> list[ExposableEntity]`（供 §3.3）。
+  - `app/integrations/modbus/profiles/*.yaml`：每个设备型号一个 YAML。首个 = `sdm120.yaml`（寄存器地址见参考文档 §4）。
+  - profile 输出统一的 `dict[key -> value]`，由 service 整个塞进 `modbus_reading.payload`（JSON）；profile 没列出的量根本不读。
+- **只读**：不写电表配置寄存器（改 Meter ID/波特率有通信中断风险）。
+- **轮询**：每个设备一个 `poll_interval_s`（默认 5s），APScheduler 一个 job 扫所有 `enabled` 设备，逐设备块读→解码→落库。9600 总线上多设备串行，5s 余量充足。
+  - 全局开关 `MODBUS_POLLING_ENABLED`（CONFIG_FIELDS）可一键停采。
+  - 采集成功/失败写入设备的 `last_poll_at` / `last_poll_ok`（供 §3.3 的 online binary_sensor）。
+  - **两级周期 / 降采样 / 保留是后续杠杆**（见 §10），本里程碑用单周期读全。
+
+**`sdm120.yaml` 示例（只含固定协议知识）**
+
+```yaml
+name: sdm120
+description: Eastron SDM120 single-phase energy meter
+function_code: 4          # input registers (FC04)
+word_order: big           # 高寄存器在前
+byte_order: big
+blocks:                   # 块读，减少 Modbus 事务
+  - { start: 0x0000, count: 0x0060 }   # 30001..30095 连续段
+  - { start: 0x0156, count: 0x0004 }   # total active/reactive energy
+metrics:
+  - { key: voltage,            address: 0x0000, type: float32, unit: V,   device_class: voltage,    ha_component: sensor }
+  - { key: current,            address: 0x0006, type: float32, unit: A,   device_class: current,    ha_component: sensor }
+  - { key: active_power,       address: 0x000C, type: float32, unit: W,   device_class: power,      ha_component: sensor }
+  - { key: power_factor,       address: 0x001E, type: float32, unit: "",  device_class: power_factor, ha_component: sensor }
+  - { key: frequency,          address: 0x0046, type: float32, unit: Hz,  device_class: frequency,  ha_component: sensor }
+  - { key: import_energy,      address: 0x0048, type: float32, unit: kWh, device_class: energy, state_class: total_increasing, ha_component: sensor }
+  - { key: export_energy,      address: 0x004A, type: float32, unit: kWh, device_class: energy, state_class: total_increasing, ha_component: sensor }
+  - { key: total_energy,       address: 0x0156, type: float32, unit: kWh, device_class: energy, state_class: total_increasing, ha_component: sensor }
+# 冷门量（demands / maxima / 无功电能…）可后续按需补进 metrics；未列出的就不读。
+```
+
+> 注意 YAML 里**没有** unit_id / friendly_name——那些是 `modbus_device` 行各自带的。
+
+### 3.2 数据模型（新增两张通用表，单库 app 链）
+
+**`modbus_device`**（设备定义 = 部署/可配置项，CRUD 管理）
+
+| 列 | 类型 | 说明 |
+| --- | --- | --- |
+| id | int PK | 内部代理主键（FK join 用）|
+| uuid | str unique | uuid4 生成的**稳定内部身份**；也作 API 路径键与 HA `unique_id` 的锚 |
+| friendly_name | str | 显示名（可改；改名重发 discovery，HA 显示名跟着变）|
+| transport | str | 现仅 `'tcp'` |
+| host | str | 网关 IP |
+| port | int | 网关端口（默认 502）|
+| unit_id | int | Modbus 从机地址 = 电表 Meter ID（设备面板可改，默认 1）|
+| profile | str | 用哪个 YAML profile（首个 `'sdm120'`）|
+| poll_interval_s | int | 采样周期（默认 5）|
+| enabled | bool | 是否轮询 |
+| last_poll_at | datetime null | 最近一次轮询时刻（online 判定）|
+| last_poll_ok | bool null | 最近一次轮询成败（online 判定）|
+| created_at / updated_at | datetime | |
+
+**`modbus_reading`**（通用遥测表，一行 = 一个设备的一次采样）
+
+| 列 | 类型 | 说明 |
+| --- | --- | --- |
+| id | int PK | |
+| device_id | int FK→modbus_device.id | **ON DELETE RESTRICT**（见 §5 删除语义）|
+| recorded_at | datetime (UTC) | 采样时刻，**真实列、带索引**（所有查询按它走时间窗）|
+| payload | JSON | profile 解码出的全部工程量 `{key: value}`；无固定列 |
+
+- 索引：`(device_id, recorded_at)`。
+- **SDM120 单相 payload 示例**：`{"voltage": 230.2, "current": 1.3, "active_power": 295.0, "power_factor": 0.98, "frequency": 50.0, "import_energy": 123.4, "export_energy": 0.0, "total_energy": 123.4}`——key 由 `sdm120.yaml` 的 `metrics[].key` 决定。
+- **接入新设备型号无需改表**：换 profile，payload 里的 key 集合随之变；表结构不动。
+- **三相电表**以后用三相 profile，payload 里多几个相位 key（如 `voltage_l1/l2/l3`），仍是同一张表。
+
+### 3.3 MQTT + HA Discovery（通用 expose 框架，元数据由 profile 派生）
+
+HA MQTT Discovery 模型 = **device → entities**：往 `<prefix>/<component>/<node>/<object>/config` 发 retained 消息定义一个 entity；config 内 `device.identifiers` 相同的 entity 归到同一个 HA device 卡片下；之后往 `state_topic` 推值。
+
+- **可暴露实体目录由 provider 动态产出**：
+  - `app/integrations/expose.py`：定义 `ExposableEntity`（`key`（稳定）、`component`（sensor/binary_sensor/switch…）、`device`（归属，决定 HA device 分组）、`device_class`、`unit`、取值来源）+ 一个 provider 注册表。
+  - **Modbus/energy provider**：每个 `enabled` 设备 = 一个 HA **device**（`identifiers` 用设备 `uuid`），其各工程量 = 一组 sensor entity——**`device_class`/`unit`/`component` 直接取自该设备 profile 的 `metrics[]`**（不再单独维护一份映射），外加一个 `binary_sensor`「online」（取 `last_poll_ok`）——这就是"**不止 sensor**"的体现。
+  - 其它 provider（如 public-ip、poo）可后续挂入，本里程碑只接 Modbus/energy provider。
+- **HA 实体身份锚定（Z2M 模型）**：discovery config 的 **`unique_id` 用设备 `uuid` + 量的 `key` 派生**（稳定，不随改名变）；可见的 **`name` 用 friendly_name**。改 friendly_name → 重发 discovery → HA 显示名跟着变、但 `unique_id` 不变故历史不丢。topic 的 object_id 也用 uuid 派生（稳定、丑无所谓）。
+- **`exposed_entities` 表**：只存"逐 key 的开关"（`key` unique + `enabled` + `updated_at`）。目录本身由 provider 计算，表只记被勾选的状态（默认未勾 = 不暴露）。
+- **MQTT 客户端**：`app/integrations/mqtt.py` 用 **paho-mqtt**，`loop_start()` 后台线程；在 lifespan 起/停；支持配置变更后**重连 + 重发 discovery**。
+- **发布时机**：
+  - discovery config（retained）：连接成功时、目录/勾选变更时全量发；取消勾选时发空 payload 清除该 entity。
+  - state：采集在每次轮询后推最新值；另有一个周期 job 兜底重发所有 enabled 实体的 state + availability（在线）topic。
+- **配置**（走现有扁平 CONFIG_FIELDS）：`MQTT_ENABLED`、`MQTT_BROKER_HOST/PORT/USERNAME/PASSWORD(secret)`、`MQTT_TLS_ENABLED`、`HA_DISCOVERY_ENABLED`、`HA_DISCOVERY_PREFIX`（默认 `homeassistant`）。
+
+### 3.4 前端
+
+- **侧边栏**：把 `AppLayout` 从顶栏重构为侧边导航（Mantine `AppShell` 或 flex sidebar），导航项：Home / Records / Energy / Config + 主题切换 + 注销；当前路由高亮；移动端可折叠。仅改 `App.tsx`（+ 可抽 `AppSidebar`/`NavItem` 组件），各页面主体不动。
+- **Energy 视图**（新页 `/energy`，首个领域视图，消费通用 `/api/modbus` 数据）：
+  - 设备管理：列表 + 新建/编辑/删除（删除有二次确认；后端对有读数的设备拒删，引导改用"禁用"）。新建/编辑表单里设 friendly_name、host、port、unit_id、profile（下拉选 `sdm120` 等）、poll_interval、enabled。
+  - 最新读数卡片（每设备当前各工程量；字段标签/单位取自 profile 的 metrics 元数据，见 `/metrics` 端点）。
+  - 走势图：用 **Recharts** 画时间序列（电压/电流/功率/电能），时间范围选择，取数走 readings API（窗口 + 上限），从 `payload` 里按 key 取序列。
+- **Expose 设置**：设置页内一块「Home Assistant Expose」——列出可暴露实体目录、逐项勾选、显示 MQTT/Discovery 连接状态、一个"重新发布 discovery"按钮。
+
+## 4. API 契约（M5 要落地的端点）
+
+> 全部 `/api` 前缀、session + CSRF（写）保护、JSON 进出。schema 经 `export_openapi.py` 固化入库。路径键用设备 `uuid`（稳定、非自增）。
+
+| 分组 | 端点 | 用途 |
+| --- | --- | --- |
+| Modbus | `GET /api/modbus/devices` | 列出设备 |
+| Modbus | `POST /api/modbus/devices` | 新建设备 |
+| Modbus | `GET /api/modbus/devices/{uuid}` | 单个设备 |
+| Modbus | `PATCH /api/modbus/devices/{uuid}` | 修改设备（含 enable/disable）|
+| Modbus | `DELETE /api/modbus/devices/{uuid}` | 删除设备；**有读数时 409**，引导改 disable |
+| Modbus | `GET /api/modbus/devices/{uuid}/metrics` | 该设备 profile 的量目录（key/label/unit/device_class），供前端渲染卡片与图表标签 |
+| Modbus | `GET /api/modbus/devices/{uuid}/latest` | 该设备最新一条读数（payload）|
+| Modbus | `GET /api/modbus/devices/{uuid}/readings` | 时间范围读数（`start/end/limit`，limit 有上限），返回 `recorded_at + payload`，供走势图 |
+| Modbus | `POST /api/modbus/devices/{uuid}/test` | 即时试读一次（验证网关连通/地址），返回解码 payload，不落库 |
+| Modbus | `GET /api/modbus/profiles` | 列出可用 profile 名 + 描述（前端建设备时的下拉选项）|
+| Expose | `GET /api/expose` | 返回可暴露实体目录 + 勾选状态 + MQTT/Discovery 状态 |
+| Expose | `PUT /api/expose` | 设置逐 key 勾选（map key→bool）|
+| Expose | `POST /api/expose/republish` | 手动重发 discovery |
+| 配置 | `POST /api/config/mqtt/test` | 试连 broker **并发布一条测试消息**（MQTT Explorer 可见），仿 SMTP 测试三态 |
+
+> MQTT broker / discovery 的**标量配置**复用现有 `GET/PUT /api/config`（只新增 CONFIG_FIELDS，不新增端点）。
+
+## 5. 已锁定决策（讨论后拍板）
+
+1. **里程碑编排**：一个 M5 文档分三段，`Depends` 串顺序（A 侧边栏 → B Modbus/Energy → C MQTT）。
+2. **两层数据模型，协议与部署分离**：`modbus_device`（部署/可配置项：friendly_name、host、port、unit_id、profile 名、poll、enabled）+ `modbus_reading`（通用遥测：device_id、recorded_at、`payload` JSON）。取代原"宽表 + 固定列"。
+3. **读数 = JSON `payload`，无固定列**；聚合走 SQLite `json_extract`（DB 端做 AVG/MAX/GROUP BY），热点量后补 generated column + 表达式索引。两级周期/降采样为后续杠杆。
+4. **协议知识 = 仓库内只读 YAML profile**（声明式"数据 + 函数"，无 OOP 继承，pydantic 启动期校验）；YAML 携带寄存器 + 每个量的 key/unit/device_class/ha_component，是**唯一真相源**，同时驱动"读 / 存 / HA 注册"。多设备可共享一个 profile。首个 profile `sdm120`。
+5. **Modbus 仅 TCP**，pymodbus，framer 配网关模式，**只读**采集。设备是网关后面的 slave，`unit_id` 寻址（设备面板可改，故落 DB）。
+6. **命名分层（方案 C）**：存储/采集/API 一律通用 `modbus_*`、`/api/modbus/devices`；前端首个领域视图叫 **Energy**（消费通用 device 数据）。
+7. **UUID = 内部生成（uuid4）的稳定身份**：既做内部索引/ API 路径键，也做 HA discovery `unique_id` 的锚；friendly_name 可改，改名重发 discovery、HA 显示名跟着变（Z2M 模型，历史不丢）。
+8. **MQTT = 通用 expose 框架**：provider 动态产出可暴露实体目录，**实体元数据（device_class/unit/component）从 profile 派生**；`exposed_entities` 只存逐 key 开关；支持 sensor/binary_sensor/switch 等多 component；设备自动注册（每设备一 device、各量为 entity + 一个 online binary_sensor）。
+9. **MQTT 库 = paho-mqtt**，lifespan 长连接，配置变更后重连 + 重发 discovery。
+10. **MQTT broker/discovery 标量配置走现有扁平 CONFIG_FIELDS**（自动渲染）；设备清单与 expose 勾选走专用表 + 自定义 UI。
+11. **图表库 = Recharts**（封在自包含组件后，仿 M2 对 Leaflet 的隔离）。
+12. **删除设备安全**：FK `ON DELETE RESTRICT`，有读数拒删（避免一键删表丢历史）；"停用"用 `enabled=false`。
+13. **Config 页用 Accordion 分区**（进页见大类、逐类展开），**不在 config 页内再放第二个 side nav**——避免与主侧栏（T01）的"双抽屉"冲突；纯前端、独立任务 M5-T01B。
+14. **CLI 手工测试工具为"受控手工链路验证"而设**（设备接市电、非随时在线，不进自动化）：`scripts/modbus_cli` 提供 `read`（profile 解码）与 `probe`（手工指定请求内容、看原始回复）两个子命令，**一律只读**（仅 FC03/04），不暴露写寄存器。
+15. **MQTT/HA 发布链路的手工验证走 UI + 外部工具，不另做 CLI**：Config 页「发送测试消息」（`mqtt/test` 发一条到测试 topic）→ 在 **MQTT Explorer** 查看；Expose 勾选实体 + 开 `HA_DISCOVERY_ENABLED` + 「重新发布 discovery」（`/api/expose/republish`）→ 到 **Home Assistant** 查看。不通则迭代配置再重发。
+
+> 项目定位：个人自用、家庭特化、不开源——可按单用户场景简化，不过度抽象。
+
+## 6. 任务依赖图
+
+```
+Phase A（独立，可最先做，纯前端）
+  M5-T01  [structural] 侧边栏布局重构
+  M5-T01B Config 页分区折叠（Accordion）   ← 与 T01 互不依赖，可并行/先后任意
+
+Phase B（Modbus 采集 + Energy 前端）
+  M5-T02 [schema] modbus_device + modbus_reading 表 + 模型
+   ├─► M5-T03 Modbus 驱动 + YAML profile 框架（pymodbus + sdm120.yaml，纯模块）
+   │     └─► M5-T04 采集 service + APScheduler 轮询（接 lifespan，落 payload）
+   └─► M5-T05 Modbus JSON API（device CRUD + readings + metrics + test）
+            └─► M5-T06 前端：设备管理 UI（依赖 T01 侧栏 + T05 API）
+            └─► M5-T07 前端：读数展示 + Recharts 走势图（依赖 T05；引入 recharts）
+
+Phase C（MQTT / Discovery，依赖 B 的设备数据与 provider 接口）
+  M5-T08 MQTT/Discovery 配置项（CONFIG_FIELDS）
+  M5-T09 [schema] exposed_entities 表 + ExposableEntity/provider 框架（modbus provider 从 profile 派生）
+   ├─► M5-T10 MQTT 客户端（paho，lifespan 连接 + 重连 + config/mqtt/test）
+   │     └─► M5-T11 Discovery 发布 + state 发布（连 T04 轮询推 state）
+   └─► M5-T12 前端：Expose 勾选 UI + /api/expose API
+
+收尾
+  M5-T13 文档 + OpenAPI + roadmap 收尾（依赖全部）
+```
+
+`T01`、`T01B`、`T02`、`T08` 无前置可先开。
+
+---
+
+## 7. 原子任务（任务卡）
+
+> 后端任务沿用校验闸门（`pytest` / `ruff` / 改路由或 schema 则 `export_openapi` 重导出入库）。前端任务闸门见 §8。新增依赖（`pymodbus`、`PyYAML`、`paho-mqtt`、`recharts`）须在对应任务里同步 `requirements.in/.txt` 或 `frontend/package.json` 并重新锁定。
+
+### M5-T01 — 侧边栏布局重构 `[structural]`
+- **Status**: `todo` · **Depends**: none
+- **Context**: 把 `AppLayout` 从顶栏改为侧边导航，给后续 Energy 等视图腾入口。纯前端，不碰各页主体。
+- **Files**: `modify frontend/src/App.tsx`；`create frontend/src/components/AppSidebar.tsx`、`frontend/src/components/NavItem.tsx`（可选）；`modify` 受影响的 `frontend/src/pages/*.test.tsx`（导航断言）
+- **Steps**:
+  1. 用 Mantine `AppShell`（或 flex sidebar）重构 `AppLayout`：左侧竖直导航（Home/Records/Config + 主题切换 + 注销），`<Outlet/>` 在右。
+  2. 当前路由高亮（`useLocation` 比对 `pathname`）；移动端可折叠（burger）。
+  3. 导航项图标沿用 `react-feather`；样式走 Mantine（暗色模式自动适配）。
+  4. 不在本任务加 Energy 项（页面还不存在，T06 加），保持导航无死链。
+- **Out of scope / 不要碰**: 不改各页面主体；不动鉴权（SessionProvider/ProtectedRoute）；不引入图表库。
+- **Acceptance criteria**:
+  - [ ] 受保护页都在侧边栏布局内；`/login`、`/change-password` 不带布局（与现状一致）。
+  - [ ] 当前路由在侧栏高亮；移动端宽度下可折叠/展开。
+  - [ ] 前端闸门全绿（`lint`/`typecheck`/`test`/`build`）。
+- **Reviewer checklist**: 布局只在 `App.tsx`/新组件内变动，未误改页面或鉴权；无死链导航项。
+
+### M5-T01B — Config 页分区折叠（Accordion）
+- **Status**: `todo` · **Depends**: none
+- **Context**: config 内容会越来越多（M5 还要加 MQTT / HA Discovery / Modbus 配置 + Expose 面板）。把 ConfigPage 从一长串 section 改为 Mantine `Accordion`：进页只见各大类标题，逐类展开编辑。**纯前端、单列内容**——它不是"第二个 app 级侧栏"，与主侧栏（T01）无冲突，故 `Depends: none`、可独立先做。
+- **Files**: `modify frontend/src/pages/ConfigPage.tsx`；`modify frontend/src/pages/ConfigPage.test.tsx`（折叠/展开断言）
+- **Steps**:
+  1. 用 Mantine `Accordion` 包住现有"按 section 分组"的渲染：每个 config section = 一个 `Accordion.Item`（标题 = section 名，面板 = 该 section 的字段表单）。
+  2. 默认折叠（或首个展开）；**保留现有保存逻辑与 config API 不变**（M2 的整页/按 section 保存语义照旧，本任务不碰后端）。
+  3. 预留：将来 T12 的「Home Assistant Expose」自定义面板也作为一个 `Accordion.Item` 接入，保持一致。
+  4. 移动端单列堆叠即可，**不引入第二个侧栏/抽屉**。
+- **Out of scope / 不要碰**: 不改后端 config API/字段（`CONFIG_FIELDS` / `_settings_payload`）；不动主侧栏（T01）；不改保存语义；不在 config 页内放任何 app 级 side nav。
+- **Acceptance criteria**:
+  - [ ] ConfigPage 以 accordion 呈现，每大类可独立展开/折叠；字段渲染与保存行为与现状一致。
+  - [ ] 与主侧栏无视觉/交互冲突（单列内容；移动端不出现双抽屉）。
+  - [ ] 前端闸门全绿（`lint`/`typecheck`/`test`/`build`）。
+- **Reviewer checklist**: 仅改 ConfigPage（+其测试），未碰 config 后端或主 layout；保存逻辑无回归；页面内**无第二个 app 级 sidebar**（accordion 是内容、非 chrome）。
+
+### M5-T02 — `modbus_device` + `modbus_reading` 表与模型 `[schema]`
+- **Status**: `todo` · **Depends**: none
+- **Context**: 单库 app 链新增两张**通用**表，建出 §3.2 结构（设备 = 部署层，读数 = JSON payload 通用遥测层）。本任务只建 schema + 模型，不写采集/接口。
+- **Files**: `create app/models/modbus.py`（`ModbusDevice`、`ModbusReading`，继承 `app.db.Base`）；`create alembic_app/versions/<date>_07_modbus_tables.py`；`modify alembic_app/env.py`（import 新模型）；`modify scripts/app_db_adopt.py`（`APP_BASELINE_REVISION` → 新 head）；`create tests/test_modbus_models.py`
+- **Steps**:
+  1. 模型按 §3.2 列定义（`Mapped[...]` 2.0 风格）：`ModbusDevice` 含 `uuid`(unique)、`friendly_name`、`transport`、`host`、`port`、`unit_id`、`profile`、`poll_interval_s`、`enabled`、`last_poll_at`/`last_poll_ok`(nullable)、时间戳；`uuid` 用 `default` 生成 uuid4 字符串。`ModbusReading` 含 `device_id` FK→`modbus_device.id`（**`ondelete="RESTRICT"`**）、`recorded_at`、`payload`（JSON 列，非 null）。
+  2. 新 revision：`down_revision` = 当前 head；`upgrade()` 用 `op.create_table` 建两表 + 索引 `(device_id, recorded_at)`；`downgrade()` 反向 drop。
+  3. 更新 `APP_BASELINE_REVISION`。
+- **Out of scope / 不要碰**: 不写 pymodbus/采集（T03/T04）；不加路由（T05）；不动其它模型。
+- **Acceptance criteria**:
+  - [ ] 全新临时 app 库 upgrade 到 head 后含 `modbus_device`、`modbus_reading` 及索引；`downgrade -1` 干净回滚。
+  - [ ] `Base.metadata.tables` 含两新表；FK 为 RESTRICT；`uuid` 唯一且自动生成。
+  - [ ] `APP_BASELINE_REVISION` == 新 head。
+  - [ ] 校验闸门全绿。
+- **Reviewer checklist**: 列/约束与 §3.2 一致；`payload` 为 JSON 列；`recorded_at` 为真实索引列（非塞进 payload）；链上单 head；`env.py` 已 import 新模型（否则 autogenerate/建表漏表）。
+
+### M5-T03 — Modbus 驱动 + YAML profile 框架（`sdm120`）
+- **Status**: `todo` · **Depends**: M5-T02
+- **Context**: 薄封装 pymodbus 的连接/块读/大端 float 解码，加 YAML profile 加载/校验/解码框架与 SDM120 profile。纯模块，mock client 单测。
+- **Files**: `create app/integrations/modbus/__init__.py`、`app/integrations/modbus/driver.py`、`app/integrations/modbus/profiles.py`、`app/integrations/modbus/profiles/sdm120.yaml`、`scripts/modbus_cli.py`；`modify requirements.in`/`requirements.txt`（加 `pymodbus`、`PyYAML`，重新锁定）；`create tests/test_modbus_driver.py`、`tests/test_modbus_profiles.py`、`tests/test_modbus_cli.py`
+- **Steps**:
+  1. `driver.py`：`read_blocks(host, port, unit_id, blocks) -> dict[int,int]`（按 profile 的 blocks 块读 input registers，FC04），用 `ModbusTcpClient`；超时/连接失败抛明确异常；大端 float32 解码 helper（`registers_to_float`，高寄存器在前）。framer 选择留可配置/可探测。
+  2. `profiles.py`：定义 pydantic `ModbusProfile` / `MetricSpec`（`key`/`address`/`type`/`unit`/`device_class`/`state_class`?/`ha_component`）；`load_profile(name) -> ModbusProfile`（读 `profiles/<name>.yaml`，校验失败抛错）；`decode(profile, registers) -> dict[key -> value]`（按各 metric 的 address+type 从寄存器对解码）；`list_profiles() -> list[(name, description)]`。**不做抽象基类/继承**，全是数据 + 模块函数。
+  3. `profiles/sdm120.yaml`：按 §3.1 示例写全核心量（参考文档 §4 地址）。
+  4. `scripts/modbus_cli.py`：纯命令行、**不依赖 DB / 不需先配设备**，供**受控手工测试**（设备接市电、非随时在线，不进自动化）。两个**只读**子命令：
+     - `read --host H --port P --unit U --profile sdm120`：按 profile 读一次、解码后**把各工程量打印成可读结果**（表格/JSON）——验证整套解码链路。
+     - `probe --host H --port P --unit U --fc 4 --address 0x0000 --count 2 [--decode float32]`：**手工指定要发送的请求内容**（功能码 FC03/04 + 起始地址 + 数量），打印**原始寄存器（hex）+ 可选大端 float 解码**——first-contact 验证网关连通、framer 模式与 unit 地址，可单读一个寄存器，不依赖 profile。
+     连接失败给清晰报错 + 非零退出。**只读**：CLI 仅暴露读功能码（FC03/04），**不提供任何写寄存器子命令**（数据红线 + 防改坏电表通信参数）。
+- **Out of scope / 不要碰**: 不连真实硬件（单测用 mock/fake 返回已知寄存器字节）；不写调度（T04）；不写电表配置寄存器；不在 profile 里放 unit_id/friendly_name。
+- **Acceptance criteria**:
+  - [ ] 单测：给定 `0x4366,0x3334` 解码为 `230.2`（参考文档实例）；字序/字节序正确。
+  - [ ] 单测：`load_profile("sdm120")` 校验通过；`decode(profile, ...)` 把已知寄存器映射到正确 key 与值。
+  - [ ] 单测：profile YAML 缺字段/类型错时 `load_profile` 抛可识别校验错。
+  - [ ] 连接失败/超时抛可识别异常，不静默返回错值。
+  - [ ] `python -m scripts.modbus_cli read ...` 能（对 mock/真实网关）打印解码后的各工程量；连接失败非零退出。
+  - [ ] `python -m scripts.modbus_cli probe --fc 4 --address 0x0000 --count 2 ...` 能打印原始寄存器与可选解码值；CLI **无任何写寄存器子命令**。
+  - [ ] 校验闸门全绿。
+- **Reviewer checklist**: 解码确为大端、高寄存器在前；地址与参考文档一致；**CLI 与 driver 都无任何写寄存器路径（仅 FC03/04）**；profile 纯协议知识、无部署项；`requirements.txt` 已同步锁定 `pymodbus`、`PyYAML`。
+
+### M5-T04 — 采集 service + APScheduler 轮询
+- **Status**: `todo` · **Depends**: M5-T03
+- **Context**: 周期扫所有 enabled 设备，调用 driver 读+解码，落 `modbus_reading.payload`。仿 public-ip 的同步 job 模式。
+- **Files**: `create app/services/modbus_poll.py`；`modify app/main.py`（lifespan 注册 job）；`create tests/test_modbus_poll.py`
+- **Steps**:
+  1. `modbus_poll.py`：`poll_device(session, device) -> ModbusReading | None`（`load_profile` → driver 读 → `decode` → 把 dict 存进 `payload` 插一行；更新 `last_poll_at`/`last_poll_ok`）；`poll_all_enabled_devices(session)` 遍历 enabled 设备；service 内吞异常并日志，不让 job 崩。
+  2. `main.py`：加同步 wrapper `_run_scheduled_modbus_poll`（自管 session），`add_job(IntervalTrigger(seconds=...), id="modbus-poll", max_instances=1, coalesce=True)`。周期取**最小 per-device interval 或一个基础 tick**（实现可用单一基础 tick + 各设备按自身 interval 取模决定本 tick 是否读，保持 job 简单）；受全局 `MODBUS_POLLING_ENABLED` 控制。
+  3. 失败的设备记 `last_poll_ok=false`（供 T11 暴露），不影响其它设备。
+- **Out of scope / 不要碰**: 不发 MQTT（T11）；不加 HTTP 路由（T05）；不引入两级周期（后续杠杆）。
+- **Acceptance criteria**:
+  - [ ] 单测：mock driver 返回已知 dict，`poll_all_enabled_devices` 后 `modbus_reading` 精确 +N 行、`payload` 内容正确。
+  - [ ] 单测：某设备读失败时其它设备仍正常落库，job 不抛；失败设备 `last_poll_ok=false`。
+  - [ ] `MODBUS_POLLING_ENABLED=false` 时不轮询。
+  - [ ] 校验闸门全绿。
+- **Reviewer checklist**: session 在 wrapper 内开关、try/finally 关闭；job `max_instances=1` 防叠加；无 N+1/每行单独 connect 的明显低效；异常不外泄崩 job；payload 为解码后的 dict（非裸寄存器）。
+
+### M5-T05 — Modbus JSON API（device CRUD + readings + metrics + test）
+- **Status**: `todo` · **Depends**: M5-T02
+- **Context**: 给前端提供设备 CRUD、最新读数、时间范围读数、量目录、即时试读。
+- **Files**: `create app/api/routes/api/modbus.py`、`app/schemas/modbus.py`；`modify app/main.py`（注册路由）；`create tests/test_api_modbus.py`
+- **Steps**:
+  1. devices：`GET`(list)/`POST`/`GET{uuid}`/`PATCH{uuid}`/`DELETE{uuid}`；session+CSRF；`POST` 校验 profile 名存在；`DELETE` 有读数 → 409。
+  2. readings：`GET {uuid}/latest`（最新一行 payload）、`GET {uuid}/readings`（`start/end/limit`，limit 有上限防全表导出，按 `recorded_at` 升序，返回 `recorded_at + payload`）。
+  3. `GET {uuid}/metrics`：返回该设备 profile 的量目录（key/label/unit/device_class），供前端渲染。
+  4. `GET /api/modbus/profiles`：`list_profiles()` 的名+描述。
+  5. `POST {uuid}/test`：用 driver 即时读一次返回解码 payload（或错误），**不落库**。
+- **Out of scope / 不要碰**: 不在此处发 MQTT；不写采集逻辑（复用 T03/T04 的 driver/service）。
+- **Acceptance criteria**:
+  - [ ] CRUD 行为正确：创建/改/删行数精确；删有读数的设备返回 409；未登录 401、缺 CSRF 403；建设备引用不存在 profile → 422。
+  - [ ] readings 时间范围 + limit 上限生效；latest 返回最新一条 payload；metrics 返回 profile 量目录。
+  - [ ] schema 经 OpenAPI 固化入库。
+  - [ ] 校验闸门全绿（含 `openapi/` 重导出）。
+- **Reviewer checklist**: 删除受 RESTRICT 保护、无批量删/清表路径；查询走 `(device_id, recorded_at)` 索引；test 端点确不落库；路径键用 `uuid`。
+
+### M5-T06 — 前端：设备管理 UI（Energy 视图）
+- **Status**: `todo` · **Depends**: M5-T01, M5-T05
+- **Context**: 在侧栏加 Energy 入口与 `/energy` 路由；设备增删改 + 试读。
+- **Files**: `create frontend/src/pages/EnergyPage.tsx`、`frontend/src/energy/DeviceForm.tsx`、`frontend/src/energy/hooks.ts`；`modify frontend/src/App.tsx`（路由）、`frontend/src/components/AppSidebar.tsx`（Energy 项）；`create` 对应 `*.test.tsx`
+- **Steps**: `useQuery`/`useMutation` 接 `/api/modbus/devices` API；列表 + 新建/编辑表单（friendly_name/host/port/unit_id/profile 下拉/poll/enabled）+ 删除二次确认（删失败 409 提示改用禁用）；"试读"按钮调 `POST {uuid}/test` 显示结果。
+- **Out of scope / 不要碰**: 走势图在 T07；不碰其它页面。
+- **Acceptance criteria**:
+  - [ ] 能增/改/删设备并即时刷新；删除有二次确认；409 有友好提示；profile 走 `/api/modbus/profiles` 下拉。
+  - [ ] 侧栏出现 Energy 入口、`/energy` 可达。
+  - [ ] 前端闸门全绿。
+- **Reviewer checklist**: 全部走生成的类型化 client；删除走确认；无与契约不符的手写请求。
+
+### M5-T07 — 前端：读数展示 + Recharts 走势图
+- **Status**: `todo` · **Depends**: M5-T05（数据）, M5-T06（页面壳）
+- **Context**: 在 Energy 页展示每设备最新读数 + 时间序列走势。
+- **Files**: `modify frontend/src/pages/EnergyPage.tsx`；`create frontend/src/energy/EnergyCharts.tsx`（封装 Recharts）；`modify frontend/package.json`（加 `recharts`，`package-lock.json` 同步）；`create` 对应测试
+- **Steps**: 最新读数卡片（接 `latest`，字段标签/单位取自 `/metrics`）；时间范围选择 + 折线图（电压/电流/功率/电能，从 `payload` 按 key 取序列），接 `readings`（窗口 + limit）；图表封在 `EnergyCharts` 内（仿 Leaflet 隔离，便于将来换库）。
+- **Out of scope / 不要碰**: 不做服务端降采样（后续）；不改后端。
+- **Acceptance criteria**:
+  - [ ] 最新读数与走势图渲染正确；时间范围只取窗口数据（不拉全量）；标签/单位来自 metrics。
+  - [ ] Recharts 封装自包含、仅此处 import。
+  - [ ] 前端闸门全绿（`build` 通过，注意 chunk 体积提示）。
+- **Reviewer checklist**: 图表组件隔离；查询有窗口/上限；空数据/加载/错误态有处理；从 payload 取 key 的逻辑容忍缺 key。
+
+### M5-T08 — MQTT / Discovery 配置项（CONFIG_FIELDS）
+- **Status**: `todo` · **Depends**: none
+- **Context**: 把 MQTT broker 与 discovery 的标量配置接入扁平配置系统（前端自动渲染）。
+- **Files**: `modify app/config.py`（新增 Settings 字段）、`app/services/config_page.py`（追加 CONFIG_FIELDS + `_settings_payload`）；`modify .env.example`；`modify tests/test_api_config.py`
+- **Steps**: 加字段 `mqtt_enabled`、`mqtt_broker_host/port/username/password`(secret)、`mqtt_tls_enabled`、`ha_discovery_enabled`、`ha_discovery_prefix`(默认 `homeassistant`)、`modbus_polling_enabled`；CONFIG_FIELDS 归入「MQTT」「Home Assistant Discovery」「Modbus」section；`_settings_payload` 补齐对应行。
+- **Out of scope / 不要碰**: 不建 MQTT 客户端（T10）；不动 expose 表（T09）。
+- **Acceptance criteria**:
+  - [ ] 新配置项在 `GET /api/config` 出现且分 section；password 为 secret（回空、留空保留）；port 为 number。
+  - [ ] 非法值（端口非数字）422 不写库。
+  - [ ] 校验闸门全绿（OpenAPI 若变化则重导出）。
+- **Reviewer checklist**: secret 不回显/不入 OpenAPI 示例；`_settings_payload` 未漏字段（否则运行期 override 丢失）。
+
+### M5-T09 — `exposed_entities` 表 + ExposableEntity/provider 框架 `[schema]`
+- **Status**: `todo` · **Depends**: M5-T02
+- **Context**: 建"可暴露实体目录"的抽象与开关存储；modbus provider 把设备映射成 device/entities，**实体元数据从 profile 派生**。
+- **Files**: `create app/integrations/expose.py`（`ExposableEntity`、provider 协议、注册表、modbus provider）；`create app/models/expose.py`（`ExposedEntityToggle`：`key` unique + `enabled` + `updated_at`）；`create alembic_app/versions/<date>_08_exposed_entities.py`；`modify alembic_app/env.py`、`scripts/app_db_adopt.py`；`create tests/test_expose_catalog.py`
+- **Steps**:
+  1. `ExposableEntity`（`key/component/device/device_class/unit/value_getter`）+ provider 接口 `enumerate(session) -> list[ExposableEntity]`。
+  2. modbus provider：每个 enabled 设备 → 一个 device（`identifiers` 用设备 `uuid`），其各量 → sensor entity（**device_class/unit/component 取自 profile 的 `metrics[]`**），加一个 `binary_sensor` online（取 `last_poll_ok`）。
+  3. `build_catalog(session)` 合并所有 provider 的实体 + 各自 `enabled`（来自 toggle 表，缺省 false）。
+  4. migration 建 toggle 表；更新 baseline 常量。
+- **Out of scope / 不要碰**: 不发 MQTT（T11）；不加 HTTP（T12）。
+- **Acceptance criteria**:
+  - [ ] 单测：建若干设备后 `build_catalog` 产出每设备对应 entity（含 online binary_sensor）+ 正确 device 分组、device_class、unit（与 profile 一致）。
+  - [ ] toggle 表 migration 可升/降；缺省 enabled=false。
+  - [ ] 至少含一个非 sensor component（online binary_sensor）。
+  - [ ] 校验闸门全绿。
+- **Reviewer checklist**: `key` 稳定（用设备 `uuid` + 量 key，不用自增 id，避免重建漂移）；component 支持多类型；目录由 provider 计算、元数据源自 profile 而非写死。
+
+### M5-T10 — MQTT 客户端（paho，lifespan 连接 + 重连）
+- **Status**: `todo` · **Depends**: M5-T08
+- **Context**: 长连接 MQTT 客户端，配置变更可重连；含连接测试端点。
+- **Files**: `create app/integrations/mqtt.py`；`modify app/main.py`（lifespan 起/停）、`app/api/routes/api/config.py`（`POST /api/config/mqtt/test`）；`modify requirements.in`/`requirements.txt`（加 `paho-mqtt` 重新锁定）；`create tests/test_mqtt_client.py`
+- **Steps**:
+  1. `MqttManager`：`is_configured()`、`connect()/disconnect()/reconnect(settings)`、`publish(topic, payload, retain)`；paho `loop_start()` 后台线程；未配置/未启用则 no-op。
+  2. lifespan：启用则 connect；shutdown disconnect。
+  3. 配置保存后若 MQTT 设置变化 → 触发 manager 重连（在 config 保存路径加 hook 或保存后比对）。
+  4. `POST /api/config/mqtt/test`：用提交/现存配置试连**并发布一条测试消息**到一个测试 topic（如 `<discovery_prefix>/home-automation/test`），返回三态（success/config-error/failed）。仿 SMTP 测试"发一封测试邮件"的语义——用户随后在 **MQTT Explorer** 里就能看到这条消息，确认 broker 发布链路通（这是 MQTT 端的手工验证手段，**不另做 CLI**）。
+- **Out of scope / 不要碰**: 不构建 discovery/state 消息（T11）。
+- **Acceptance criteria**:
+  - [ ] 单测（fake broker/paho mock）：configured 时 connect 调用正确；未配置 no-op；publish 透传 topic/payload/retain。
+  - [ ] `POST /api/config/mqtt/test` 试连**并发布一条测试消息**（可在 MQTT Explorer 看到）；三态有明确返回；session+CSRF 保护。
+  - [ ] 校验闸门全绿。
+- **Reviewer checklist**: 断网/连接失败不崩主进程；线程在 shutdown 正确停止；`requirements.txt` 同步锁定 `paho-mqtt`；密码不进日志。
+
+### M5-T11 — Discovery 发布 + state 发布
+- **Status**: `todo` · **Depends**: M5-T09, M5-T10
+- **Context**: 把 enabled 实体发成 HA discovery config（retained）并周期推 state；采集轮询后推最新值。
+- **Files**: `create app/services/ha_discovery.py`；`modify app/services/modbus_poll.py`（轮询后推 state）、`app/main.py`（state 周期 job + 连接后/勾选变更后发 discovery）、`app/api/routes/api/...`（`/api/expose/republish` 在 T12 接，本任务提供 service）；`create tests/test_ha_discovery.py`
+- **Steps**:
+  1. `build_discovery_payload(entity)` → HA 规范 config（`<prefix>/<component>/<node>/<object>/config`，含 `device` 块、`state_topic`、`device_class`、`unit_of_measurement`、`availability`；**`unique_id` 用设备 uuid + 量 key 派生，`name` 用 friendly_name**）。
+  2. `publish_discovery(session)`：对 enabled 实体发 retained config；对取消勾选的发空 payload 清除。
+  3. `publish_states(session)`：取各实体当前值发 state；采集在 `poll_device` 成功后顺带推该设备实体 state + online。
+  4. lifespan：连接成功 / 目录或勾选变更后 `publish_discovery`；周期 job 兜底 `publish_states` + availability。
+- **Out of scope / 不要碰**: 不做前端（T12）；不改采集解码逻辑。
+- **Acceptance criteria**:
+  - [ ] 单测：discovery payload 符合 HA 结构（device 分组正确、topic/device_class/unit 正确、`unique_id` 取自 uuid）；取消勾选发空 payload。
+  - [ ] 单测：采集轮询成功后推对应 state topic；失败推 online=false。
+  - [ ] discovery 用 retained。
+  - [ ] 校验闸门全绿。
+- **Reviewer checklist**: 仅发 enabled 实体；entity `unique_id` 稳定（源自 uuid，不随改名变）；MQTT 未启用时整链 no-op；不阻塞轮询。
+
+### M5-T12 — 前端：Expose 勾选 UI + `/api/expose`
+- **Status**: `todo` · **Depends**: M5-T09, M5-T11
+- **Context**: 后端 expose 读写端点 + 设置页勾选界面。
+- **Files**: `create app/api/routes/api/expose.py`、`app/schemas/expose.py`；`modify app/main.py`；`create tests/test_api_expose.py`；前端 `create frontend/src/pages/.../ExposeSettings.tsx`（或并入 ConfigPage）、`modify` 路由/设置入口；`create` 前端测试
+- **Steps**: `GET /api/expose`（目录 + 勾选 + MQTT/Discovery 状态）、`PUT /api/expose`（key→bool）、`POST /api/expose/republish`（调 T11 service）；前端列出目录、按 device 分组、逐项开关、显示连接状态、"重新发布"按钮。
+- **Out of scope / 不要碰**: 不改采集/发布逻辑（T11）。
+- **Acceptance criteria**:
+  - [ ] `GET/PUT /api/expose` 正确读写勾选；session+CSRF；OpenAPI 固化。
+  - [ ] 勾选变更后（或点重新发布）触发 discovery 重发。
+  - [ ] 前端能勾选并显示状态；前后端闸门全绿。
+- **Reviewer checklist**: PUT 只改 toggle 不误碰其它配置；republish 真触发 T11；类型化 client。
+
+### M5-T13 — 文档 + OpenAPI + roadmap 收尾
+- **Status**: `todo` · **Depends**: 全部
+- **Files**: `modify README.md`（Modbus/Energy/MQTT 段、新依赖）、`docs/roadmap.md`（M5 行 + 把"MQTT/IoT"从"下一阶段"毕业、新增 Modbus 采集方向）、`docs/architecture-overview.md`（新增 MQTT 通道与 Modbus 采集）；`modify docs/design/README.md`（列入 m5）；`run python scripts/export_openapi.py` 并提交 `openapi/`
+- **Acceptance criteria**:
+  - [ ] 文档反映新链路；`git diff --exit-code openapi/` 无未提交差异。
+  - [ ] 校验闸门全绿。
+- **Reviewer checklist**: 无残留旧描述（含旧 `energy_meters`/`/api/energy` 字样）；OpenAPI 已入库。
+
+---
+
+## 8. 前端校验闸门（前端任务每次结束都要全绿）
+
+在 `frontend/` 下：
+```bash
+npm ci
+npm run lint
+npm run typecheck
+npm run test
+npm run build        # 必须产出 dist；留意 chunk 体积告警
+```
+- 后端若同任务改了路由/schema，仍需根目录 `python scripts/export_openapi.py` 并提交 `openapi/`。
+- 新增前端依赖（recharts）须提交 `package.json` + `package-lock.json`。
+
+## 9. 构建上下文完整性（M1 教训）
+
+- 本里程碑**不删/移文件**，但新增了 Python 依赖（`pymodbus`、`PyYAML`、`paho-mqtt`）与前端依赖（`recharts`）：必须同步 `requirements.in/.txt` 的重新锁定与 `package-lock.json`，否则镜像构建会缺包。
+- 新增源文件都在 `app/`、`scripts/` 与 `frontend/` 既有 COPY 范围内，无需改 `Dockerfile` 的 `COPY`；**但 `app/integrations/modbus/profiles/*.yaml` 是非 .py 资源**——确认 `COPY app ...` 把整个目录（含 YAML）带进镜像，且运行期能按相对路径定位 YAML（建议用 `importlib.resources` 或基于 `__file__` 的路径，别用 CWD 相对路径）。`scripts/modbus_cli.py` 须在镜像里可 `python -m scripts.modbus_cli` 调用；`tests/test_deployment.py::test_dockerfile_copy_sources_exist` 仍应通过。
+- 发版前置走查（见 CLAUDE.md）：真起 app 跑一次轮询、真连一次 broker、前端 Energy 视图人工瞄一眼渲染，再打 tag。
+
+## 10. 后续杠杆（本里程碑不做，文档留痕）
+
+- **热点量的 generated column + 索引**：JSON `payload` 默认无法对单个量建索引；某个量若需高频聚合，给 `modbus_reading` 加一列 `GENERATED ALWAYS AS (json_extract(payload,'$.<key>'))` 并建索引——非破坏性、要哪个补哪个。
+- **两级采样周期**：瞬时量（V/I/P/PF/Hz）快、累计电能慢——9600 总线吃紧或要把功率压到 5s 以下时再开（device 表加 `slow_interval_s`，读数表是通用 payload、无需改结构）。
+- **保留 / 降采样**：5s 采样长期行数大（≈630 万行/年/设备），加定期降采样或保留窗口任务（独立于本里程碑），`GROUP BY` + `AVG(json_extract(...))` 在 SQL 端做。
+- **更多 device profile**：三相电表（如 SDM630）新增一个 YAML profile（payload 里多几个相位 key），不改表、不改采集主链。
+- **更多 expose provider**：public-ip / poo 等挂入 expose 框架。
+- **写 Modbus 寄存器**：当前只读；如需经 MQTT/UI 控制设备（switch 类），再单独评估安全边界。
+
+## 11. 人工验收 walkthrough（实现完成后）
+
+> 重点：用命令行**手工**读到设备数据并展示结果。这些是**受控手工测试**——设备接市电、并非随时在线，故**不纳入自动化测试**（自动化只用 mock）；CLI 工具（`modbus_cli read/probe`）就是为这种"我想测的时候手动测一次"而设。可用 `docker compose` 起环境，命令在容器内或容器外跑均可。
+
+**前提**：网关（Waveshare RTU↔TCP）已上电接入网络，电表 Meter ID 已知（默认 1）。
+
+**1) 命令行直接试读（不依赖 DB，最快验证）**
+- 容器内：`docker compose exec <app> python -m scripts.modbus_cli read --host <网关IP> --port 502 --unit 1 --profile sdm120`
+- 或容器外：`source .venv/bin/activate && python -m scripts.modbus_cli read --host <网关IP> --port 502 --unit 1 --profile sdm120`
+- 预期：打印解码后的工程量（电压/电流/有功功率/功率因数/频率/导入导出电能等），数值合理；连不上则清晰报错。
+- **first-contact / 原始验证**：`python -m scripts.modbus_cli probe --host <网关IP> --port 502 --unit 1 --fc 4 --address 0x0000 --count 2 --decode float32` —— 手工指定请求内容、看原始回复（电压寄存器应解出约 230V），用于确认网关 framer 模式与 unit 地址；建 profile / 设备前就能跑。
+
+**2)（可选）经 API 试读已配置的设备**
+- 先在前端 Energy 页或 `POST /api/modbus/devices` 建一个设备；
+- `POST /api/modbus/devices/{uuid}/test` 即时试读，返回解码 payload（不落库）。
+
+**3)（可选）验证后台轮询落库**
+- 确认 `MODBUS_POLLING_ENABLED=true` 且设备 `enabled`；等一个采样周期；
+- 看前端 Energy 视图的最新读数/走势图，或查 `GET /api/modbus/devices/{uuid}/readings` 有新行。
+
+**4)（可选）手工验证 MQTT / HA Discovery 发布链路**（受控手工步骤，不进自动化）
+- **broker 发布链路**：配好 MQTT broker 后，在 Config 页点「发送测试消息」（`POST /api/config/mqtt/test`）——它试连并发一条测试消息；打开 **MQTT Explorer** 确认能收到，即链路通。
+- **HA Discovery**：在 Expose 设置勾选若干实体、开 `HA_DISCOVERY_ENABLED`，点「重新发布 discovery」（`POST /api/expose/republish`）；到 **Home Assistant** 确认对应 device/entity 出现、值正确。
+- **改名验证**：改某设备 friendly_name 后重发，确认 HA 显示名跟着变、历史不丢（`unique_id` 稳定）。
+- 不通则按需调整配置/勾选再重发即可。
+
+## 12. 里程碑完成定义（DoD）
+
+- 后端能按 per-device 周期静默轮询 Modbus-TCP 设备、按 YAML profile 解码落 `modbus_reading.payload`，支持多设备 CRUD。
+- MQTT 启用时，勾选的实体以 HA Discovery 注册成 device/entities（含非 sensor），state 周期发布；配置变更可重连重发；改 friendly_name 重发后 HA 显示名跟着变、`unique_id` 不变。
+- 前端侧边栏可切换功能；Energy 视图能管理设备、看最新读数与走势图；设置页可勾选 expose。
+- 后端 `pytest`/`ruff`/`export_openapi` + 前端 `lint/typecheck/test/build` 全绿且 `openapi/` 已入库。
+- README / architecture / roadmap / design 索引反映 M5 现实。

docs/references/SDM120-Modbus-Protocol.md

@@ -0,0 +1,134 @@
+# SDM120 Modbus 协议（从官方 PDF 提取）
+
+> 来源：`docs/references/SDM120-MODBUS_Protocol.pdf`
+> （Eastron SDM120 Modbus Smart Meter Modbus Protocol Implementation **V2.4**）
+> 本文件是 PDF 的可读化提取，供本项目的 Modbus 采集驱动设计参考。**以官方 PDF 为准**，本文件如有出入以 PDF 为准。
+
+## 0. 本项目的接入方式（重要）
+
+SDM120 物理层是 **Modbus RTU（RS-485 串口）**。本项目通过一个 **Modbus-TCP 网关**接入：
+
+- 后端用 **Modbus TCP**（`IP:port`）连到网关，网关在串口侧转成 RTU 与电表通信。
+- TCP 帧用 MBAP header、**无 CRC**（CRC/Error Check 由网关在 RTU 侧处理）。本文档里 RTU 帧的 `Error Check (Lo/Hi)` 字段在 TCP 模式下不需要我们关心。
+- **Slave Address / Unit ID = 电表的 Meter ID**（默认 `1`，范围 1–247），在 TCP 请求里作为 unit id 传入。
+- 若以后直连串口（RTU），才需要管波特率 / 校验位 / CRC（见 §5 holding 寄存器）。
+
+## 1. 协议帧格式
+
+MODBUS 定义 master 查询 / slave 响应的格式。Eastron 电表用 16-bit 寄存器在主从间传值，**但实际数据是 32-bit IEEE-754 浮点**，因此每个测量参数占**两个相邻的 16-bit 寄存器**。
+
+### Query（master → slave，RTU 帧）
+
+| 字段 | 说明 |
+| --- | --- |
+| Slave Address | 8-bit，目标从机地址 1–247（0=广播，Eastron 不支持广播） |
+| Function Code | 8-bit，功能码（Eastron 支持 **03 / 04 / 08 / 16(=0x10)**） |
+| Start Address (Hi/Lo) | 16-bit 起始寄存器地址；**寄存器成对使用、从 0 开始，所以起始地址必须是偶数** |
+| Number of Points (Hi/Lo) | 16-bit 请求的寄存器数量；**也必须是偶数**（成对读浮点） |
+| Error Check (Lo/Hi) | 16-bit CRC（RTU 模式；TCP 网关模式无此字段） |
+
+### Response（slave → master）
+
+| 字段 | 说明 |
+| --- | --- |
+| Slave Address | 响应从机地址 |
+| Function Code | 与查询相同的功能码（表示识别并已响应） |
+| Byte Count | 8-bit，本次返回的数据字节数 |
+| Data (寄存器对) | 每个寄存器 Hi byte / Lo byte，按"高寄存器在前"排列 |
+| Error Check (Lo/Hi) | 16-bit CRC（RTU 模式） |
+
+### Exception Response（异常响应）
+
+- 异常响应的 Function Code = 查询功能码 **OR 0x80**（最高位置 1）。
+- 数据是单字节 Error Code（异常码）。
+- PDF 正文写了"见后文 Table Of Exception Codes"，但提取的 8 页里**没有附上该表**。标准 Modbus 异常码（供参考，非本 PDF 内容）：`01` 非法功能、`02` 非法数据地址、`03` 非法数据值、`04` 从机设备故障。
+
+## 2. 功能码
+
+| 功能码 | 作用 | 寄存器区 |
+| --- | --- | --- |
+| **04** | Read Input Registers（读输入寄存器，3X）—— **所有测量值都在这里** | 30001+ |
+| **03** | Read Holding Registers（读保持寄存器，4X）—— 配置项 | 40001+ |
+| **16 / 0x10** | Write Holding Registers（写保持寄存器，4X）—— 改配置 | 40001+ |
+| 08 | Diagnostics（诊断） | — |
+
+> ⚠️ **测量值用 FC 04（输入寄存器），不是 FC 03。** 这是最常见的踩坑点。
+
+## 3. 浮点数据编码
+
+- 每个参数 = **32-bit IEEE-754 float**，占两个相邻 16-bit 寄存器。
+- **字序（word order）= 大端：高寄存器在前。**
+- **字节序（byte order）= 大端：寄存器内高字节在前。**
+- 即整体就是标准大端 float（`>f`），4 字节顺序 = `[Reg1 Hi][Reg1 Lo][Reg2 Hi][Reg2 Lo]`。
+
+**实例（来自 PDF）**
+
+| 含义 | 原始 4 字节 (hex) | 解码值 |
+| --- | --- | --- |
+| Volts 1 | `43 66 33 34` | `230.2` V |
+| Demand Time | `3F 80 00 00` | `1.0` |
+| Network Node | `42 70 00 00` | `60.0` |
+
+> Python 解码：`struct.unpack('>f', bytes([0x43,0x66,0x33,0x34]))[0]` → `230.2`。
+> pymodbus 用 `BinaryPayloadDecoder.fromRegisters(regs, byteorder=Endian.BIG, wordorder=Endian.BIG)`。
+
+## 4. 输入寄存器表（FC 04 读测量值）
+
+全部为 `Float`，长度 4 字节，每项占 2 个寄存器。"Hex 起始"是 Modbus 协议起始地址（即 Start Address Hi/Lo）。
+
+| 寄存器 | 参数 | 单位 | Hex 起始 |
+| --- | --- | --- | --- |
+| 30001 | Voltage（电压） | Volts | `0000` |
+| 30007 | Current（电流） | Amps | `0006` |
+| 30013 | Active power（有功功率） | Watts | `000C` |
+| 30019 | Apparent power（视在功率） | VA | `0012` |
+| 30025 | Reactive power（无功功率） | VAr | `0018` |
+| 30031 | Power factor（功率因数） | — | `001E` |
+| 30071 | Frequency（频率） | Hz | `0046` |
+| 30073 | Import active energy（导入有功电能） | kWh | `0048` |
+| 30075 | Export active energy（导出有功电能） | kWh | `004A` |
+| 30077 | Import reactive energy（导入无功电能） | kvarh | `004C` |
+| 30079 | Export reactive energy（导出无功电能） | kvarh | `004E` |
+| 30085 | Total system power demand | W | `0054` |
+| 30087 | Maximum total system power demand | W | `0056` |
+| 30089 | Import system power demand | W | `0058` |
+| 30091 | Maximum import system power demand | W | `005A` |
+| 30093 | Export system power demand | W | `005C` |
+| 30095 | Maximum export system power demand | W | `005E` |
+| 30259 | Current demand | Amps | `0102` |
+| 30265 | Maximum current demand | Amps | `0108` |
+| 30343 | Total active energy（总有功电能） | kWh | `0156` |
+| 30345 | Total reactive energy（总无功电能） | kvarh | `0158` |
+
+**读取分块建议**：`0x0000–0x005E`（30001–30095）地址连续，可一次块读；`0x0102/0x0108`、`0x0156/0x0158` 各为独立小块。整表用 2–3 次块读即可覆盖，减少 Modbus 事务数。
+
+### 常用核心子集（日常监控够用）
+
+电压 `0000`、电流 `0006`、有功功率 `000C`、功率因数 `001E`、频率 `0046`、导入有功电能 `0048`、导出有功电能 `004A`、总有功电能 `0156`。
+
+## 5. 保持寄存器表（FC 03 读 / FC 16 写配置）
+
+| 寄存器 | 参数 | Hex 起始 | 格式 | 说明 |
+| --- | --- | --- | --- | --- |
+| 40013 | Relay Pulse Width | `000C` | Float | 继电器脉宽 60/100/200 ms，默认 100ms |
+| 40019 | Network Parity Stop | `0012` | Float | 0=1停止位无校验(默认)，1=1停止位偶校验，2=1停止位奇校验，3=2停止位无校验；**改后需重启生效** |
+| 40021 | Meter ID | `0014` | Float | 从机地址 1–247，默认 1 |
+| 40029 | Baud rate | `001C` | Float | 0=2400(默认)，1=4800，2=9600，5=1200 |
+| 40087 | Pulse 1 output mode | `0056` | Float | 0001 导入有功，0002 导入+导出有功，0004 导出有功(默认)，0005 导入无功，0006 导入+导出无功，0008 导出无功 |
+| 463745 | Time of scroll display | `F900` | HEX 2字节 | 滚动显示时间 0–30s，默认 0（不滚动） |
+| 463761 | Pulse 1 output | `F910` | HEX 2字节 | 0000:0.001kWh/imp(默认)，0001:0.01，0002:0.1，0003:1 kWh/imp |
+| 463777 | Measurement mode | `F920` | HEX 2字节 | 1:total=import，2:total=import+export(默认)，3:total=import-export |
+| 464513 | Serial number | `FC00` | uint32 4字节 | 序列号，**只读** |
+| 464515 | Meter code | `FC02` | Hex 2字节 | 设备码=`0020`，**只读** |
+| 464516 | Software version | `FC03` | Hex 2字节 | 软件版本，**只读** |
+
+> ⚠️ 写保持寄存器（改 Meter ID / 波特率 / 校验位等）会改变电表通信参数，配错可能导致**通信中断**。本项目默认**只读采集**，不建议在自动化链路里写电表配置。
+
+## 6. 给本项目采集驱动的要点小结
+
+1. 走 **Modbus TCP 网关**：`ModbusTcpClient(host, port)`，`slave=<Meter ID>`。
+2. 测量值用 **FC 04 / 输入寄存器**，按 §4 地址读。
+3. 解码 **大端 float32**（word & byte 都大端，高寄存器在前）。
+4. 起始地址与数量**都用偶数**（成对读）。
+5. 默认**只读**，不写电表配置寄存器。
+6. 不同型号电表 → 不同"寄存器 profile"。本表是 `sdm120` 这一个 profile 的定义。

docs/roadmap.md

@@ -2,7 +2,7 @@
 
 本文档记录 `home-automation` 在 `v1.0.3` 之后的下一阶段规划。这一阶段不是小修补，而是几次较大的结构性改动：单库化、前端重写、以及远期的移动端试水。
 
-> 每个里程碑的**可执行原子任务**展开在 [`docs/design/`](./design/README.md)：M1 [`m1-db-consolidation.md`](./design/m1-db-consolidation.md)、M2 [`m2-frontend-v2.md`](./design/m2-frontend-v2.md)、M3 [`m3-token-mobile.md`](./design/m3-token-mobile.md)。这些文档为 Orchestrator→Implementer→Reviewer 的多模型流水线设计。
+> 每个里程碑的**可执行原子任务**展开在 [`docs/design/`](./design/README.md)：M1 [`m1-db-consolidation.md`](./design/m1-db-consolidation.md)、M2 [`m2-frontend-v2.md`](./design/m2-frontend-v2.md)、M3 [`m3-token-mobile.md`](./design/m3-token-mobile.md)、M4 [`m4-login-hardening.md`](./design/m4-login-hardening.md)。这些文档为 Orchestrator→Implementer→Reviewer 的多模型流水线设计。
 
 ## 当前基线（v1.0.3）
 
@@ -36,9 +36,11 @@
 | --- | --- | --- |
 | **M1** ✅ | 单库化地基 | 把三库合并成单一 `app.db`，清理散落数据层，删掉 Grafana |
 | **M2** ✅ | 前端 v2 | React SPA 取代 Jinja，承载 config + 可视化 + 记录增删改 |
+| **M4** ✅ | 登录加固 | 防爆破/指数退避 + CLI 逃生通道 + 可选 TOTP 二次验证（**先于 M5**） |
+| **M5** | IoT / 能耗采集 | Modbus/Energy + MQTT/HA Discovery + 前端侧边栏 |
 | **M3** | 开放与移动端（远期试水） | token 鉴权 + React Native 移动端 |
 
-排序原则：**先清地基，再在干净结构上盖楼。** M2 的新 API 和 React 必须建立在合并后的单库之上，否则就是在准备推倒的旧数据层上盖新楼、之后回头返工。
+排序原则：**先清地基，再在干净结构上盖楼。** M2 的新 API 和 React 必须建立在合并后的单库之上；M4 是公网安全加固，在 M5 IoT 集成之前先堵住裸密码这个洞；M5 在安全基座就绪后再做 IoT 接入。
 
 ---
 
@@ -132,6 +134,22 @@
 
 ---
 
+## M4 — 登录加固（✅ 已完成，排在 M5 之前）
+
+### 目标
+
+给暴露在公网的单 admin 登录做纵深防御，先于 M5 IoT 集成关闭暴力枚举和单因子风险。
+
+### 范围
+
+- **防爆破 / 指数退避**：失败登录按双键（IP + username）指数增长延迟，成功即清零。退避是延迟（429 + Retry-After），不是永久封号。全局开关 `AUTH_LOGIN_THROTTLE_ENABLED`（CONFIG_FIELDS）；反代后需开 `AUTH_TRUST_FORWARDED_FOR`（`.env` 部署级）。
+- **CLI 逃生通道**：`python -m scripts.admin_cli` 直连本地 DB，无需 HTTP 服务、无需任何已存凭据；支持重置密码、解锁退避、关停/重发 TOTP、查看用户。
+- **可选 TOTP 二次验证**：admin 可自选启用；启用后两步登录（密码 + 6 位动态码或一次性恢复码）；不启用维持纯密码。CLI `disable-totp` 是连恢复码都丢了时的最终逃生口。
+
+> 详细设计与任务卡：[`docs/design/m4-login-hardening.md`](./design/m4-login-hardening.md)
+
+---
+
 ## M3 — 开放与移动端（远期试水）
 
 ### 目标
@@ -149,22 +167,28 @@
 - 移动端是这一阶段最远期、最不确定的部分。
 - token 主要是移动端的前置条件；Web 端 React 用现有 session cookie 即可，不需要为它提前引入 token。
 
-## Future Ideas（暂不排期，想到先记下）
+## 下一阶段：已确定要做（尚未拆解为任务卡）
 
-> 这里收集**还没排进里程碑**的想法。不是承诺、也没有先后顺序；想做时再从这里捞出来细化成 `docs/design/` 的任务卡。**明确不开 M2.5**——下列条目一律先躺在 Future Ideas，之后再说。
+> 这些是 M4 之后**已经定下来要做**的方向——区别于下面的 Future Ideas（仅备忘、未必做）。这里只记到 roadmap 粒度：确定**做什么、为什么**；具体排期、依赖与原子任务，等动手时再展开成 `docs/design/` 的任务卡。**先后顺序未定**，具体排期等动手时再定。
 
-### TOTP 二次验证（Dashboard 加固）
+### 1. 前端优化
 
-**动机**：M2 之后多了一个 Web Dashboard。它虽有单 admin 密码保护，但**大概率会暴露在公网**上，只靠密码这一层不够。给登录再叠一层 **TOTP（基于时间的一次性密码，RFC 6238）** 作为第二因子，做纵深防御。
+**动机**：M2 的 React SPA 先把功能跑通，性能 / 体验层面的打磨还没做。这一项**确定要做，但具体优化什么还没定**。
+
+**范围（待定）**：方向先留空，想清楚再细化。可能的候选（仅占位、非承诺）：打包体积与代码分割（M2 构建已提示存在 > 500 kB 的单 chunk）、首屏加载、热力图 / 地图的渲染性能、移动端适配、可访问性等。等确定具体目标后再拆任务卡。
+
+### 2. 设置页生成 Long-lived Token（供 API 调用）
+
+**动机**：浏览器端走 session cookie 即可，但**脚本 / 设备 / 外部程序调用 API** 需要一种长期有效、可随身携带的凭据。在设置页加一组功能，由 admin **手动签发 long-lived token**，之后用它来调 API。
 
 **范围（粗略，待细化）**：
 
-- 在现有单 admin（Argon2 + server-side session）登录之上，叠加 TOTP 第二步：密码校验通过后再验 6 位动态码，通过才发 session cookie。
+- 设置页新增「API Token」区：生成 / 命名 / 吊销 long-lived token；明文只在**生成时展示一次**，此后只存哈希。
-- 首次启用时生成 TOTP secret，给出可导入 Authenticator 的二维码 / 可手输密钥；同时生成一组一次性**恢复码（recovery codes）**。
+- 后端支持用该 token 鉴权访问 API（与现有 session cookie 并存，互不影响）。
+- 与 [M3](#m3--开放与移动端远期试水) 的 token 主题相关，但**这条是 Web 设置页手动签发的 PAT 风格**，不依赖移动端 OAuth 流程；两者实现时可复用同一套 token 存储 / 校验。
 
-**运维 / 命令行要求（关键，实现时必须满足）**：
+## Future Ideas（暂不排期，想到先记下）
 
-1. **忘记密码**：不需要任何 Web 端“找回密码”流程——直接在命令行里重置 admin 密码即可（沿用现有 CLI 思路）。
+> 这里收集**还没排进里程碑、也还没决定要不要做**的想法。不是承诺、也没有先后顺序；想做时再从这里捞出来——先升进上面的「下一阶段」，再细化成 `docs/design/` 的任务卡。
-2. **TOTP 重置 / 恢复**：必须提供**命令行重置入口**。要覆盖最坏情况——**连恢复码（restore key）都丢了**，也能纯靠 CLI 把 TOTP 关掉 / 重新发放新的 secret，从而恢复登录。即：**CLI 是不依赖任何已存恢复凭据的最终逃生通道**，不能出现“密钥丢了就彻底锁死”的死角。
 
-**先不做**：本条仅记入 Future Ideas，不进 M2.5、不排期；之后再细化为 design 任务卡。
+_（暂无条目。）_

frontend/package-lock.json

@@ -17,6 +17,7 @@
         "leaflet.heat": "^0.2.0",
         "leaflet.markercluster": "^1.5.3",
         "openapi-fetch": "^0.17.0",
+        "qrcode.react": "^4.2.0",
         "react": "^18.3.1",
         "react-dom": "^18.3.1",
         "react-feather": "^2.0.10",
@@ -5661,6 +5662,15 @@
         "node": ">=6"
       }
     },
+    "node_modules/qrcode.react": {
+      "version": "4.2.0",
+      "resolved": "https://registry.npmjs.org/qrcode.react/-/qrcode.react-4.2.0.tgz",
+      "integrity": "sha512-QpgqWi8rD9DsS9EP3z7BT+5lY5SFhsqGjpgW5DY/i3mK4M9DTBNz3ErMi8BWYEfI3L0d8GIbGmcdFAS1uIRGjA==",
+      "license": "ISC",
+      "peerDependencies": {
+        "react": "^16.8.0 || ^17.0.0 || ^18.0.0 || ^19.0.0"
+      }
+    },
     "node_modules/react": {
       "version": "18.3.1",
       "resolved": "https://registry.npmjs.org/react/-/react-18.3.1.tgz",

frontend/package.json

@@ -22,6 +22,7 @@
     "leaflet.heat": "^0.2.0",
     "leaflet.markercluster": "^1.5.3",
     "openapi-fetch": "^0.17.0",
+    "qrcode.react": "^4.2.0",
     "react": "^18.3.1",
     "react-dom": "^18.3.1",
     "react-feather": "^2.0.10",

frontend/src/api/schema.d.ts

@@ -245,6 +245,7 @@ export interface paths {
          *
          *     On success, sets an HttpOnly session cookie and returns the session user + CSRF token.
          *     On failure, returns 401 with no cookie set.
+         *     Repeated failures trigger exponential back-off (429 + Retry-After).
          *     No X-CSRF-Token required (unauthenticated endpoint).
          */
         post: operations["post_login_api_auth_login_post"];
@@ -300,6 +301,112 @@ export interface paths {
         patch?: never;
         trace?: never;
     };
+    "/api/auth/totp/setup": {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        get?: never;
+        put?: never;
+        /**
+         * Post Totp Setup
+         * @description Generate a new pending TOTP secret, otpauth URI, and one-time recovery codes.
+         *
+         *     The secret is stored in the DB but TOTP is NOT yet enabled (totp_enabled stays
+         *     False until the user confirms with POST /api/auth/totp/enable).
+         *
+         *     Recovery codes are returned here as plaintext exactly once; their Argon2 hashes
+         *     are persisted immediately so enable only needs to flip the enabled flag.
+         *
+         *     Repeating this call replaces any prior pending secret and regenerates codes.
+         *
+         *     Requires: session cookie + X-CSRF-Token.
+         */
+        post: operations["post_totp_setup_api_auth_totp_setup_post"];
+        delete?: never;
+        options?: never;
+        head?: never;
+        patch?: never;
+        trace?: never;
+    };
+    "/api/auth/totp/enable": {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        get?: never;
+        put?: never;
+        /**
+         * Post Totp Enable
+         * @description Enable TOTP by confirming with the current 6-digit code from the authenticator app.
+         *
+         *     Requires a prior call to POST /api/auth/totp/setup (so that a pending secret
+         *     exists).  On success, totp_enabled becomes True.
+         *
+         *     Returns 400 if the code is wrong or there is no pending secret.
+         *     Requires: session cookie + X-CSRF-Token.
+         */
+        post: operations["post_totp_enable_api_auth_totp_enable_post"];
+        delete?: never;
+        options?: never;
+        head?: never;
+        patch?: never;
+        trace?: never;
+    };
+    "/api/auth/totp/disable": {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        get?: never;
+        put?: never;
+        /**
+         * Post Totp Disable
+         * @description Disable TOTP.  The caller must provide exactly one of:
+         *     - ``password``: the user's current login password, OR
+         *     - ``code``: the current 6-digit TOTP code.
+         *
+         *     On success: totp_enabled=False, totp_secret cleared, all recovery codes deleted.
+         *     Returns 400 if neither credential matches or neither is provided.
+         *     Requires: session cookie + X-CSRF-Token.
+         */
+        post: operations["post_totp_disable_api_auth_totp_disable_post"];
+        delete?: never;
+        options?: never;
+        head?: never;
+        patch?: never;
+        trace?: never;
+    };
+    "/api/auth/totp": {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        /**
+         * Get Totp Status
+         * @description Return the current TOTP status for the authenticated user.
+         *
+         *     Response contains only ``{"enabled": bool}``.
+         *     Secret and recovery codes are NEVER returned here.
+         *     Requires: session cookie only (no CSRF — read-only).
+         */
+        get: operations["get_totp_status_api_auth_totp_get"];
+        put?: never;
+        post?: never;
+        delete?: never;
+        options?: never;
+        head?: never;
+        patch?: never;
+        trace?: never;
+    };
     "/homeassistant/publish": {
         parameters: {
             query?: never;
@@ -510,6 +617,8 @@ export interface components {
             username: string;
             /** Password */
             password: string;
+            /** Totp Code */
+            totp_code?: string | null;
         };
         /** PasswordChangeRequest */
         PasswordChangeRequest: {
@@ -647,6 +756,50 @@ export interface components {
             /** Status */
             status: string;
         };
+        /**
+         * TotpDisableRequest
+         * @description Disable TOTP by proving identity.
+         *
+         *     Exactly one of ``password`` or ``code`` must be provided.
+         */
+        TotpDisableRequest: {
+            /** Password */
+            password?: string | null;
+            /** Code */
+            code?: string | null;
+        };
+        /**
+         * TotpEnableRequest
+         * @description The user confirms setup by providing the 6-digit TOTP code.
+         */
+        TotpEnableRequest: {
+            /** Code */
+            code: string;
+        };
+        /**
+         * TotpSetupResponse
+         * @description Returned once after a setup call.
+         *
+         *     ``secret`` and ``recovery_codes`` are **one-time plaintext values**.
+         *     They are never returned again by any subsequent API call.
+         *     The frontend must display and instruct the user to save them before confirming.
+         */
+        TotpSetupResponse: {
+            /** Secret */
+            secret: string;
+            /** Otpauth Uri */
+            otpauth_uri: string;
+            /** Recovery Codes */
+            recovery_codes: string[];
+        };
+        /**
+         * TotpStatusResponse
+         * @description Minimal status response — never exposes secret or recovery codes.
+         */
+        TotpStatusResponse: {
+            /** Enabled */
+            enabled: boolean;
+        };
         /** ValidationError */
         ValidationError: {
             /** Location */
@@ -1143,6 +1296,123 @@ export interface operations {
             };
         };
     };
+    post_totp_setup_api_auth_totp_setup_post: {
+        parameters: {
+            query?: never;
+            header?: {
+                "X-CSRF-Token"?: string | null;
+            };
+            path?: never;
+            cookie?: never;
+        };
+        requestBody?: never;
+        responses: {
+            /** @description Successful Response */
+            200: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content: {
+                    "application/json": components["schemas"]["TotpSetupResponse"];
+                };
+            };
+            /** @description Validation Error */
+            422: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content: {
+                    "application/json": components["schemas"]["HTTPValidationError"];
+                };
+            };
+        };
+    };
+    post_totp_enable_api_auth_totp_enable_post: {
+        parameters: {
+            query?: never;
+            header?: {
+                "X-CSRF-Token"?: string | null;
+            };
+            path?: never;
+            cookie?: never;
+        };
+        requestBody: {
+            content: {
+                "application/json": components["schemas"]["TotpEnableRequest"];
+            };
+        };
+        responses: {
+            /** @description Successful Response */
+            204: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content?: never;
+            };
+            /** @description Validation Error */
+            422: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content: {
+                    "application/json": components["schemas"]["HTTPValidationError"];
+                };
+            };
+        };
+    };
+    post_totp_disable_api_auth_totp_disable_post: {
+        parameters: {
+            query?: never;
+            header?: {
+                "X-CSRF-Token"?: string | null;
+            };
+            path?: never;
+            cookie?: never;
+        };
+        requestBody: {
+            content: {
+                "application/json": components["schemas"]["TotpDisableRequest"];
+            };
+        };
+        responses: {
+            /** @description Successful Response */
+            204: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content?: never;
+            };
+            /** @description Validation Error */
+            422: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content: {
+                    "application/json": components["schemas"]["HTTPValidationError"];
+                };
+            };
+        };
+    };
+    get_totp_status_api_auth_totp_get: {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        requestBody?: never;
+        responses: {
+            /** @description Successful Response */
+            200: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content: {
+                    "application/json": components["schemas"]["TotpStatusResponse"];
+                };
+            };
+        };
+    };
     publish_from_homeassistant_homeassistant_publish_post: {
         parameters: {
             query?: never;

frontend/src/auth/totp.ts

@@ -0,0 +1,77 @@
+/**
+ * TOTP API helpers (M4-T08).
+ *
+ * All calls go through the typed apiClient (openapi-fetch + generated schema).
+ * CSRF tokens are injected automatically by the client middleware for write
+ * endpoints; GET /api/auth/totp needs no CSRF (read-only).
+ *
+ * Security notes:
+ * - The TotpSetupResponse contains secret + recovery_codes in plaintext.
+ *   These are kept only in component state; callers must NOT write them to
+ *   localStorage, sessionStorage, or log them.
+ * - After the setup phase is over the data leaves memory when the component
+ *   re-renders (one-time display only).
+ */
+
+import apiClient, { ApiError } from '../api/client'
+import type { components } from '../api/schema.d.ts'
+
+export type TotpSetupResponse = components['schemas']['TotpSetupResponse']
+export type TotpStatusResponse = components['schemas']['TotpStatusResponse']
+
+// ---------------------------------------------------------------------------
+// getTotpStatus — GET /api/auth/totp
+// ---------------------------------------------------------------------------
+
+/** Return the current TOTP enabled state for the authenticated user. */
+export async function getTotpStatus(): Promise<TotpStatusResponse> {
+  const res = await apiClient.GET('/api/auth/totp')
+  if (!res.data) {
+    throw new ApiError(res.response.status, null)
+  }
+  return res.data
+}
+
+// ---------------------------------------------------------------------------
+// setupTotp — POST /api/auth/totp/setup
+// ---------------------------------------------------------------------------
+
+/**
+ * Generate a new pending TOTP secret + recovery codes.
+ * Returns the one-time plaintext values; caller is responsible for displaying
+ * them exactly once and never persisting them outside component state.
+ */
+export async function setupTotp(): Promise<TotpSetupResponse> {
+  const res = await apiClient.POST('/api/auth/totp/setup')
+  if (!res.data) {
+    throw new ApiError(res.response.status, null)
+  }
+  return res.data
+}
+
+// ---------------------------------------------------------------------------
+// enableTotp — POST /api/auth/totp/enable
+// ---------------------------------------------------------------------------
+
+/** Confirm TOTP setup with the current 6-digit code from the authenticator app. */
+export async function enableTotp(code: string): Promise<void> {
+  // The endpoint returns 204 on success; openapi-fetch yields { data: undefined }.
+  await apiClient.POST('/api/auth/totp/enable', { body: { code } })
+}
+
+// ---------------------------------------------------------------------------
+// disableTotp — POST /api/auth/totp/disable
+// ---------------------------------------------------------------------------
+
+/**
+ * Disable TOTP.  Exactly one of password or code must be provided.
+ * On success the backend clears the secret and all recovery codes.
+ */
+export async function disableTotp(opts: {
+  password?: string | null
+  code?: string | null
+}): Promise<void> {
+  await apiClient.POST('/api/auth/totp/disable', {
+    body: { password: opts.password ?? null, code: opts.code ?? null },
+  })
+}

frontend/src/pages/ConfigPage.tsx

@@ -35,6 +35,7 @@ import {
 } from '@mantine/core'
 import apiClient, { ApiError } from '../api/client'
 import type { components } from '../api/schema.d.ts'
+import { TotpSettings } from './TotpSettings'
 
 // ---------------------------------------------------------------------------
 // Types
@@ -393,6 +394,11 @@ export function ConfigPage() {
           </Text>
         </Stack>
       )}
+
+      {/* TOTP two-factor auth management */}
+      <Stack mt="xl">
+        <TotpSettings />
+      </Stack>
     </Container>
   )
 }

frontend/src/pages/LoginPage.test.tsx

@@ -1,16 +1,21 @@
 /**
- * Tests for LoginPage (M2-T07).
+ * Tests for LoginPage (M2-T07, extended M4-T08).
  *
  * Strategy: vi.mock the apiClient module so we can control POST /api/auth/login
  * responses without a real server.  We also mock useSession so tests can control
  * the authentication state.
  *
  * Coverage:
- * 1. Renders the login form.
+ * 1. Renders the login form (step 1).
- * 2. Successful login → invalidates session query + navigates.
+ * 2. Successful login (step 1) → invalidates session query + navigates.
  * 3. 401 bad credentials → shows inline error, does not navigate.
  * 4. Already-authenticated users visiting /login → redirected to '/'.
  * 5. Unexpected error → shows generic error message.
+ * 6. (M4-T08) 401 with totp_required → switches to step 2 TOTP form.
+ * 7. (M4-T08) Step 2 submit with correct code → success + navigate.
+ * 8. (M4-T08) Step 2 submit with wrong code → shows error, stays on step 2.
+ * 9. (M4-T08) Step 2 "Back to login" button → returns to step 1.
+ * 10. (M4-T08) 429 response → shows "too many attempts" message.
  */
 
 import { describe, it, expect, vi, beforeEach } from 'vitest'
@@ -86,6 +91,10 @@ describe('LoginPage', () => {
     mockUseSession.mockReturnValue({ status: 'unauthenticated', user: null })
   })
 
+  // -------------------------------------------------------------------------
+  // 1. Renders the form
+  // -------------------------------------------------------------------------
+
   it('renders the login form with username and password fields', () => {
     renderLogin()
     expect(screen.getByTestId('login-form')).toBeInTheDocument()
@@ -99,8 +108,11 @@ describe('LoginPage', () => {
     expect(screen.getByRole('heading', { name: /sign in/i })).toBeInTheDocument()
   })
 
+  // -------------------------------------------------------------------------
+  // 2. Successful login
+  // -------------------------------------------------------------------------
+
   it('navigates to "/" on successful login', async () => {
-    // Simulate a successful POST /api/auth/login response
     mockPost.mockResolvedValueOnce({
       data: { user: { username: 'admin', force_password_change: false }, csrf_token: 'tok123' },
       response: { status: 200, ok: true },
@@ -130,10 +142,14 @@ describe('LoginPage', () => {
     })
   })
 
+  // -------------------------------------------------------------------------
+  // 3. 401 bad credentials
+  // -------------------------------------------------------------------------
+
   it('shows inline error on 401 and does NOT navigate', async () => {
-    // Simulate 401: openapi-fetch returns { data: undefined, response: { status: 401 } }
     mockPost.mockResolvedValueOnce({
       data: undefined,
+      error: { detail: 'invalid username or password' },
       response: { status: 401, ok: false },
     })
 
@@ -154,6 +170,7 @@ describe('LoginPage', () => {
   it('does not include the password in the error message', async () => {
     mockPost.mockResolvedValueOnce({
       data: undefined,
+      error: { detail: 'invalid username or password' },
       response: { status: 401, ok: false },
     })
 
@@ -167,6 +184,10 @@ describe('LoginPage', () => {
     expect(screen.getByTestId('login-error')).not.toHaveTextContent('super-secret-password')
   })
 
+  // -------------------------------------------------------------------------
+  // 5. Unexpected error
+  // -------------------------------------------------------------------------
+
   it('shows generic error on unexpected network failure', async () => {
     mockPost.mockRejectedValueOnce(new Error('Network error'))
 
@@ -180,6 +201,10 @@ describe('LoginPage', () => {
     expect(screen.getByTestId('login-error')).toHaveTextContent(/login failed/i)
   })
 
+  // -------------------------------------------------------------------------
+  // 4. Already authenticated
+  // -------------------------------------------------------------------------
+
   it('redirects already-authenticated users to "/"', async () => {
     mockUseSession.mockReturnValue({
       status: 'authenticated',
@@ -192,4 +217,148 @@ describe('LoginPage', () => {
       expect(screen.getByTestId('home-page')).toBeInTheDocument()
     })
   })
+
+  // -------------------------------------------------------------------------
+  // 6. (M4-T08) 401 with totp_required → step 2
+  // -------------------------------------------------------------------------
+
+  it('switches to TOTP step on 401 with totp_required=true', async () => {
+    mockPost.mockResolvedValueOnce({
+      data: undefined,
+      error: { detail: { totp_required: true } },
+      response: { status: 401, ok: false },
+    })
+
+    renderLogin()
+    fillAndSubmit('admin', 'correct-password')
+
+    await waitFor(() => {
+      expect(screen.getByTestId('login-totp-form')).toBeInTheDocument()
+    })
+
+    // Step 1 form should be gone
+    expect(screen.queryByTestId('login-form')).not.toBeInTheDocument()
+    // TOTP code input should be visible
+    expect(screen.getByTestId('totp-code-input')).toBeInTheDocument()
+  })
+
+  // -------------------------------------------------------------------------
+  // 7. (M4-T08) Step 2 correct code → success
+  // -------------------------------------------------------------------------
+
+  it('completes login on correct TOTP code (step 2)', async () => {
+    // Step 1: trigger totp_required
+    mockPost.mockResolvedValueOnce({
+      data: undefined,
+      error: { detail: { totp_required: true } },
+      response: { status: 401, ok: false },
+    })
+    // Step 2: success
+    mockPost.mockResolvedValueOnce({
+      data: { user: { username: 'admin', force_password_change: false }, csrf_token: 'tok456' },
+      response: { status: 200, ok: true },
+    })
+
+    renderLogin()
+    fillAndSubmit('admin', 'correct-password')
+
+    await waitFor(() => {
+      expect(screen.getByTestId('login-totp-form')).toBeInTheDocument()
+    })
+
+    // Enter the TOTP code and submit step 2
+    fireEvent.change(screen.getByTestId('totp-code-input'), { target: { value: '123456' } })
+    fireEvent.submit(screen.getByTestId('login-totp-form'))
+
+    await waitFor(() => {
+      expect(screen.getByTestId('home-page')).toBeInTheDocument()
+    })
+
+    // Check step 2 request included totp_code
+    expect(mockPost).toHaveBeenCalledWith('/api/auth/login', {
+      body: { username: 'admin', password: 'correct-password', totp_code: '123456' },
+    })
+  })
+
+  // -------------------------------------------------------------------------
+  // 8. (M4-T08) Step 2 wrong code → error, stays on step 2
+  // -------------------------------------------------------------------------
+
+  it('shows error on wrong TOTP code and stays on step 2', async () => {
+    // Step 1: totp_required
+    mockPost.mockResolvedValueOnce({
+      data: undefined,
+      error: { detail: { totp_required: true } },
+      response: { status: 401, ok: false },
+    })
+    // Step 2: wrong code → 401
+    mockPost.mockResolvedValueOnce({
+      data: undefined,
+      error: { detail: 'invalid totp code' },
+      response: { status: 401, ok: false },
+    })
+
+    renderLogin()
+    fillAndSubmit('admin', 'correct-password')
+
+    await waitFor(() => {
+      expect(screen.getByTestId('login-totp-form')).toBeInTheDocument()
+    })
+
+    fireEvent.change(screen.getByTestId('totp-code-input'), { target: { value: '000000' } })
+    fireEvent.submit(screen.getByTestId('login-totp-form'))
+
+    await waitFor(() => {
+      expect(screen.getByTestId('login-error')).toBeInTheDocument()
+    })
+
+    // Should still be on TOTP form
+    expect(screen.getByTestId('login-totp-form')).toBeInTheDocument()
+    expect(screen.getByTestId('login-error')).toHaveTextContent(/incorrect verification code/i)
+  })
+
+  // -------------------------------------------------------------------------
+  // 9. (M4-T08) Step 2 "Back to login" → returns to step 1
+  // -------------------------------------------------------------------------
+
+  it('returns to step 1 when "Back to login" is clicked', async () => {
+    mockPost.mockResolvedValueOnce({
+      data: undefined,
+      error: { detail: { totp_required: true } },
+      response: { status: 401, ok: false },
+    })
+
+    renderLogin()
+    fillAndSubmit('admin', 'correct-password')
+
+    await waitFor(() => {
+      expect(screen.getByTestId('login-totp-form')).toBeInTheDocument()
+    })
+
+    fireEvent.click(screen.getByTestId('totp-back'))
+
+    await waitFor(() => {
+      expect(screen.getByTestId('login-form')).toBeInTheDocument()
+    })
+
+    expect(screen.queryByTestId('login-totp-form')).not.toBeInTheDocument()
+  })
+
+  // -------------------------------------------------------------------------
+  // 10. (M4-T08) 429 → "too many attempts" message
+  // -------------------------------------------------------------------------
+
+  it('shows "too many attempts" message on 429', async () => {
+    const { ApiError } = await import('../api/client')
+    mockPost.mockRejectedValueOnce(new ApiError(429, { detail: 'Too Many Requests' }))
+
+    renderLogin()
+    fillAndSubmit('admin', 'password')
+
+    await waitFor(() => {
+      expect(screen.getByTestId('login-error')).toBeInTheDocument()
+    })
+
+    expect(screen.getByTestId('login-error')).toHaveTextContent(/too many login attempts/i)
+  })
 })

frontend/src/pages/LoginPage.tsx

@@ -1,13 +1,21 @@
 /**
- * LoginPage — real login form (M2-T07).
+ * LoginPage — real login form (M2-T07, extended M4-T08).
  *
  * Behaviours:
- * - Renders a Mantine form with username + password fields.
+ * - Renders a Mantine form with username + password fields (step 1).
  * - On submit → POST /api/auth/login via apiClient (no CSRF needed; unauthenticated endpoint).
  * - On success → invalidate ['session'] so SessionProvider re-fetches, then navigate to the
  *   originally-requested route (from location.state.from) or fall back to '/'.
- * - On 401 (bad credentials) → show an inline error without leaking the password.
+ * - On 401 with totp_required=true → switch to step 2 (TOTP code entry).
+ * - On step 2 submit → POST /api/auth/login with {username, password, totp_code}.
+ * - On 401 plain (bad credentials / wrong TOTP code) → show inline error.
+ * - On 429 (rate-limited) → show "too many attempts, please try again later".
  * - Already-authenticated users visiting /login → redirect to '/'.
+ *
+ * Security:
+ * - Password is retained in state for the step-2 re-submission but is NEVER
+ *   written to localStorage, sessionStorage, or console.
+ * - totp_code / recovery codes are NEVER logged or persisted.
  */
 
 import { useState } from 'react'
@@ -23,9 +31,10 @@ import {
   Alert,
   Stack,
   Center,
+  Text,
 } from '@mantine/core'
 import { useSession } from '../auth/SessionProvider'
-import apiClient from '../api/client'
+import apiClient, { ApiError } from '../api/client'
 import { setCsrfToken } from '../api/csrf'
 
 // ---------------------------------------------------------------------------
@@ -36,6 +45,9 @@ interface LocationState {
   from?: { pathname: string }
 }
 
+/** The two UI states for the login flow. */
+type LoginStep = 'credentials' | 'totp'
+
 // ---------------------------------------------------------------------------
 // Component
 // ---------------------------------------------------------------------------
@@ -48,6 +60,8 @@ export function LoginPage() {
 
   const [username, setUsername] = useState('')
   const [password, setPassword] = useState('')
+  const [totpCode, setTotpCode] = useState('')
+  const [step, setStep] = useState<LoginStep>('credentials')
   const [error, setError] = useState<string | null>(null)
   const [loading, setLoading] = useState(false)
 
@@ -57,7 +71,11 @@ export function LoginPage() {
     return <Navigate to={from} replace />
   }
 
-  async function handleSubmit(e: React.FormEvent) {
+  // ---------------------------------------------------------------------------
+  // handleCredentialsSubmit — step 1
+  // ---------------------------------------------------------------------------
+
+  async function handleCredentialsSubmit(e: React.FormEvent) {
     e.preventDefault()
     setError(null)
     setLoading(true)
@@ -68,31 +86,84 @@ export function LoginPage() {
       })
 
       if (res.response.status === 401 || !res.data) {
-        // Bad credentials — do not leak the password in the message.
+        // Check if the backend wants a TOTP code.
+        // openapi-fetch puts 401 body in res.error (not res.data).
+        // The shape is { detail: { totp_required: true } } for TOTP-required,
+        // or { detail: "invalid username or password" } for bad credentials.
+        const detail = (res.error as { detail?: unknown } | undefined)?.detail
+        if (detail && typeof detail === 'object' && (detail as { totp_required?: boolean }).totp_required === true) {
+          // Transition to step 2 — TOTP entry. Password stays in state for re-send.
+          setStep('totp')
+          setTotpCode('')
+          return
+        }
+        // Ordinary bad credentials.
         setError('Incorrect username or password.')
         return
       }
 
-      // Success: store the CSRF token returned by login (same shape as session response).
+      // Success: store the CSRF token returned by login.
       if (res.data.csrf_token) {
         setCsrfToken(res.data.csrf_token)
       }
 
-      // Refresh session state: invalidate the ['session'] query so SessionProvider
+      // Refresh session state.
-      // picks up the new authenticated state (which may include force_password_change).
       await queryClient.invalidateQueries({ queryKey: ['session'] })
-
-      // Navigate to the originally-requested route or home.
       const from = (location.state as LocationState)?.from?.pathname ?? '/'
       navigate(from, { replace: true })
-    } catch {
+    } catch (err) {
-      // Any unexpected error (network, 5xx, etc.)
+      if (err instanceof ApiError && err.status === 429) {
+        setError('Too many login attempts. Please try again later.')
+      } else {
         setError('Login failed. Please try again.')
+      }
     } finally {
       setLoading(false)
     }
   }
 
+  // ---------------------------------------------------------------------------
+  // handleTotpSubmit — step 2
+  // ---------------------------------------------------------------------------
+
+  async function handleTotpSubmit(e: React.FormEvent) {
+    e.preventDefault()
+    setError(null)
+    setLoading(true)
+
+    try {
+      const res = await apiClient.POST('/api/auth/login', {
+        body: { username, password, totp_code: totpCode },
+      })
+
+      if (res.response.status === 401 || !res.data) {
+        // Wrong code (or expired) — stay on step 2 and show error.
+        setError('Incorrect verification code. Please try again.')
+        return
+      }
+
+      // Success.
+      if (res.data.csrf_token) {
+        setCsrfToken(res.data.csrf_token)
+      }
+      await queryClient.invalidateQueries({ queryKey: ['session'] })
+      const from = (location.state as LocationState)?.from?.pathname ?? '/'
+      navigate(from, { replace: true })
+    } catch (err) {
+      if (err instanceof ApiError && err.status === 429) {
+        setError('Too many login attempts. Please try again later.')
+      } else {
+        setError('Login failed. Please try again.')
+      }
+    } finally {
+      setLoading(false)
+    }
+  }
+
+  // ---------------------------------------------------------------------------
+  // Render
+  // ---------------------------------------------------------------------------
+
   return (
     <Center mih="100vh">
       <Container size="xs" w="100%">
@@ -107,7 +178,8 @@ export function LoginPage() {
             </Alert>
           )}
 
-          <form onSubmit={handleSubmit} data-testid="login-form">
+          {step === 'credentials' && (
+            <form onSubmit={handleCredentialsSubmit} data-testid="login-form">
               <Stack gap="md">
                 <TextInput
                   label="Username"
@@ -140,6 +212,51 @@ export function LoginPage() {
                 </Button>
               </Stack>
             </form>
+          )}
+
+          {step === 'totp' && (
+            <form onSubmit={handleTotpSubmit} data-testid="login-totp-form">
+              <Stack gap="md">
+                <Text size="sm" c="dimmed">
+                  Enter the 6-digit code from your authenticator app, or a recovery code.
+                </Text>
+
+                <TextInput
+                  label="Verification Code"
+                  placeholder="6-digit code or recovery code"
+                  value={totpCode}
+                  onChange={(e) => setTotpCode(e.currentTarget.value)}
+                  required
+                  autoComplete="one-time-code"
+                  inputMode="numeric"
+                  data-testid="totp-code-input"
+                />
+
+                <Button
+                  type="submit"
+                  fullWidth
+                  loading={loading}
+                  mt="sm"
+                  data-testid="totp-submit"
+                >
+                  Verify
+                </Button>
+
+                <Button
+                  variant="subtle"
+                  size="sm"
+                  onClick={() => {
+                    setStep('credentials')
+                    setError(null)
+                    setTotpCode('')
+                  }}
+                  data-testid="totp-back"
+                >
+                  Back to login
+                </Button>
+              </Stack>
+            </form>
+          )}
         </Paper>
       </Container>
     </Center>

frontend/src/pages/TotpSettings.test.tsx

@@ -0,0 +1,344 @@
+/**
+ * Tests for TotpSettings (M4-T08).
+ *
+ * Strategy: vi.mock the auth/totp module so we can control API responses
+ * without a real server. We also mock the api/client module for ApiError.
+ *
+ * Coverage:
+ * 1. Shows loading spinner while fetching TOTP status.
+ * 2. Shows error when TOTP status fetch fails.
+ * 3. Shows "Disabled" badge and enable button when TOTP is disabled.
+ * 4. Enable flow: clicking setup → shows QR code + recovery codes + confirm form.
+ * 5. Confirm with correct code → TOTP enabled (status refetched).
+ * 6. Confirm with wrong code → shows error, stays on confirm form.
+ * 7. Shows "Enabled" badge and disable form when TOTP is enabled.
+ * 8. Disable with password → TOTP disabled.
+ * 9. Disable with TOTP code → TOTP disabled.
+ * 10. Disable without providing password or code → shows validation error.
+ * 11. Setup error shows error message.
+ */
+
+import { describe, it, expect, vi, beforeEach } from 'vitest'
+import { screen, waitFor, fireEvent } from '@testing-library/react'
+import { renderWithProviders } from '../test-utils'
+import { TotpSettings } from './TotpSettings'
+
+// ---------------------------------------------------------------------------
+// Mock auth/totp
+// ---------------------------------------------------------------------------
+
+const mockGetTotpStatus = vi.fn()
+const mockSetupTotp = vi.fn()
+const mockEnableTotp = vi.fn()
+const mockDisableTotp = vi.fn()
+
+vi.mock('../auth/totp', () => ({
+  getTotpStatus: (...args: unknown[]) => mockGetTotpStatus(...args),
+  setupTotp: (...args: unknown[]) => mockSetupTotp(...args),
+  enableTotp: (...args: unknown[]) => mockEnableTotp(...args),
+  disableTotp: (...args: unknown[]) => mockDisableTotp(...args),
+}))
+
+// ---------------------------------------------------------------------------
+// Mock api/client (for ApiError)
+// ---------------------------------------------------------------------------
+
+vi.mock('../api/client', () => ({
+  default: {
+    GET: vi.fn(),
+    POST: vi.fn(),
+    PUT: vi.fn(),
+  },
+  ApiError: class ApiError extends Error {
+    status: number
+    body: unknown
+    constructor(status: number, body: unknown) {
+      super(`API error ${status}`)
+      this.name = 'ApiError'
+      this.status = status
+      this.body = body
+    }
+  },
+  registerLoginRedirect: vi.fn(),
+}))
+
+// ---------------------------------------------------------------------------
+// Fixture
+// ---------------------------------------------------------------------------
+
+const MOCK_SETUP_DATA = {
+  secret: 'JBSWY3DPEHPK3PXP',
+  otpauth_uri: 'otpauth://totp/TestApp:admin?secret=JBSWY3DPEHPK3PXP&issuer=TestApp',
+  recovery_codes: ['abcd-1234', 'efgh-5678', 'ijkl-9012'],
+}
+
+// ---------------------------------------------------------------------------
+// Helpers
+// ---------------------------------------------------------------------------
+
+function renderTotp() {
+  return renderWithProviders(<TotpSettings />, { initialPath: '/config' })
+}
+
+// ---------------------------------------------------------------------------
+// Tests
+// ---------------------------------------------------------------------------
+
+describe('TotpSettings', () => {
+  beforeEach(() => {
+    vi.clearAllMocks()
+  })
+
+  // -------------------------------------------------------------------------
+  // 1. Loading state
+  // -------------------------------------------------------------------------
+
+  it('shows loading spinner while fetching TOTP status', async () => {
+    // Never resolves during this test
+    mockGetTotpStatus.mockReturnValue(new Promise(() => {}))
+    renderTotp()
+    await waitFor(() => {
+      expect(screen.getByTestId('totp-status-loading')).toBeInTheDocument()
+    })
+  })
+
+  // -------------------------------------------------------------------------
+  // 2. Error state
+  // -------------------------------------------------------------------------
+
+  it('shows error when TOTP status fetch fails', async () => {
+    mockGetTotpStatus.mockRejectedValue(new Error('Network error'))
+    renderTotp()
+    await waitFor(() => {
+      expect(screen.getByTestId('totp-status-error')).toBeInTheDocument()
+    })
+  })
+
+  // -------------------------------------------------------------------------
+  // 3. Disabled state
+  // -------------------------------------------------------------------------
+
+  it('shows Disabled badge and enable button when TOTP is disabled', async () => {
+    mockGetTotpStatus.mockResolvedValue({ enabled: false })
+    renderTotp()
+
+    await waitFor(() => {
+      expect(screen.getByTestId('totp-status-badge')).toBeInTheDocument()
+    })
+
+    expect(screen.getByTestId('totp-status-badge')).toHaveTextContent(/disabled/i)
+    expect(screen.getByTestId('totp-setup-button')).toBeInTheDocument()
+  })
+
+  // -------------------------------------------------------------------------
+  // 4. Enable flow: setup shows QR + recovery codes + confirm form
+  // -------------------------------------------------------------------------
+
+  it('shows QR code, secret, and recovery codes after clicking setup', async () => {
+    mockGetTotpStatus.mockResolvedValue({ enabled: false })
+    mockSetupTotp.mockResolvedValue(MOCK_SETUP_DATA)
+
+    renderTotp()
+
+    await waitFor(() => {
+      expect(screen.getByTestId('totp-setup-button')).toBeInTheDocument()
+    })
+
+    fireEvent.click(screen.getByTestId('totp-setup-button'))
+
+    await waitFor(() => {
+      expect(screen.getByTestId('totp-qr-code')).toBeInTheDocument()
+    })
+
+    expect(screen.getByTestId('totp-secret')).toHaveTextContent('JBSWY3DPEHPK3PXP')
+    expect(screen.getByTestId('totp-recovery-codes')).toBeInTheDocument()
+    expect(screen.getByTestId('totp-recovery-codes')).toHaveTextContent('abcd-1234')
+    expect(screen.getByTestId('totp-confirm-form')).toBeInTheDocument()
+    expect(screen.getByTestId('totp-confirm-input')).toBeInTheDocument()
+  })
+
+  // -------------------------------------------------------------------------
+  // 5. Confirm with correct code → enabled
+  // -------------------------------------------------------------------------
+
+  it('enables TOTP after confirming with correct code', async () => {
+    mockGetTotpStatus
+      .mockResolvedValueOnce({ enabled: false })
+      .mockResolvedValueOnce({ enabled: true })
+    mockSetupTotp.mockResolvedValue(MOCK_SETUP_DATA)
+    mockEnableTotp.mockResolvedValue(undefined)
+
+    renderTotp()
+
+    await waitFor(() => {
+      expect(screen.getByTestId('totp-setup-button')).toBeInTheDocument()
+    })
+
+    fireEvent.click(screen.getByTestId('totp-setup-button'))
+
+    await waitFor(() => {
+      expect(screen.getByTestId('totp-confirm-form')).toBeInTheDocument()
+    })
+
+    fireEvent.change(screen.getByTestId('totp-confirm-input'), { target: { value: '123456' } })
+    fireEvent.submit(screen.getByTestId('totp-confirm-form'))
+
+    await waitFor(() => {
+      expect(mockEnableTotp).toHaveBeenCalledWith('123456')
+    })
+
+    // After enable the status refetches → shows enabled
+    await waitFor(() => {
+      expect(screen.getByTestId('totp-status-badge')).toHaveTextContent(/enabled/i)
+    })
+  })
+
+  // -------------------------------------------------------------------------
+  // 6. Confirm with wrong code → error
+  // -------------------------------------------------------------------------
+
+  it('shows error on wrong TOTP code during enable', async () => {
+    mockGetTotpStatus.mockResolvedValue({ enabled: false })
+    mockSetupTotp.mockResolvedValue(MOCK_SETUP_DATA)
+
+    const { ApiError } = await import('../api/client')
+    mockEnableTotp.mockRejectedValue(new ApiError(400, { detail: 'invalid code' }))
+
+    renderTotp()
+
+    await waitFor(() => {
+      expect(screen.getByTestId('totp-setup-button')).toBeInTheDocument()
+    })
+
+    fireEvent.click(screen.getByTestId('totp-setup-button'))
+
+    await waitFor(() => {
+      expect(screen.getByTestId('totp-confirm-form')).toBeInTheDocument()
+    })
+
+    fireEvent.change(screen.getByTestId('totp-confirm-input'), { target: { value: '000000' } })
+    fireEvent.submit(screen.getByTestId('totp-confirm-form'))
+
+    await waitFor(() => {
+      expect(screen.getByTestId('totp-confirm-error')).toBeInTheDocument()
+    })
+
+    // Should still show confirm form (not enabled)
+    expect(screen.getByTestId('totp-confirm-form')).toBeInTheDocument()
+  })
+
+  // -------------------------------------------------------------------------
+  // 7. Enabled state
+  // -------------------------------------------------------------------------
+
+  it('shows Enabled badge and disable form when TOTP is enabled', async () => {
+    mockGetTotpStatus.mockResolvedValue({ enabled: true })
+    renderTotp()
+
+    await waitFor(() => {
+      expect(screen.getByTestId('totp-status-badge')).toBeInTheDocument()
+    })
+
+    expect(screen.getByTestId('totp-status-badge')).toHaveTextContent(/enabled/i)
+    expect(screen.getByTestId('totp-disable-form')).toBeInTheDocument()
+    expect(screen.getByTestId('totp-disable-button')).toBeInTheDocument()
+  })
+
+  // -------------------------------------------------------------------------
+  // 8. Disable with password
+  // -------------------------------------------------------------------------
+
+  it('disables TOTP when correct password is provided', async () => {
+    mockGetTotpStatus
+      .mockResolvedValueOnce({ enabled: true })
+      .mockResolvedValueOnce({ enabled: false })
+    mockDisableTotp.mockResolvedValue(undefined)
+
+    renderTotp()
+
+    await waitFor(() => {
+      expect(screen.getByTestId('totp-disable-form')).toBeInTheDocument()
+    })
+
+    fireEvent.change(screen.getByTestId('totp-disable-password'), {
+      target: { value: 'my-password' },
+    })
+    fireEvent.submit(screen.getByTestId('totp-disable-form'))
+
+    await waitFor(() => {
+      expect(mockDisableTotp).toHaveBeenCalledWith({ password: 'my-password', code: null })
+    })
+
+    await waitFor(() => {
+      expect(screen.getByTestId('totp-status-badge')).toHaveTextContent(/disabled/i)
+    })
+  })
+
+  // -------------------------------------------------------------------------
+  // 9. Disable with TOTP code
+  // -------------------------------------------------------------------------
+
+  it('disables TOTP when correct TOTP code is provided', async () => {
+    mockGetTotpStatus
+      .mockResolvedValueOnce({ enabled: true })
+      .mockResolvedValueOnce({ enabled: false })
+    mockDisableTotp.mockResolvedValue(undefined)
+
+    renderTotp()
+
+    await waitFor(() => {
+      expect(screen.getByTestId('totp-disable-form')).toBeInTheDocument()
+    })
+
+    fireEvent.change(screen.getByTestId('totp-disable-code'), { target: { value: '654321' } })
+    fireEvent.submit(screen.getByTestId('totp-disable-form'))
+
+    await waitFor(() => {
+      expect(mockDisableTotp).toHaveBeenCalledWith({ password: null, code: '654321' })
+    })
+  })
+
+  // -------------------------------------------------------------------------
+  // 10. Disable without credentials → validation error
+  // -------------------------------------------------------------------------
+
+  it('shows validation error when disable submitted without password or code', async () => {
+    mockGetTotpStatus.mockResolvedValue({ enabled: true })
+
+    renderTotp()
+
+    await waitFor(() => {
+      expect(screen.getByTestId('totp-disable-form')).toBeInTheDocument()
+    })
+
+    // Submit without filling in anything
+    fireEvent.submit(screen.getByTestId('totp-disable-form'))
+
+    await waitFor(() => {
+      expect(screen.getByTestId('totp-disable-error')).toBeInTheDocument()
+    })
+
+    expect(mockDisableTotp).not.toHaveBeenCalled()
+  })
+
+  // -------------------------------------------------------------------------
+  // 11. Setup error
+  // -------------------------------------------------------------------------
+
+  it('shows error when TOTP setup request fails', async () => {
+    mockGetTotpStatus.mockResolvedValue({ enabled: false })
+    mockSetupTotp.mockRejectedValue(new Error('Network error'))
+
+    renderTotp()
+
+    await waitFor(() => {
+      expect(screen.getByTestId('totp-setup-button')).toBeInTheDocument()
+    })
+
+    fireEvent.click(screen.getByTestId('totp-setup-button'))
+
+    await waitFor(() => {
+      expect(screen.getByTestId('totp-enable-error')).toBeInTheDocument()
+    })
+  })
+})

frontend/src/pages/TotpSettings.tsx

@@ -0,0 +1,360 @@
+/**
+ * TotpSettings — TOTP two-factor auth management panel (M4-T08).
+ *
+ * Intended to be embedded in the settings/config area of the app.
+ * Handles the full TOTP lifecycle:
+ *   - Status probe  : GET /api/auth/totp
+ *   - Enable flow   : setup → show QR code + one-time recovery codes → confirm with code
+ *   - Disable flow  : prompt for current password or TOTP code → disable
+ *
+ * Security notes:
+ * - secret and recovery_codes from TotpSetupResponse are held ONLY in component
+ *   state; they are NEVER written to localStorage, sessionStorage, or console.
+ * - Recovery codes are displayed exactly once (during the setup flow); after the
+ *   user confirms with a code and TOTP is enabled the data is discarded.
+ * - Passwords typed in the disable form are NEVER logged or persisted.
+ */
+
+import { useState, useCallback } from 'react'
+import { useQuery, useQueryClient } from '@tanstack/react-query'
+import {
+  Paper,
+  Title,
+  Text,
+  Button,
+  Alert,
+  Stack,
+  Group,
+  TextInput,
+  PasswordInput,
+  Divider,
+  Badge,
+  Loader,
+  Center,
+  Code,
+  List,
+} from '@mantine/core'
+import { QRCodeSVG } from 'qrcode.react'
+import { getTotpStatus, setupTotp, enableTotp, disableTotp } from '../auth/totp'
+import { ApiError } from '../api/client'
+import type { TotpSetupResponse } from '../auth/totp'
+
+// ---------------------------------------------------------------------------
+// Sub-panel: Enable flow
+// ---------------------------------------------------------------------------
+
+interface EnablePanelProps {
+  onEnabled: () => void
+}
+
+function EnablePanel({ onEnabled }: EnablePanelProps) {
+  const [setupData, setSetupData] = useState<TotpSetupResponse | null>(null)
+  const [confirmCode, setConfirmCode] = useState('')
+  const [status, setStatus] = useState<'idle' | 'setup-loading' | 'confirm' | 'confirming' | 'error'>('idle')
+  const [errorMsg, setErrorMsg] = useState<string | null>(null)
+
+  async function handleSetup() {
+    setStatus('setup-loading')
+    setErrorMsg(null)
+    try {
+      const data = await setupTotp()
+      // data contains one-time secret + recovery_codes — keep only in state.
+      setSetupData(data)
+      setStatus('confirm')
+    } catch {
+      setErrorMsg('Failed to start TOTP setup. Please try again.')
+      setStatus('error')
+    }
+  }
+
+  async function handleEnable(e: React.FormEvent) {
+    e.preventDefault()
+    setErrorMsg(null)
+    setStatus('confirming')
+    try {
+      await enableTotp(confirmCode)
+      // Discard secret + recovery codes from memory now that TOTP is enabled.
+      setSetupData(null)
+      setConfirmCode('')
+      onEnabled()
+    } catch (err) {
+      let msg = 'Failed to enable TOTP. Check the code and try again.'
+      if (err instanceof ApiError && err.status === 400) {
+        msg = 'Incorrect code or no pending setup. Please start the setup again.'
+      }
+      setErrorMsg(msg)
+      setStatus('confirm')
+    }
+  }
+
+  // ---------------------------------------------------------------------------
+  // Render: idle — show "Start Setup" button
+  // ---------------------------------------------------------------------------
+
+  if (status === 'idle' || status === 'error') {
+    return (
+      <Stack gap="sm">
+        <Text size="sm">
+          Two-factor authentication is not enabled. Enable it to require a verification code
+          each time you sign in.
+        </Text>
+        {errorMsg && (
+          <Alert color="red" data-testid="totp-enable-error">
+            {errorMsg}
+          </Alert>
+        )}
+        <Button onClick={handleSetup} data-testid="totp-setup-button">
+          Enable Two-Factor Authentication
+        </Button>
+      </Stack>
+    )
+  }
+
+  // ---------------------------------------------------------------------------
+  // Render: setup-loading
+  // ---------------------------------------------------------------------------
+
+  if (status === 'setup-loading') {
+    return (
+      <Center>
+        <Loader data-testid="totp-setup-loading" />
+      </Center>
+    )
+  }
+
+  // ---------------------------------------------------------------------------
+  // Render: confirm — show QR code, secret, recovery codes, confirmation form
+  // ---------------------------------------------------------------------------
+
+  if ((status === 'confirm' || status === 'confirming') && setupData) {
+    return (
+      <Stack gap="md">
+        <Alert color="yellow" data-testid="totp-setup-warning">
+          Save your recovery codes now — they will only be shown once.
+        </Alert>
+
+        <Text size="sm" fw={500}>
+          1. Scan this QR code with your authenticator app (e.g. Google Authenticator, Authy):
+        </Text>
+
+        <Center data-testid="totp-qr-code">
+          <QRCodeSVG value={setupData.otpauth_uri} size={180} />
+        </Center>
+
+        <Text size="sm" c="dimmed" ta="center">
+          Or enter the key manually:
+        </Text>
+        <Code block data-testid="totp-secret">
+          {setupData.secret}
+        </Code>
+
+        <Divider />
+
+        <Text size="sm" fw={500}>
+          2. Save these recovery codes in a secure place. Each code can be used once
+          if you lose access to your authenticator:
+        </Text>
+        <Paper withBorder p="sm" data-testid="totp-recovery-codes">
+          <List spacing="xs" size="sm">
+            {setupData.recovery_codes.map((code) => (
+              <List.Item key={code}>
+                <Code>{code}</Code>
+              </List.Item>
+            ))}
+          </List>
+        </Paper>
+
+        <Divider />
+
+        <Text size="sm" fw={500}>
+          3. Enter the 6-digit code from your authenticator app to confirm setup:
+        </Text>
+
+        {errorMsg && (
+          <Alert color="red" data-testid="totp-confirm-error">
+            {errorMsg}
+          </Alert>
+        )}
+
+        <form onSubmit={handleEnable} data-testid="totp-confirm-form">
+          <Stack gap="sm">
+            <TextInput
+              label="Verification Code"
+              placeholder="6-digit code"
+              value={confirmCode}
+              onChange={(e) => setConfirmCode(e.currentTarget.value)}
+              required
+              autoComplete="one-time-code"
+              inputMode="numeric"
+              data-testid="totp-confirm-input"
+            />
+            <Button
+              type="submit"
+              loading={status === 'confirming'}
+              data-testid="totp-confirm-button"
+            >
+              Confirm and Enable
+            </Button>
+          </Stack>
+        </form>
+      </Stack>
+    )
+  }
+
+  return null
+}
+
+// ---------------------------------------------------------------------------
+// Sub-panel: Disable flow
+// ---------------------------------------------------------------------------
+
+interface DisablePanelProps {
+  onDisabled: () => void
+}
+
+function DisablePanel({ onDisabled }: DisablePanelProps) {
+  const [disablePassword, setDisablePassword] = useState('')
+  const [disableCode, setDisableCode] = useState('')
+  const [errorMsg, setErrorMsg] = useState<string | null>(null)
+  const [isSubmitting, setIsSubmitting] = useState(false)
+
+  async function handleDisable(e: React.FormEvent) {
+    e.preventDefault()
+    setErrorMsg(null)
+
+    // At least one of password or code must be provided.
+    if (!disablePassword && !disableCode) {
+      setErrorMsg('Please provide your current password or a TOTP code.')
+      return
+    }
+
+    setIsSubmitting(true)
+    try {
+      await disableTotp({
+        password: disablePassword || null,
+        code: disableCode || null,
+      })
+      setDisablePassword('')
+      setDisableCode('')
+      onDisabled()
+    } catch (err) {
+      let msg = 'Failed to disable TOTP. Please try again.'
+      if (err instanceof ApiError && err.status === 400) {
+        msg = 'Incorrect password or code. Please try again.'
+      }
+      setErrorMsg(msg)
+    } finally {
+      setIsSubmitting(false)
+    }
+  }
+
+  return (
+    <Stack gap="md">
+      <Alert color="blue" data-testid="totp-enabled-notice">
+        Two-factor authentication is enabled. To disable it, enter your current password
+        or a valid verification code.
+      </Alert>
+
+      {errorMsg && (
+        <Alert color="red" data-testid="totp-disable-error">
+          {errorMsg}
+        </Alert>
+      )}
+
+      <form onSubmit={handleDisable} data-testid="totp-disable-form">
+        <Stack gap="sm">
+          <PasswordInput
+            label="Current Password"
+            placeholder="Enter your password (or use a TOTP code below)"
+            value={disablePassword}
+            onChange={(e) => setDisablePassword(e.currentTarget.value)}
+            autoComplete="current-password"
+            data-testid="totp-disable-password"
+          />
+
+          <Text size="xs" c="dimmed" ta="center">
+            — or —
+          </Text>
+
+          <TextInput
+            label="TOTP Code"
+            placeholder="6-digit code from your authenticator"
+            value={disableCode}
+            onChange={(e) => setDisableCode(e.currentTarget.value)}
+            autoComplete="one-time-code"
+            inputMode="numeric"
+            data-testid="totp-disable-code"
+          />
+
+          <Button
+            type="submit"
+            color="red"
+            variant="outline"
+            loading={isSubmitting}
+            data-testid="totp-disable-button"
+          >
+            Disable Two-Factor Authentication
+          </Button>
+        </Stack>
+      </form>
+    </Stack>
+  )
+}
+
+// ---------------------------------------------------------------------------
+// TotpSettings — main exported component
+// ---------------------------------------------------------------------------
+
+export function TotpSettings() {
+  const queryClient = useQueryClient()
+
+  const {
+    data: totpStatus,
+    isLoading,
+    isError,
+  } = useQuery({
+    queryKey: ['totp-status'],
+    queryFn: getTotpStatus,
+    // No automatic retries — the test QueryClient is configured with retry: false.
+    // Network errors are shown immediately via the error state.
+    retry: false,
+  })
+
+  // Invalidate the TOTP status query to refetch after enable/disable.
+  const refetchStatus = useCallback(() => {
+    void queryClient.invalidateQueries({ queryKey: ['totp-status'] })
+  }, [queryClient])
+
+  return (
+    <Paper withBorder p="md" radius="md" data-testid="totp-settings">
+      <Group justify="space-between" mb="md" wrap="nowrap">
+        <Title order={4}>Two-Factor Authentication (TOTP)</Title>
+        {totpStatus && (
+          <Badge
+            color={totpStatus.enabled ? 'green' : 'gray'}
+            variant="outline"
+            data-testid="totp-status-badge"
+          >
+            {totpStatus.enabled ? 'Enabled' : 'Disabled'}
+          </Badge>
+        )}
+      </Group>
+
+      {isLoading && (
+        <Center>
+          <Loader data-testid="totp-status-loading" />
+        </Center>
+      )}
+
+      {isError && (
+        <Alert color="red" data-testid="totp-status-error">
+          Failed to load TOTP status. Please refresh the page.
+        </Alert>
+      )}
+
+      {totpStatus && !totpStatus.enabled && <EnablePanel onEnabled={refetchStatus} />}
+
+      {totpStatus && totpStatus.enabled && <DisablePanel onDisabled={refetchStatus} />}
+    </Paper>
+  )
+}

openapi/openapi.json

@@ -653,7 +653,7 @@
           "api-session"
         ],
         "summary": "Post Login",
-        "description": "Authenticate with username and password.\n\nOn success, sets an HttpOnly session cookie and returns the session user + CSRF token.\nOn failure, returns 401 with no cookie set.\nNo X-CSRF-Token required (unauthenticated endpoint).",
+        "description": "Authenticate with username and password.\n\nOn success, sets an HttpOnly session cookie and returns the session user + CSRF token.\nOn failure, returns 401 with no cookie set.\nRepeated failures trigger exponential back-off (429 + Retry-After).\nNo X-CSRF-Token required (unauthenticated endpoint).",
         "operationId": "post_login_api_auth_login_post",
         "requestBody": {
           "content": {
@@ -795,6 +795,184 @@
         }
       }
     },
+    "/api/auth/totp/setup": {
+      "post": {
+        "tags": [
+          "api-session"
+        ],
+        "summary": "Post Totp Setup",
+        "description": "Generate a new pending TOTP secret, otpauth URI, and one-time recovery codes.\n\nThe secret is stored in the DB but TOTP is NOT yet enabled (totp_enabled stays\nFalse until the user confirms with POST /api/auth/totp/enable).\n\nRecovery codes are returned here as plaintext exactly once; their Argon2 hashes\nare persisted immediately so enable only needs to flip the enabled flag.\n\nRepeating this call replaces any prior pending secret and regenerates codes.\n\nRequires: session cookie + X-CSRF-Token.",
+        "operationId": "post_totp_setup_api_auth_totp_setup_post",
+        "parameters": [
+          {
+            "name": "X-CSRF-Token",
+            "in": "header",
+            "required": false,
+            "schema": {
+              "anyOf": [
+                {
+                  "type": "string"
+                },
+                {
+                  "type": "null"
+                }
+              ],
+              "title": "X-Csrf-Token"
+            }
+          }
+        ],
+        "responses": {
+          "200": {
+            "description": "Successful Response",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/TotpSetupResponse"
+                }
+              }
+            }
+          },
+          "422": {
+            "description": "Validation Error",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/HTTPValidationError"
+                }
+              }
+            }
+          }
+        }
+      }
+    },
+    "/api/auth/totp/enable": {
+      "post": {
+        "tags": [
+          "api-session"
+        ],
+        "summary": "Post Totp Enable",
+        "description": "Enable TOTP by confirming with the current 6-digit code from the authenticator app.\n\nRequires a prior call to POST /api/auth/totp/setup (so that a pending secret\nexists).  On success, totp_enabled becomes True.\n\nReturns 400 if the code is wrong or there is no pending secret.\nRequires: session cookie + X-CSRF-Token.",
+        "operationId": "post_totp_enable_api_auth_totp_enable_post",
+        "parameters": [
+          {
+            "name": "X-CSRF-Token",
+            "in": "header",
+            "required": false,
+            "schema": {
+              "anyOf": [
+                {
+                  "type": "string"
+                },
+                {
+                  "type": "null"
+                }
+              ],
+              "title": "X-Csrf-Token"
+            }
+          }
+        ],
+        "requestBody": {
+          "required": true,
+          "content": {
+            "application/json": {
+              "schema": {
+                "$ref": "#/components/schemas/TotpEnableRequest"
+              }
+            }
+          }
+        },
+        "responses": {
+          "204": {
+            "description": "Successful Response"
+          },
+          "422": {
+            "description": "Validation Error",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/HTTPValidationError"
+                }
+              }
+            }
+          }
+        }
+      }
+    },
+    "/api/auth/totp/disable": {
+      "post": {
+        "tags": [
+          "api-session"
+        ],
+        "summary": "Post Totp Disable",
+        "description": "Disable TOTP.  The caller must provide exactly one of:\n- ``password``: the user's current login password, OR\n- ``code``: the current 6-digit TOTP code.\n\nOn success: totp_enabled=False, totp_secret cleared, all recovery codes deleted.\nReturns 400 if neither credential matches or neither is provided.\nRequires: session cookie + X-CSRF-Token.",
+        "operationId": "post_totp_disable_api_auth_totp_disable_post",
+        "parameters": [
+          {
+            "name": "X-CSRF-Token",
+            "in": "header",
+            "required": false,
+            "schema": {
+              "anyOf": [
+                {
+                  "type": "string"
+                },
+                {
+                  "type": "null"
+                }
+              ],
+              "title": "X-Csrf-Token"
+            }
+          }
+        ],
+        "requestBody": {
+          "required": true,
+          "content": {
+            "application/json": {
+              "schema": {
+                "$ref": "#/components/schemas/TotpDisableRequest"
+              }
+            }
+          }
+        },
+        "responses": {
+          "204": {
+            "description": "Successful Response"
+          },
+          "422": {
+            "description": "Validation Error",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/HTTPValidationError"
+                }
+              }
+            }
+          }
+        }
+      }
+    },
+    "/api/auth/totp": {
+      "get": {
+        "tags": [
+          "api-session"
+        ],
+        "summary": "Get Totp Status",
+        "description": "Return the current TOTP status for the authenticated user.\n\nResponse contains only ``{\"enabled\": bool}``.\nSecret and recovery codes are NEVER returned here.\nRequires: session cookie only (no CSRF — read-only).",
+        "operationId": "get_totp_status_api_auth_totp_get",
+        "responses": {
+          "200": {
+            "description": "Successful Response",
+            "content": {
+              "application/json": {
+                "schema": {
+                  "$ref": "#/components/schemas/TotpStatusResponse"
+                }
+              }
+            }
+          }
+        }
+      }
+    },
     "/homeassistant/publish": {
       "post": {
         "tags": [
@@ -1169,6 +1347,17 @@
           "password": {
             "type": "string",
             "title": "Password"
+          },
+          "totp_code": {
+            "anyOf": [
+              {
+                "type": "string"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "title": "Totp Code"
           }
         },
         "type": "object",
@@ -1549,6 +1738,90 @@
         ],
         "title": "StatusResponse"
       },
+      "TotpDisableRequest": {
+        "properties": {
+          "password": {
+            "anyOf": [
+              {
+                "type": "string"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "title": "Password"
+          },
+          "code": {
+            "anyOf": [
+              {
+                "type": "string"
+              },
+              {
+                "type": "null"
+              }
+            ],
+            "title": "Code"
+          }
+        },
+        "type": "object",
+        "title": "TotpDisableRequest",
+        "description": "Disable TOTP by proving identity.\n\nExactly one of ``password`` or ``code`` must be provided."
+      },
+      "TotpEnableRequest": {
+        "properties": {
+          "code": {
+            "type": "string",
+            "title": "Code"
+          }
+        },
+        "type": "object",
+        "required": [
+          "code"
+        ],
+        "title": "TotpEnableRequest",
+        "description": "The user confirms setup by providing the 6-digit TOTP code."
+      },
+      "TotpSetupResponse": {
+        "properties": {
+          "secret": {
+            "type": "string",
+            "title": "Secret"
+          },
+          "otpauth_uri": {
+            "type": "string",
+            "title": "Otpauth Uri"
+          },
+          "recovery_codes": {
+            "items": {
+              "type": "string"
+            },
+            "type": "array",
+            "title": "Recovery Codes"
+          }
+        },
+        "type": "object",
+        "required": [
+          "secret",
+          "otpauth_uri",
+          "recovery_codes"
+        ],
+        "title": "TotpSetupResponse",
+        "description": "Returned once after a setup call.\n\n``secret`` and ``recovery_codes`` are **one-time plaintext values**.\nThey are never returned again by any subsequent API call.\nThe frontend must display and instruct the user to save them before confirming."
+      },
+      "TotpStatusResponse": {
+        "properties": {
+          "enabled": {
+            "type": "boolean",
+            "title": "Enabled"
+          }
+        },
+        "type": "object",
+        "required": [
+          "enabled"
+        ],
+        "title": "TotpStatusResponse",
+        "description": "Minimal status response — never exposes secret or recovery codes."
+      },
       "ValidationError": {
         "properties": {
           "loc": {

openapi/openapi.yaml

@@ -480,6 +480,8 @@ paths:
 
         On failure, returns 401 with no cookie set.
 
+        Repeated failures trigger exponential back-off (429 + Retry-After).
+
         No X-CSRF-Token required (unauthenticated endpoint).'
       operationId: post_login_api_auth_login_post
       requestBody:
@@ -575,6 +577,157 @@ paths:
             application/json:
               schema:
                 $ref: '#/components/schemas/HTTPValidationError'
+  /api/auth/totp/setup:
+    post:
+      tags:
+      - api-session
+      summary: Post Totp Setup
+      description: 'Generate a new pending TOTP secret, otpauth URI, and one-time
+        recovery codes.
+
+
+        The secret is stored in the DB but TOTP is NOT yet enabled (totp_enabled stays
+
+        False until the user confirms with POST /api/auth/totp/enable).
+
+
+        Recovery codes are returned here as plaintext exactly once; their Argon2 hashes
+
+        are persisted immediately so enable only needs to flip the enabled flag.
+
+
+        Repeating this call replaces any prior pending secret and regenerates codes.
+
+
+        Requires: session cookie + X-CSRF-Token.'
+      operationId: post_totp_setup_api_auth_totp_setup_post
+      parameters:
+      - name: X-CSRF-Token
+        in: header
+        required: false
+        schema:
+          anyOf:
+          - type: string
+          - type: 'null'
+          title: X-Csrf-Token
+      responses:
+        '200':
+          description: Successful Response
+          content:
+            application/json:
+              schema:
+                $ref: '#/components/schemas/TotpSetupResponse'
+        '422':
+          description: Validation Error
+          content:
+            application/json:
+              schema:
+                $ref: '#/components/schemas/HTTPValidationError'
+  /api/auth/totp/enable:
+    post:
+      tags:
+      - api-session
+      summary: Post Totp Enable
+      description: 'Enable TOTP by confirming with the current 6-digit code from the
+        authenticator app.
+
+
+        Requires a prior call to POST /api/auth/totp/setup (so that a pending secret
+
+        exists).  On success, totp_enabled becomes True.
+
+
+        Returns 400 if the code is wrong or there is no pending secret.
+
+        Requires: session cookie + X-CSRF-Token.'
+      operationId: post_totp_enable_api_auth_totp_enable_post
+      parameters:
+      - name: X-CSRF-Token
+        in: header
+        required: false
+        schema:
+          anyOf:
+          - type: string
+          - type: 'null'
+          title: X-Csrf-Token
+      requestBody:
+        required: true
+        content:
+          application/json:
+            schema:
+              $ref: '#/components/schemas/TotpEnableRequest'
+      responses:
+        '204':
+          description: Successful Response
+        '422':
+          description: Validation Error
+          content:
+            application/json:
+              schema:
+                $ref: '#/components/schemas/HTTPValidationError'
+  /api/auth/totp/disable:
+    post:
+      tags:
+      - api-session
+      summary: Post Totp Disable
+      description: 'Disable TOTP.  The caller must provide exactly one of:
+
+        - ``password``: the user''s current login password, OR
+
+        - ``code``: the current 6-digit TOTP code.
+
+
+        On success: totp_enabled=False, totp_secret cleared, all recovery codes deleted.
+
+        Returns 400 if neither credential matches or neither is provided.
+
+        Requires: session cookie + X-CSRF-Token.'
+      operationId: post_totp_disable_api_auth_totp_disable_post
+      parameters:
+      - name: X-CSRF-Token
+        in: header
+        required: false
+        schema:
+          anyOf:
+          - type: string
+          - type: 'null'
+          title: X-Csrf-Token
+      requestBody:
+        required: true
+        content:
+          application/json:
+            schema:
+              $ref: '#/components/schemas/TotpDisableRequest'
+      responses:
+        '204':
+          description: Successful Response
+        '422':
+          description: Validation Error
+          content:
+            application/json:
+              schema:
+                $ref: '#/components/schemas/HTTPValidationError'
+  /api/auth/totp:
+    get:
+      tags:
+      - api-session
+      summary: Get Totp Status
+      description: 'Return the current TOTP status for the authenticated user.
+
+
+        Response contains only ``{"enabled": bool}``.
+
+        Secret and recovery codes are NEVER returned here.
+
+        Requires: session cookie only (no CSRF — read-only).'
+      operationId: get_totp_status_api_auth_totp_get
+      responses:
+        '200':
+          description: Successful Response
+          content:
+            application/json:
+              schema:
+                $ref: '#/components/schemas/TotpStatusResponse'
   /homeassistant/publish:
     post:
       tags:
@@ -824,6 +977,11 @@ components:
         password:
           type: string
           title: Password
+        totp_code:
+          anyOf:
+          - type: string
+          - type: 'null'
+          title: Totp Code
       type: object
       required:
       - username
@@ -1077,6 +1235,71 @@ components:
       required:
       - status
       title: StatusResponse
+    TotpDisableRequest:
+      properties:
+        password:
+          anyOf:
+          - type: string
+          - type: 'null'
+          title: Password
+        code:
+          anyOf:
+          - type: string
+          - type: 'null'
+          title: Code
+      type: object
+      title: TotpDisableRequest
+      description: 'Disable TOTP by proving identity.
+
+
+        Exactly one of ``password`` or ``code`` must be provided.'
+    TotpEnableRequest:
+      properties:
+        code:
+          type: string
+          title: Code
+      type: object
+      required:
+      - code
+      title: TotpEnableRequest
+      description: The user confirms setup by providing the 6-digit TOTP code.
+    TotpSetupResponse:
+      properties:
+        secret:
+          type: string
+          title: Secret
+        otpauth_uri:
+          type: string
+          title: Otpauth Uri
+        recovery_codes:
+          items:
+            type: string
+          type: array
+          title: Recovery Codes
+      type: object
+      required:
+      - secret
+      - otpauth_uri
+      - recovery_codes
+      title: TotpSetupResponse
+      description: 'Returned once after a setup call.
+
+
+        ``secret`` and ``recovery_codes`` are **one-time plaintext values**.
+
+        They are never returned again by any subsequent API call.
+
+        The frontend must display and instruct the user to save them before confirming.'
+    TotpStatusResponse:
+      properties:
+        enabled:
+          type: boolean
+          title: Enabled
+      type: object
+      required:
+      - enabled
+      title: TotpStatusResponse
+      description: Minimal status response — never exposes secret or recovery codes.
     ValidationError:
       properties:
         loc:

requirements.in

@@ -4,6 +4,7 @@ argon2-cffi>=25.1,<26.0
 fastapi>=0.115,<0.116
 httpx>=0.28,<1.0
 pydantic-settings>=2.6,<3.0
+pyotp>=2.9,<3.0
 python-multipart>=0.0.12,<1.0
 pyyaml>=6.0,<7.0
 sqlalchemy>=2.0,<3.0

requirements.txt

@@ -59,6 +59,8 @@ pydantic-core==2.46.2
     # via pydantic
 pydantic-settings==2.13.1
     # via -r requirements.in
+pyotp==2.10.0
+    # via -r requirements.in
 python-dotenv==1.2.2
     # via
     #   pydantic-settings

scripts/admin_cli.py

@@ -0,0 +1,387 @@
+"""Admin CLI — escape hatch for local server administration.
+
+Entry point::
+
+    python -m scripts.admin_cli <command> [args]
+
+Commands
+--------
+reset-password <username> [--password <pwd>]
+    Reset the Argon2 password hash for the given admin user.
+    If ``--password`` is omitted the new password is read interactively
+    with ``getpass.getpass`` (no echo).  Prompts twice for confirmation.
+
+unlock [--all | --ip <ip> | --username <u>]
+    Delete rows from ``auth_login_throttle``.
+    --all       clear every row (all IPs and usernames)
+    --ip        clear the row for a specific IP address (scope='ip')
+    --username  clear the row for a specific username (scope='user')
+    Prints the number of rows deleted.
+
+disable-totp <username>
+    Disable TOTP for the given user without requiring any existing credential
+    (recovery codes, current TOTP code, or password).  Sets totp_enabled=False,
+    clears totp_secret, and deletes ALL recovery codes for that user.
+    This is the final escape hatch when all recovery codes are lost.
+
+reissue-totp <username>
+    Generate a fresh TOTP secret for the given user and print the new
+    otpauth:// URI so the user can re-enroll in their Authenticator app.
+    totp_enabled is LEFT UNCHANGED (caller must re-run enable flow if needed).
+    Existing recovery codes are NOT deleted.  Recovery codes are independent
+    random values with no cryptographic binding to the TOTP secret, so they
+    remain valid after reissue and continue to serve as backup login credentials.
+    Silently removing them would unexpectedly strip the user of that backup;
+    use disable-totp for a full cleanup (secret + recovery codes).
+
+list-admin
+    Print all users with their auth status columns (id, username,
+    is_active, force_password_change).
+
+Design contract
+---------------
+- Direct DB access via ``get_session_local()`` — no HTTP, no network.
+- Only touches ``auth_users`` (totp_secret/totp_enabled/password_hash columns)
+  and ``auth_login_throttle`` / ``auth_recovery_code`` (delete rows).
+  Never touches any user-data table (location, poo, public_ip, app_config, etc.).
+- Non-zero exit code on user-facing errors (missing user, bad args, …).
+"""
+
+from __future__ import annotations
+
+import argparse
+import getpass
+import sys
+from pathlib import Path
+
+# Ensure project root is on sys.path when invoked as `python -m scripts.admin_cli`
+# or directly as a script.
+PROJECT_ROOT = Path(__file__).resolve().parents[1]
+if str(PROJECT_ROOT) not in sys.path:
+    sys.path.insert(0, str(PROJECT_ROOT))
+
+import pyotp
+from sqlalchemy import delete, select
+from sqlalchemy.orm import Session
+
+from app.config import get_settings
+from app.db import get_session_local
+from app.models.auth import AuthUser, RecoveryCode
+from app.models.auth_throttle import LoginThrottle
+from app.services.auth import hash_password, verify_password  # noqa: F401 (used below)
+
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+
+def _get_session() -> Session:
+    """Return a new SQLAlchemy session for the configured app database."""
+    SessionLocal = get_session_local()
+    return SessionLocal()
+
+
+def _find_user(session: Session, username: str) -> AuthUser | None:
+    return session.scalar(select(AuthUser).where(AuthUser.username == username).limit(1))
+
+
+def _die(message: str, code: int = 1) -> None:
+    print(f"Error: {message}", file=sys.stderr)
+    sys.exit(code)
+
+
+# ---------------------------------------------------------------------------
+# Sub-command implementations
+# ---------------------------------------------------------------------------
+
+
+def cmd_reset_password(args: argparse.Namespace) -> None:
+    """Reset the password for <username>."""
+    username: str = args.username
+
+    if args.password is not None:
+        new_password: str = args.password
+    else:
+        # Interactive prompt — getpass suppresses echo.
+        new_password = getpass.getpass(f"New password for '{username}': ")
+        confirm = getpass.getpass("Confirm new password: ")
+        if new_password != confirm:
+            _die("Passwords do not match.")
+
+    if not new_password:
+        _die("Password must not be empty.")
+
+    session = _get_session()
+    try:
+        user = _find_user(session, username)
+        if user is None:
+            _die(f"User '{username}' not found.")
+
+        user.password_hash = hash_password(new_password)
+        session.commit()
+        print(f"Password for '{username}' has been reset.")
+    finally:
+        session.close()
+
+
+def cmd_unlock(args: argparse.Namespace) -> None:
+    """Delete rows from auth_login_throttle according to the given filter."""
+    use_all: bool = args.all
+    ip: str | None = args.ip
+    username: str | None = args.username
+
+    # Validate: exactly one of --all / --ip / --username must be provided.
+    flags = [use_all, ip is not None, username is not None]
+    if sum(flags) == 0:
+        _die("One of --all, --ip, or --username is required.")
+    if sum(flags) > 1:
+        _die("Only one of --all, --ip, or --username may be specified at a time.")
+
+    session = _get_session()
+    try:
+        if use_all:
+            stmt = delete(LoginThrottle)
+            result = session.execute(stmt)
+            count = result.rowcount
+        elif ip is not None:
+            stmt = delete(LoginThrottle).where(
+                LoginThrottle.scope == "ip",
+                LoginThrottle.key == ip,
+            )
+            result = session.execute(stmt)
+            count = result.rowcount
+        else:  # username is not None
+            stmt = delete(LoginThrottle).where(
+                LoginThrottle.scope == "user",
+                LoginThrottle.key == username,
+            )
+            result = session.execute(stmt)
+            count = result.rowcount
+
+        session.commit()
+        noun = "row" if count == 1 else "rows"
+        print(f"Cleared {count} throttle {noun}.")
+    finally:
+        session.close()
+
+
+def cmd_disable_totp(args: argparse.Namespace) -> None:
+    """Disable TOTP for <username> — no credential required (escape hatch).
+
+    Sets totp_enabled=False, clears totp_secret, and deletes ALL recovery
+    codes for this user.  This is the final escape route when all recovery
+    codes have been lost: whoever holds server CLI access can always turn off
+    TOTP, restoring plain-password login.
+    """
+    username: str = args.username
+
+    session = _get_session()
+    try:
+        user = _find_user(session, username)
+        if user is None:
+            _die(f"User '{username}' not found.")
+
+        assert user.id is not None
+        # Delete all recovery codes for this user.
+        deleted = session.execute(
+            delete(RecoveryCode).where(RecoveryCode.user_id == user.id)
+        )
+        rc_count = deleted.rowcount
+
+        # Clear TOTP state.
+        user.totp_enabled = False
+        user.totp_secret = None
+        session.commit()
+
+        print(f"TOTP disabled for '{username}'. {rc_count} recovery code(s) deleted.")
+        print("The user can now log in with password only.")
+    finally:
+        session.close()
+
+
+def cmd_reissue_totp(args: argparse.Namespace) -> None:
+    """Generate a fresh TOTP secret for <username> and print the otpauth URI.
+
+    Design choices:
+    - ``totp_enabled`` is LEFT UNCHANGED.  If the user currently has TOTP
+      enabled, the new secret takes effect immediately on the next login
+      (old code invalid, new code accepted) — no web enable step required.
+      If TOTP was disabled, the secret is updated but login remains
+      password-only until the web setup/enable flow is completed.
+    - Existing recovery codes are NOT deleted here.  Recovery codes are
+      independent random values stored as Argon2 hashes in
+      ``auth_recovery_code.code_hash``; login verification compares only those
+      hashes and has no dependency on ``totp_secret``.  Reissuing the secret
+      therefore leaves all existing recovery codes fully valid — they remain
+      the user's backup login path.  Silently deleting them would
+      unexpectedly strip that backup; ``disable-totp`` is the explicit full
+      cleanup path (secret + all recovery codes).
+    - This command's purpose is to hand the new secret URI to the admin so
+      they can re-enroll their Authenticator — useful when the device is lost
+      but the account credentials are still intact.
+    """
+    username: str = args.username
+
+    session = _get_session()
+    try:
+        user = _find_user(session, username)
+        if user is None:
+            _die(f"User '{username}' not found.")
+
+        # Generate a new base-32 secret using pyotp (same library as the web TOTP flow).
+        new_secret = pyotp.random_base32()
+
+        # Build the provisioning URI using the same issuer logic as the web service.
+        settings = get_settings()
+        issuer = settings.effective_totp_issuer
+        otpauth_uri = pyotp.TOTP(new_secret).provisioning_uri(
+            name=user.username,
+            issuer_name=issuer,
+        )
+
+        user.totp_secret = new_secret
+        session.commit()
+
+        print(f"New TOTP secret issued for '{username}'.")
+        print(f"Secret : {new_secret}")
+        print(f"URI    : {otpauth_uri}")
+        print()
+        if user.totp_enabled:
+            print(
+                "NOTE: TOTP is still ENABLED.  Scan the URI above in your Authenticator app;"
+                " the new code is active immediately on next login — the old code is now invalid."
+                "  To fully disable TOTP instead, use 'disable-totp'."
+            )
+        else:
+            print(
+                "NOTE: TOTP is currently DISABLED for this user — login remains password-only."
+                "  The new secret has been saved, but will only take effect once TOTP is"
+                " enabled via the web settings page."
+            )
+    finally:
+        session.close()
+
+
+def cmd_list_admin(args: argparse.Namespace) -> None:  # noqa: ARG001
+    """List all admin users with their status columns."""
+    session = _get_session()
+    try:
+        users = session.scalars(select(AuthUser).order_by(AuthUser.id)).all()
+        if not users:
+            print("No users found.")
+            return
+
+        # Header
+        print(f"{'ID':<5} {'USERNAME':<30} {'IS_ACTIVE':<10} {'FORCE_PW_CHANGE':<16}")
+        print("-" * 65)
+        for u in users:
+            print(
+                f"{u.id:<5} {u.username:<30} {str(u.is_active):<10} "
+                f"{str(u.force_password_change):<16}"
+            )
+    finally:
+        session.close()
+
+
+# ---------------------------------------------------------------------------
+# Argument parser
+# ---------------------------------------------------------------------------
+
+
+def build_parser() -> argparse.ArgumentParser:
+    parser = argparse.ArgumentParser(
+        prog="python -m scripts.admin_cli",
+        description=(
+            "Admin CLI — local escape hatch for password reset, throttle unlock, "
+            "and user inspection.  Connects directly to the configured app DB; "
+            "no HTTP server required."
+        ),
+    )
+    subparsers = parser.add_subparsers(dest="command", metavar="<command>")
+    subparsers.required = True
+
+    # -- reset-password --
+    rp = subparsers.add_parser(
+        "reset-password",
+        help="Reset a user's password (prompts interactively if --password is omitted).",
+    )
+    rp.add_argument("username", help="Username whose password to reset.")
+    rp.add_argument(
+        "--password",
+        default=None,
+        help="New password (omit to be prompted interactively with no echo).",
+    )
+    rp.set_defaults(func=cmd_reset_password)
+
+    # -- unlock --
+    ul = subparsers.add_parser(
+        "unlock",
+        help="Clear login-throttle rows (escape when locked out by exponential back-off).",
+    )
+    unlock_group = ul.add_mutually_exclusive_group(required=True)
+    unlock_group.add_argument(
+        "--all",
+        action="store_true",
+        default=False,
+        help="Clear ALL throttle rows.",
+    )
+    unlock_group.add_argument(
+        "--ip",
+        metavar="<ip>",
+        default=None,
+        help="Clear the throttle row for this specific IP address.",
+    )
+    unlock_group.add_argument(
+        "--username",
+        metavar="<username>",
+        default=None,
+        help="Clear the throttle row for this specific username.",
+    )
+    ul.set_defaults(func=cmd_unlock)
+
+    # -- disable-totp --
+    dt = subparsers.add_parser(
+        "disable-totp",
+        help=(
+            "Disable TOTP for a user (escape hatch — no credential required). "
+            "Clears totp_secret, sets totp_enabled=False, deletes all recovery codes."
+        ),
+    )
+    dt.add_argument("username", help="Username whose TOTP to disable.")
+    dt.set_defaults(func=cmd_disable_totp)
+
+    # -- reissue-totp --
+    rt = subparsers.add_parser(
+        "reissue-totp",
+        help=(
+            "Generate a new TOTP secret for a user and print the otpauth:// URI. "
+            "totp_enabled is left unchanged; re-enroll via the web UI."
+        ),
+    )
+    rt.add_argument("username", help="Username for whom to reissue the TOTP secret.")
+    rt.set_defaults(func=cmd_reissue_totp)
+
+    # -- list-admin --
+    la = subparsers.add_parser(
+        "list-admin",
+        help="List all users and their auth status (for diagnostics).",
+    )
+    la.set_defaults(func=cmd_list_admin)
+
+    return parser
+
+
+# ---------------------------------------------------------------------------
+# Entry point
+# ---------------------------------------------------------------------------
+
+
+def main(argv: list[str] | None = None) -> None:
+    parser = build_parser()
+    args = parser.parse_args(argv)
+    args.func(args)
+
+
+if __name__ == "__main__":
+    main()

scripts/app_db_adopt.py

@@ -15,7 +15,7 @@ if str(PROJECT_ROOT) not in sys.path:
 
 from app.config import get_settings
 
-APP_BASELINE_REVISION = "20260611_06_merge_location_poo_tables"
+APP_BASELINE_REVISION = "20260621_08_totp"
 
 
 class AppDatabaseAdoptionError(RuntimeError):

tests/conftest.py

@@ -26,6 +26,9 @@ def test_database_urls(tmp_path: Path, monkeypatch: pytest.MonkeyPatch):
     monkeypatch.setenv("AUTH_BOOTSTRAP_USERNAME", "admin")
     monkeypatch.setenv("AUTH_BOOTSTRAP_PASSWORD", "test-password")
     monkeypatch.setenv("AUTH_COOKIE_SECURE_OVERRIDE", "false")
+    # Disable throttle by default so existing tests are unaffected; throttle
+    # tests opt in explicitly via their own monkeypatch / env override.
+    monkeypatch.setenv("AUTH_LOGIN_THROTTLE_ENABLED", "false")
     get_settings.cache_clear()
     reset_db_caches()
 

tests/test_admin_cli.py

@@ -0,0 +1,591 @@
+"""Tests for M4-T03 + M4-T07: scripts/admin_cli.py — CLI escape hatch.
+
+Covers:
+- reset-password: new password verifies, old password fails.
+- unlock: clears specified/all throttle rows; prints count; idempotent.
+- disable-totp: clears TOTP state + recovery codes without requiring any credential.
+- reissue-totp: writes new secret; prints otpauth URI; secret changes.
+- list-admin: prints user table without crashing.
+- Non-existent user → friendly error + SystemExit with non-zero code.
+- Data-safety: CLI must never reference user-data tables in delete/drop.
+
+DB isolation: each test gets a fresh temporary SQLite database via
+``cli_db`` fixture (mirrors the pattern in test_login_throttle.py).
+"""
+
+from __future__ import annotations
+
+import subprocess
+import sys
+from datetime import UTC, datetime
+from pathlib import Path
+
+import pytest
+from alembic import command
+from alembic.config import Config
+from sqlalchemy import create_engine, select
+from sqlalchemy.orm import Session, sessionmaker
+
+from app.db import reset_db_caches
+from app.config import get_settings
+from app.models.auth import AuthUser, RecoveryCode
+from app.models.auth_throttle import LoginThrottle
+from app.services.auth import hash_password, verify_password
+from scripts.admin_cli import (
+    build_parser,
+    cmd_disable_totp,
+    cmd_list_admin,
+    cmd_reset_password,
+    cmd_reissue_totp,
+    cmd_unlock,
+)
+
+
+# ---------------------------------------------------------------------------
+# Fixtures
+# ---------------------------------------------------------------------------
+
+
+def _make_app_alembic_config(database_url: str) -> Config:
+    config = Config("alembic_app.ini")
+    config.set_main_option("sqlalchemy.url", database_url)
+    return config
+
+
+@pytest.fixture()
+def cli_db(tmp_path: Path, monkeypatch: pytest.MonkeyPatch):
+    """Fresh migrated SQLite DB wired as the active app DB for the CLI.
+
+    Monkeypatches APP_DATABASE_URL so that ``get_session_local()`` inside
+    admin_cli reaches this temp file, not the real data store.
+    """
+    db_path = tmp_path / "cli_test.db"
+    db_url = f"sqlite:///{db_path}"
+
+    monkeypatch.setenv("APP_DATABASE_URL", db_url)
+    monkeypatch.setenv("AUTH_BOOTSTRAP_USERNAME", "admin")
+    monkeypatch.setenv("AUTH_BOOTSTRAP_PASSWORD", "original-password")
+    monkeypatch.setenv("AUTH_COOKIE_SECURE_OVERRIDE", "false")
+    monkeypatch.setenv("AUTH_LOGIN_THROTTLE_ENABLED", "false")
+    get_settings.cache_clear()
+    reset_db_caches()
+
+    command.upgrade(_make_app_alembic_config(db_url), "head")
+    reset_db_caches()
+
+    # Bootstrap the admin user manually so the test controls the initial hash.
+    engine = create_engine(db_url, connect_args={"check_same_thread": False})
+    SessionLocal = sessionmaker(bind=engine, autoflush=False, autocommit=False, class_=Session)
+    with SessionLocal() as session:
+        # The Alembic upgrade does NOT create a user (that is app startup logic).
+        # Insert the initial admin user here for tests that need one.
+        existing = session.scalar(select(AuthUser).limit(1))
+        if existing is None:
+            user = AuthUser(
+                username="admin",
+                password_hash=hash_password("original-password"),
+                is_active=True,
+                force_password_change=False,
+                created_at=datetime.now(UTC),
+            )
+            session.add(user)
+            session.commit()
+    engine.dispose()
+    reset_db_caches()
+
+    yield db_url
+
+    get_settings.cache_clear()
+    reset_db_caches()
+
+
+@pytest.fixture()
+def cli_session(cli_db: str) -> Session:
+    """SQLAlchemy session bound to the CLI test DB."""
+    engine = create_engine(cli_db, connect_args={"check_same_thread": False})
+    SessionLocal = sessionmaker(bind=engine, autoflush=False, autocommit=False, class_=Session)
+    session = SessionLocal()
+    try:
+        yield session
+    finally:
+        session.close()
+        engine.dispose()
+
+
+def _add_throttle_rows(session: Session, rows: list[tuple[str, str]]) -> None:
+    """Helper: insert (scope, key) throttle rows into the test DB."""
+    now = datetime.now(UTC)
+    for scope, key in rows:
+        session.add(
+            LoginThrottle(
+                scope=scope,
+                key=key,
+                failures=5,
+                first_failed_at=now,
+                last_failed_at=now,
+                next_allowed_at=None,
+            )
+        )
+    session.commit()
+
+
+def _throttle_count(session: Session) -> int:
+    return session.query(LoginThrottle).count()
+
+
+def _enable_totp_for_user(session: Session, username: str, secret: str = "JBSWY3DPEHPK3PXP") -> None:
+    """Helper: set TOTP enabled + secret for a user and add some recovery codes."""
+    user = session.scalar(select(AuthUser).where(AuthUser.username == username))
+    assert user is not None, f"User '{username}' not found in _enable_totp_for_user"
+    user.totp_secret = secret
+    user.totp_enabled = True
+    # Add two fake recovery codes so we can verify they get deleted.
+    session.add(RecoveryCode(user_id=user.id, code_hash=hash_password("aaaa-bbbb")))
+    session.add(RecoveryCode(user_id=user.id, code_hash=hash_password("cccc-dddd")))
+    session.commit()
+    session.expire_all()
+
+
+def _recovery_code_count(session: Session, username: str) -> int:
+    user = session.scalar(select(AuthUser).where(AuthUser.username == username))
+    assert user is not None
+    return session.query(RecoveryCode).filter(RecoveryCode.user_id == user.id).count()
+
+
+# ---------------------------------------------------------------------------
+# reset-password tests
+# ---------------------------------------------------------------------------
+
+
+class TestResetPassword:
+    """reset-password sub-command."""
+
+    def test_new_password_verifies(self, cli_db: str, cli_session: Session):
+        """After reset, new password passes verify_password; old one fails."""
+        parser = build_parser()
+        args = parser.parse_args(["reset-password", "admin", "--password", "new-secret-123"])
+        cmd_reset_password(args)
+
+        # Re-query from the session (committed by CLI, so flush needed).
+        cli_session.expire_all()
+        user = cli_session.scalar(select(AuthUser).where(AuthUser.username == "admin"))
+        assert user is not None
+        assert verify_password("new-secret-123", user.password_hash), (
+            "New password should verify against stored hash"
+        )
+        assert not verify_password("original-password", user.password_hash), (
+            "Old password must no longer verify"
+        )
+
+    def test_nonexistent_user_exits_nonzero(self, cli_db: str):
+        """reset-password for an unknown username must SystemExit with code != 0."""
+        parser = build_parser()
+        args = parser.parse_args(
+            ["reset-password", "ghost-user", "--password", "doesnt-matter"]
+        )
+        with pytest.raises(SystemExit) as exc_info:
+            cmd_reset_password(args)
+        assert exc_info.value.code != 0
+
+    def test_empty_password_exits_nonzero(self, cli_db: str):
+        """Passing an empty --password string must fail with non-zero exit."""
+        parser = build_parser()
+        args = parser.parse_args(["reset-password", "admin", "--password", ""])
+        with pytest.raises(SystemExit) as exc_info:
+            cmd_reset_password(args)
+        assert exc_info.value.code != 0
+
+    def test_multiple_resets_are_idempotent(self, cli_db: str, cli_session: Session):
+        """Calling reset-password multiple times leaves the last password active."""
+        parser = build_parser()
+
+        args1 = parser.parse_args(["reset-password", "admin", "--password", "first-pass"])
+        cmd_reset_password(args1)
+
+        args2 = parser.parse_args(["reset-password", "admin", "--password", "second-pass"])
+        cmd_reset_password(args2)
+
+        cli_session.expire_all()
+        user = cli_session.scalar(select(AuthUser).where(AuthUser.username == "admin"))
+        assert user is not None
+        assert verify_password("second-pass", user.password_hash)
+        assert not verify_password("first-pass", user.password_hash)
+
+
+# ---------------------------------------------------------------------------
+# unlock tests
+# ---------------------------------------------------------------------------
+
+
+class TestUnlock:
+    """unlock sub-command."""
+
+    def test_unlock_all_clears_all_rows(self, cli_db: str, cli_session: Session):
+        """unlock --all should delete every row in auth_login_throttle."""
+        _add_throttle_rows(
+            cli_session,
+            [("ip", "1.2.3.4"), ("ip", "5.6.7.8"), ("user", "admin")],
+        )
+        assert _throttle_count(cli_session) == 3
+
+        parser = build_parser()
+        args = parser.parse_args(["unlock", "--all"])
+        cmd_unlock(args)
+
+        cli_session.expire_all()
+        assert _throttle_count(cli_session) == 0
+
+    def test_unlock_ip_clears_only_target_ip(self, cli_db: str, cli_session: Session):
+        """unlock --ip should remove only the matching scope='ip' row."""
+        _add_throttle_rows(
+            cli_session,
+            [("ip", "10.0.0.1"), ("ip", "10.0.0.2"), ("user", "admin")],
+        )
+
+        parser = build_parser()
+        args = parser.parse_args(["unlock", "--ip", "10.0.0.1"])
+        cmd_unlock(args)
+
+        cli_session.expire_all()
+        assert _throttle_count(cli_session) == 2
+        remaining = cli_session.scalars(select(LoginThrottle)).all()
+        keys = [(r.scope, r.key) for r in remaining]
+        assert ("ip", "10.0.0.1") not in keys
+        assert ("ip", "10.0.0.2") in keys
+        assert ("user", "admin") in keys
+
+    def test_unlock_username_clears_only_target_user(self, cli_db: str, cli_session: Session):
+        """unlock --username should remove only the matching scope='user' row."""
+        _add_throttle_rows(
+            cli_session,
+            [("ip", "1.2.3.4"), ("user", "admin"), ("user", "other")],
+        )
+
+        parser = build_parser()
+        args = parser.parse_args(["unlock", "--username", "admin"])
+        cmd_unlock(args)
+
+        cli_session.expire_all()
+        assert _throttle_count(cli_session) == 2
+        remaining = cli_session.scalars(select(LoginThrottle)).all()
+        keys = [(r.scope, r.key) for r in remaining]
+        assert ("user", "admin") not in keys
+        assert ("ip", "1.2.3.4") in keys
+        assert ("user", "other") in keys
+
+    def test_unlock_all_idempotent(self, cli_db: str, cli_session: Session):
+        """Calling unlock --all when no rows exist must not raise."""
+        assert _throttle_count(cli_session) == 0
+
+        parser = build_parser()
+        args = parser.parse_args(["unlock", "--all"])
+        cmd_unlock(args)  # must not raise
+
+        cli_session.expire_all()
+        assert _throttle_count(cli_session) == 0
+
+    def test_unlock_ip_idempotent_when_absent(self, cli_db: str, cli_session: Session):
+        """Calling unlock --ip for a non-existent IP is a no-op (0 rows deleted)."""
+        parser = build_parser()
+        args = parser.parse_args(["unlock", "--ip", "99.99.99.99"])
+        cmd_unlock(args)  # must not raise; 0 rows deleted is fine
+
+    def test_unlock_username_idempotent_when_absent(self, cli_db: str, cli_session: Session):
+        """Calling unlock --username for a non-existent username is a no-op."""
+        parser = build_parser()
+        args = parser.parse_args(["unlock", "--username", "nobody"])
+        cmd_unlock(args)  # must not raise
+
+    def test_unlock_all_prints_row_count(self, cli_db: str, cli_session: Session, capsys):
+        """unlock --all must print the number of rows deleted."""
+        _add_throttle_rows(cli_session, [("ip", "1.2.3.4"), ("user", "admin")])
+
+        parser = build_parser()
+        args = parser.parse_args(["unlock", "--all"])
+        cmd_unlock(args)
+
+        captured = capsys.readouterr()
+        assert "2" in captured.out
+
+    def test_unlock_does_not_touch_auth_users(self, cli_db: str, cli_session: Session):
+        """unlock must not delete or alter auth_users rows."""
+        _add_throttle_rows(cli_session, [("ip", "1.2.3.4")])
+        user_count_before = cli_session.query(AuthUser).count()
+
+        parser = build_parser()
+        args = parser.parse_args(["unlock", "--all"])
+        cmd_unlock(args)
+
+        cli_session.expire_all()
+        assert cli_session.query(AuthUser).count() == user_count_before
+
+
+# ---------------------------------------------------------------------------
+# list-admin tests
+# ---------------------------------------------------------------------------
+
+
+class TestListAdmin:
+    """list-admin sub-command."""
+
+    def test_list_admin_prints_header_and_user(self, cli_db: str, capsys):
+        """list-admin must print the admin user row."""
+        parser = build_parser()
+        args = parser.parse_args(["list-admin"])
+        cmd_list_admin(args)
+
+        captured = capsys.readouterr()
+        assert "admin" in captured.out
+
+    def test_list_admin_no_users(
+        self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch, capsys
+    ):
+        """list-admin on an empty user table should say 'No users found.'"""
+        db_path = tmp_path / "empty_test.db"
+        db_url = f"sqlite:///{db_path}"
+
+        monkeypatch.setenv("APP_DATABASE_URL", db_url)
+        monkeypatch.setenv("AUTH_BOOTSTRAP_USERNAME", "admin")
+        monkeypatch.setenv("AUTH_BOOTSTRAP_PASSWORD", "x")
+        monkeypatch.setenv("AUTH_COOKIE_SECURE_OVERRIDE", "false")
+        get_settings.cache_clear()
+        reset_db_caches()
+        command.upgrade(_make_app_alembic_config(db_url), "head")
+        reset_db_caches()
+
+        # Do NOT bootstrap a user — table is empty.
+        parser = build_parser()
+        args = parser.parse_args(["list-admin"])
+        cmd_list_admin(args)
+
+        captured = capsys.readouterr()
+        assert "No users found" in captured.out
+
+        get_settings.cache_clear()
+        reset_db_caches()
+
+
+# ---------------------------------------------------------------------------
+# disable-totp tests
+# ---------------------------------------------------------------------------
+
+
+class TestDisableTotp:
+    """disable-totp sub-command (M4-T07)."""
+
+    def test_disable_totp_clears_state(self, cli_db: str, cli_session: Session):
+        """After disable-totp: totp_enabled=False, totp_secret=None, recovery codes gone."""
+        _enable_totp_for_user(cli_session, "admin")
+
+        # Verify preconditions.
+        user = cli_session.scalar(select(AuthUser).where(AuthUser.username == "admin"))
+        assert user is not None
+        assert user.totp_enabled is True
+        assert user.totp_secret is not None
+        assert _recovery_code_count(cli_session, "admin") == 2
+
+        # Run disable-totp — NO credential supplied (escape hatch semantics).
+        parser = build_parser()
+        args = parser.parse_args(["disable-totp", "admin"])
+        cmd_disable_totp(args)
+
+        # Verify state.
+        cli_session.expire_all()
+        user = cli_session.scalar(select(AuthUser).where(AuthUser.username == "admin"))
+        assert user is not None
+        assert user.totp_enabled is False
+        assert user.totp_secret is None
+        assert _recovery_code_count(cli_session, "admin") == 0
+
+    def test_disable_totp_no_credential_required(self, cli_db: str, cli_session: Session):
+        """disable-totp must succeed without any recovery code or password argument."""
+        _enable_totp_for_user(cli_session, "admin")
+
+        # The argparse spec for disable-totp takes only username — no credential flags.
+        # Parsing must succeed with just the username.
+        parser = build_parser()
+        args = parser.parse_args(["disable-totp", "admin"])
+        # No exception means no credential was needed.
+        cmd_disable_totp(args)  # must not raise
+
+    def test_disable_totp_already_disabled_is_noop(self, cli_db: str, cli_session: Session):
+        """disable-totp on a user who never had TOTP is a harmless no-op."""
+        user = cli_session.scalar(select(AuthUser).where(AuthUser.username == "admin"))
+        assert user is not None
+        assert user.totp_enabled is False
+
+        parser = build_parser()
+        args = parser.parse_args(["disable-totp", "admin"])
+        cmd_disable_totp(args)  # must not raise
+
+        cli_session.expire_all()
+        user = cli_session.scalar(select(AuthUser).where(AuthUser.username == "admin"))
+        assert user is not None
+        assert user.totp_enabled is False
+        assert user.totp_secret is None
+
+    def test_disable_totp_nonexistent_user_exits_nonzero(self, cli_db: str):
+        """disable-totp for an unknown user must SystemExit with code != 0."""
+        parser = build_parser()
+        args = parser.parse_args(["disable-totp", "ghost-user"])
+        with pytest.raises(SystemExit) as exc_info:
+            cmd_disable_totp(args)
+        assert exc_info.value.code != 0
+
+    def test_disable_totp_prints_confirmation(
+        self, cli_db: str, cli_session: Session, capsys
+    ):
+        """disable-totp must print a confirmation message mentioning the user."""
+        _enable_totp_for_user(cli_session, "admin")
+
+        parser = build_parser()
+        args = parser.parse_args(["disable-totp", "admin"])
+        cmd_disable_totp(args)
+
+        captured = capsys.readouterr()
+        assert "admin" in captured.out
+
+    def test_disable_totp_does_not_touch_password(
+        self, cli_db: str, cli_session: Session
+    ):
+        """disable-totp must not alter the user's password_hash."""
+        _enable_totp_for_user(cli_session, "admin")
+
+        user_before = cli_session.scalar(select(AuthUser).where(AuthUser.username == "admin"))
+        assert user_before is not None
+        pw_hash_before = user_before.password_hash
+
+        parser = build_parser()
+        args = parser.parse_args(["disable-totp", "admin"])
+        cmd_disable_totp(args)
+
+        cli_session.expire_all()
+        user_after = cli_session.scalar(select(AuthUser).where(AuthUser.username == "admin"))
+        assert user_after is not None
+        assert user_after.password_hash == pw_hash_before
+
+
+# ---------------------------------------------------------------------------
+# reissue-totp tests
+# ---------------------------------------------------------------------------
+
+
+class TestReissueTotp:
+    """reissue-totp sub-command (M4-T07)."""
+
+    def test_reissue_totp_changes_secret(self, cli_db: str, cli_session: Session):
+        """After reissue-totp, totp_secret must be a new non-empty value."""
+        old_secret = "JBSWY3DPEHPK3PXP"
+        _enable_totp_for_user(cli_session, "admin", secret=old_secret)
+
+        parser = build_parser()
+        args = parser.parse_args(["reissue-totp", "admin"])
+        cmd_reissue_totp(args)
+
+        cli_session.expire_all()
+        user = cli_session.scalar(select(AuthUser).where(AuthUser.username == "admin"))
+        assert user is not None
+        assert user.totp_secret is not None
+        assert user.totp_secret != old_secret, "Secret must be regenerated"
+
+    def test_reissue_totp_prints_uri(self, cli_db: str, cli_session: Session, capsys):
+        """reissue-totp must print an otpauth:// URI and the secret."""
+        _enable_totp_for_user(cli_session, "admin")
+
+        parser = build_parser()
+        args = parser.parse_args(["reissue-totp", "admin"])
+        cmd_reissue_totp(args)
+
+        captured = capsys.readouterr()
+        assert "otpauth://" in captured.out
+        assert "Secret" in captured.out
+
+    def test_reissue_totp_totp_enabled_unchanged(self, cli_db: str, cli_session: Session):
+        """reissue-totp must leave totp_enabled in its current state (True stays True)."""
+        _enable_totp_for_user(cli_session, "admin")
+
+        user_before = cli_session.scalar(select(AuthUser).where(AuthUser.username == "admin"))
+        assert user_before is not None
+        assert user_before.totp_enabled is True
+
+        parser = build_parser()
+        args = parser.parse_args(["reissue-totp", "admin"])
+        cmd_reissue_totp(args)
+
+        cli_session.expire_all()
+        user_after = cli_session.scalar(select(AuthUser).where(AuthUser.username == "admin"))
+        assert user_after is not None
+        assert user_after.totp_enabled is True, "totp_enabled must remain unchanged"
+
+    def test_reissue_totp_disabled_user_stays_disabled(
+        self, cli_db: str, cli_session: Session
+    ):
+        """reissue-totp on a user with totp_enabled=False leaves it False."""
+        # Don't enable TOTP — user is in disabled state.
+        parser = build_parser()
+        args = parser.parse_args(["reissue-totp", "admin"])
+        cmd_reissue_totp(args)
+
+        cli_session.expire_all()
+        user = cli_session.scalar(select(AuthUser).where(AuthUser.username == "admin"))
+        assert user is not None
+        assert user.totp_enabled is False, "totp_enabled must remain False"
+        assert user.totp_secret is not None, "A new secret should have been written"
+
+    def test_reissue_totp_nonexistent_user_exits_nonzero(self, cli_db: str):
+        """reissue-totp for an unknown user must SystemExit with code != 0."""
+        parser = build_parser()
+        args = parser.parse_args(["reissue-totp", "ghost-user"])
+        with pytest.raises(SystemExit) as exc_info:
+            cmd_reissue_totp(args)
+        assert exc_info.value.code != 0
+
+
+# ---------------------------------------------------------------------------
+# Argument-parser edge-cases
+# ---------------------------------------------------------------------------
+
+
+class TestParserEdgeCases:
+    """Ensure argparse flags behave as expected."""
+
+    def test_unlock_requires_flag(self):
+        """unlock without any flag should exit with a parse error (non-zero)."""
+        parser = build_parser()
+        with pytest.raises(SystemExit) as exc_info:
+            parser.parse_args(["unlock"])
+        assert exc_info.value.code != 0
+
+    def test_unlock_ip_and_all_mutually_exclusive(self):
+        """--ip and --all together should produce a parse error."""
+        parser = build_parser()
+        with pytest.raises(SystemExit) as exc_info:
+            parser.parse_args(["unlock", "--all", "--ip", "1.2.3.4"])
+        assert exc_info.value.code != 0
+
+    def test_unknown_command_exits_nonzero(self):
+        """An unknown sub-command must exit non-zero."""
+        parser = build_parser()
+        with pytest.raises(SystemExit) as exc_info:
+            parser.parse_args(["nonexistent-command"])
+        assert exc_info.value.code != 0
+
+
+# ---------------------------------------------------------------------------
+# Module-level smoke: python -m scripts.admin_cli --help
+# ---------------------------------------------------------------------------
+
+
+def test_help_flag_exits_zero():
+    """python -m scripts.admin_cli --help must exit 0."""
+    result = subprocess.run(
+        [sys.executable, "-m", "scripts.admin_cli", "--help"],
+        capture_output=True,
+        text=True,
+    )
+    assert result.returncode == 0
+    assert "reset-password" in result.stdout
+    assert "unlock" in result.stdout
+    assert "disable-totp" in result.stdout
+    assert "reissue-totp" in result.stdout
+    assert "list-admin" in result.stdout

tests/test_api_session.py

@@ -1,7 +1,24 @@
-"""Tests for M2-T02: GET /api/session, POST /api/auth/login, /logout, /password."""
+"""Tests for M2-T02: GET /api/session, POST /api/auth/login, /logout, /password.
+
+Also covers M4-T06: two-factor login regression tests (TOTP not enabled → same
+as before; TOTP enabled → two-step protocol).
+"""
 from __future__ import annotations
 
+from pathlib import Path
+
+import pyotp
+import pytest
+from alembic import command
+from alembic.config import Config
 from fastapi.testclient import TestClient
+from sqlalchemy import create_engine
+from sqlalchemy.orm import Session
+
+from app.db import reset_db_caches
+from app.config import get_settings
+from app.main import create_app
+from app.services.login_throttle import N_FREE
 
 
 # ---------------------------------------------------------------------------
@@ -330,3 +347,427 @@ def test_session_response_has_no_secret_fields(client: TestClient) -> None:
     assert set(body.keys()) == {"user", "csrf_token"}
     # user must have exactly these keys
     assert set(body["user"].keys()) == {"username", "force_password_change"}
+
+
+# ---------------------------------------------------------------------------
+# M4-T06 — Two-factor login (TOTP second step)
+# Fixtures & helpers
+# ---------------------------------------------------------------------------
+
+
+def _make_app_alembic_config(database_url: str) -> Config:
+    cfg = Config("alembic_app.ini")
+    cfg.set_main_option("sqlalchemy.url", database_url)
+    return cfg
+
+
+@pytest.fixture
+def totp_login_client(tmp_path: Path, monkeypatch: pytest.MonkeyPatch):
+    """Full client with TOTP-capable DB (throttle disabled for simplicity)."""
+    db_path = tmp_path / "totp_login_test.db"
+    db_url = f"sqlite:///{db_path}"
+
+    monkeypatch.setenv("APP_DATABASE_URL", db_url)
+    monkeypatch.setenv("AUTH_BOOTSTRAP_USERNAME", "admin")
+    monkeypatch.setenv("AUTH_BOOTSTRAP_PASSWORD", "test-password")
+    monkeypatch.setenv("AUTH_COOKIE_SECURE_OVERRIDE", "false")
+    monkeypatch.setenv("AUTH_LOGIN_THROTTLE_ENABLED", "false")
+    get_settings.cache_clear()
+    reset_db_caches()
+
+    command.upgrade(_make_app_alembic_config(db_url), "head")
+    reset_db_caches()
+
+    app = create_app()
+    with TestClient(app) as tc:
+        yield tc
+
+    get_settings.cache_clear()
+    reset_db_caches()
+
+
+@pytest.fixture
+def totp_login_throttle_client(tmp_path: Path, monkeypatch: pytest.MonkeyPatch):
+    """Full client with TOTP + throttle ENABLED (for throttle-interaction tests)."""
+    db_path = tmp_path / "totp_throttle_test.db"
+    db_url = f"sqlite:///{db_path}"
+
+    monkeypatch.setenv("APP_DATABASE_URL", db_url)
+    monkeypatch.setenv("AUTH_BOOTSTRAP_USERNAME", "admin")
+    monkeypatch.setenv("AUTH_BOOTSTRAP_PASSWORD", "test-password")
+    monkeypatch.setenv("AUTH_COOKIE_SECURE_OVERRIDE", "false")
+    monkeypatch.setenv("AUTH_LOGIN_THROTTLE_ENABLED", "true")
+    get_settings.cache_clear()
+    reset_db_caches()
+
+    command.upgrade(_make_app_alembic_config(db_url), "head")
+    reset_db_caches()
+
+    engine = create_engine(db_url, connect_args={"check_same_thread": False})
+    app = create_app()
+    with TestClient(app) as tc:
+        yield tc, engine
+    engine.dispose()
+    get_settings.cache_clear()
+    reset_db_caches()
+
+
+def _enable_totp(client: TestClient) -> tuple[str, list[str]]:
+    """Log in, setup TOTP, enable it; return (secret, recovery_codes)."""
+    login_resp = client.post(
+        "/api/auth/login",
+        json={"username": "admin", "password": "test-password"},
+    )
+    assert login_resp.status_code == 200, login_resp.text
+    csrf = login_resp.json()["csrf_token"]
+
+    setup_resp = client.post("/api/auth/totp/setup", headers={"X-CSRF-Token": csrf})
+    assert setup_resp.status_code == 200, setup_resp.text
+    setup_body = setup_resp.json()
+    secret = setup_body["secret"]
+    recovery_codes = setup_body["recovery_codes"]
+
+    enable_resp = client.post(
+        "/api/auth/totp/enable",
+        json={"code": pyotp.TOTP(secret).now()},
+        headers={"X-CSRF-Token": csrf},
+    )
+    assert enable_resp.status_code == 204, enable_resp.text
+
+    # Log out so subsequent login tests start fresh
+    client.post("/api/auth/logout", headers={"X-CSRF-Token": csrf})
+    return secret, recovery_codes
+
+
+# ---------------------------------------------------------------------------
+# M4-T06 — Regression: TOTP not enabled → login unchanged
+# ---------------------------------------------------------------------------
+
+
+class TestLoginWithoutTotpEnabled:
+    """TOTP not enabled: login behaves exactly as before T06."""
+
+    def test_login_without_totp_code_succeeds(self, totp_login_client: TestClient) -> None:
+        """Standard login (no totp_code field at all) still works when TOTP is off."""
+        resp = totp_login_client.post(
+            "/api/auth/login",
+            json={"username": "admin", "password": "test-password"},
+        )
+        assert resp.status_code == 200
+        body = resp.json()
+        assert "user" in body
+        assert "csrf_token" in body
+
+    def test_login_with_null_totp_code_succeeds(self, totp_login_client: TestClient) -> None:
+        """Explicitly sending totp_code=null works fine when TOTP is off."""
+        resp = totp_login_client.post(
+            "/api/auth/login",
+            json={"username": "admin", "password": "test-password", "totp_code": None},
+        )
+        assert resp.status_code == 200
+
+    def test_login_sets_session_cookie(self, totp_login_client: TestClient) -> None:
+        resp = totp_login_client.post(
+            "/api/auth/login",
+            json={"username": "admin", "password": "test-password"},
+        )
+        assert resp.status_code == 200
+        assert "home_automation_session=" in resp.headers.get("set-cookie", "").lower()
+
+    def test_wrong_password_still_401(self, totp_login_client: TestClient) -> None:
+        resp = totp_login_client.post(
+            "/api/auth/login",
+            json={"username": "admin", "password": "wrong"},
+        )
+        assert resp.status_code == 401
+        # Must be the standard generic message
+        assert resp.json()["detail"] == "invalid username or password"
+
+    def test_wrong_password_no_cookie(self, totp_login_client: TestClient) -> None:
+        resp = totp_login_client.post(
+            "/api/auth/login",
+            json={"username": "admin", "password": "wrong"},
+        )
+        assert resp.status_code == 401
+        assert "home_automation_session=" not in resp.headers.get("set-cookie", "").lower()
+
+
+# ---------------------------------------------------------------------------
+# M4-T06 — TOTP enabled: two-step protocol
+# ---------------------------------------------------------------------------
+
+
+class TestLoginWithTotpEnabled:
+    """TOTP enabled: missing/wrong/correct code semantics."""
+
+    def test_missing_totp_code_returns_401_with_totp_required(
+        self, totp_login_client: TestClient
+    ) -> None:
+        """Correct password but no totp_code → 401 with totp_required signal."""
+        _enable_totp(totp_login_client)
+        resp = totp_login_client.post(
+            "/api/auth/login",
+            json={"username": "admin", "password": "test-password"},
+        )
+        assert resp.status_code == 401
+        body = resp.json()
+        # detail must be a dict with totp_required=true
+        assert isinstance(body["detail"], dict)
+        assert body["detail"].get("totp_required") is True
+
+    def test_totp_required_response_has_no_session_cookie(
+        self, totp_login_client: TestClient
+    ) -> None:
+        """totp_required 401 must NOT set a session cookie."""
+        _enable_totp(totp_login_client)
+        resp = totp_login_client.post(
+            "/api/auth/login",
+            json={"username": "admin", "password": "test-password"},
+        )
+        assert resp.status_code == 401
+        assert "home_automation_session=" not in resp.headers.get("set-cookie", "").lower()
+
+    def test_wrong_totp_code_returns_generic_401(
+        self, totp_login_client: TestClient
+    ) -> None:
+        """Wrong TOTP code → generic 401 (no totp_required signal)."""
+        _enable_totp(totp_login_client)
+        resp = totp_login_client.post(
+            "/api/auth/login",
+            json={"username": "admin", "password": "test-password", "totp_code": "000000"},
+        )
+        assert resp.status_code == 401
+        detail = resp.json()["detail"]
+        # Must be a plain string (not dict with totp_required)
+        assert isinstance(detail, str)
+        assert detail == "invalid username or password"
+
+    def test_wrong_totp_code_no_cookie(self, totp_login_client: TestClient) -> None:
+        """Wrong TOTP → no cookie."""
+        _enable_totp(totp_login_client)
+        resp = totp_login_client.post(
+            "/api/auth/login",
+            json={"username": "admin", "password": "test-password", "totp_code": "000000"},
+        )
+        assert resp.status_code == 401
+        assert "home_automation_session=" not in resp.headers.get("set-cookie", "").lower()
+
+    def test_correct_totp_code_issues_session(self, totp_login_client: TestClient) -> None:
+        """Correct TOTP code → 200 + cookie + csrf."""
+        secret, _ = _enable_totp(totp_login_client)
+        resp = totp_login_client.post(
+            "/api/auth/login",
+            json={
+                "username": "admin",
+                "password": "test-password",
+                "totp_code": pyotp.TOTP(secret).now(),
+            },
+        )
+        assert resp.status_code == 200
+        body = resp.json()
+        assert "user" in body
+        assert "csrf_token" in body
+        assert "home_automation_session=" in resp.headers.get("set-cookie", "").lower()
+
+    def test_totp_required_detail_not_on_wrong_password(
+        self, totp_login_client: TestClient
+    ) -> None:
+        """Wrong password must NOT carry totp_required (cannot reveal TOTP state)."""
+        _enable_totp(totp_login_client)
+        resp = totp_login_client.post(
+            "/api/auth/login",
+            json={"username": "admin", "password": "wrong-password"},
+        )
+        assert resp.status_code == 401
+        detail = resp.json()["detail"]
+        assert isinstance(detail, str), "Wrong-password 401 must not expose totp_required"
+        if isinstance(detail, dict):
+            assert "totp_required" not in detail
+
+
+# ---------------------------------------------------------------------------
+# M4-T06 — Recovery codes can be used for login (one-time)
+# ---------------------------------------------------------------------------
+
+
+class TestRecoveryCodeLogin:
+    """Recovery codes work as second factor and are consumed after use."""
+
+    def test_recovery_code_login_succeeds(self, totp_login_client: TestClient) -> None:
+        _, recovery_codes = _enable_totp(totp_login_client)
+        code = recovery_codes[0]
+        resp = totp_login_client.post(
+            "/api/auth/login",
+            json={"username": "admin", "password": "test-password", "totp_code": code},
+        )
+        assert resp.status_code == 200, resp.text
+
+    def test_recovery_code_is_one_time_only(self, totp_login_client: TestClient) -> None:
+        """Using the same recovery code twice: second use must fail (401)."""
+        _, recovery_codes = _enable_totp(totp_login_client)
+        code = recovery_codes[0]
+
+        # First use: success
+        resp1 = totp_login_client.post(
+            "/api/auth/login",
+            json={"username": "admin", "password": "test-password", "totp_code": code},
+        )
+        assert resp1.status_code == 200, resp1.text
+
+        # Log out
+        csrf = resp1.json()["csrf_token"]
+        totp_login_client.post("/api/auth/logout", headers={"X-CSRF-Token": csrf})
+
+        # Second use: must fail
+        resp2 = totp_login_client.post(
+            "/api/auth/login",
+            json={"username": "admin", "password": "test-password", "totp_code": code},
+        )
+        assert resp2.status_code == 401, (
+            "Recovery code must be rejected on second use"
+        )
+
+    def test_recovery_code_used_at_set_in_db(self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> None:
+        """After consuming a recovery code, its used_at is set in the DB."""
+        from app.models.auth import RecoveryCode
+        from sqlalchemy import select
+
+        db_path = tmp_path / "rc_used_test.db"
+        db_url = f"sqlite:///{db_path}"
+
+        monkeypatch.setenv("APP_DATABASE_URL", db_url)
+        monkeypatch.setenv("AUTH_BOOTSTRAP_USERNAME", "admin")
+        monkeypatch.setenv("AUTH_BOOTSTRAP_PASSWORD", "test-password")
+        monkeypatch.setenv("AUTH_COOKIE_SECURE_OVERRIDE", "false")
+        monkeypatch.setenv("AUTH_LOGIN_THROTTLE_ENABLED", "false")
+        get_settings.cache_clear()
+        reset_db_caches()
+        command.upgrade(_make_app_alembic_config(db_url), "head")
+        reset_db_caches()
+
+        engine = create_engine(db_url, connect_args={"check_same_thread": False})
+        app = create_app()
+        with TestClient(app) as tc:
+            secret, recovery_codes = _enable_totp(tc)
+            code = recovery_codes[0]
+
+            tc.post(
+                "/api/auth/login",
+                json={"username": "admin", "password": "test-password", "totp_code": code},
+            )
+
+            with Session(engine) as db:
+                used_codes = db.execute(
+                    select(RecoveryCode).where(RecoveryCode.used_at.isnot(None))
+                ).scalars().all()
+                assert len(used_codes) >= 1, "At least one recovery code should have used_at set"
+
+        engine.dispose()
+        get_settings.cache_clear()
+        reset_db_caches()
+
+    def test_other_recovery_codes_still_work_after_one_consumed(
+        self, totp_login_client: TestClient
+    ) -> None:
+        """Consuming one recovery code leaves the others intact."""
+        _, recovery_codes = _enable_totp(totp_login_client)
+
+        # Use first code
+        resp1 = totp_login_client.post(
+            "/api/auth/login",
+            json={"username": "admin", "password": "test-password", "totp_code": recovery_codes[0]},
+        )
+        assert resp1.status_code == 200
+        csrf = resp1.json()["csrf_token"]
+        totp_login_client.post("/api/auth/logout", headers={"X-CSRF-Token": csrf})
+
+        # Second code must still work
+        resp2 = totp_login_client.post(
+            "/api/auth/login",
+            json={"username": "admin", "password": "test-password", "totp_code": recovery_codes[1]},
+        )
+        assert resp2.status_code == 200, (
+            "Second recovery code should still be valid after first is consumed"
+        )
+
+
+# ---------------------------------------------------------------------------
+# M4-T06 — Throttle interaction with TOTP second step
+# ---------------------------------------------------------------------------
+
+
+class TestThrottleWithTotpEnabled:
+    """Verify throttle semantics when TOTP is enabled:
+    - Missing code (totp_required) does NOT count as a throttle failure.
+    - Wrong TOTP/recovery code DOES count as a throttle failure.
+    - Throttle check at step 1 still gates step 2.
+    """
+
+    def test_missing_code_does_not_register_throttle_failure(
+        self, totp_login_throttle_client
+    ) -> None:
+        """Sending correct password without totp_code must not count as a failure.
+
+        If this test sends N_FREE+1 such requests without accumulating throttle
+        failures, the next request will still get 401 (totp_required) not 429.
+        """
+        client, engine = totp_login_throttle_client
+        _enable_totp(client)
+
+        # Send N_FREE + 2 requests with correct password but no totp_code.
+        for i in range(N_FREE + 2):
+            resp = client.post(
+                "/api/auth/login",
+                json={"username": "admin", "password": "test-password"},
+            )
+            # Must never become 429 — each is a legitimate first-step
+            assert resp.status_code == 401, (
+                f"Iteration {i}: expected 401 (totp_required), got {resp.status_code}"
+            )
+            body = resp.json()
+            assert isinstance(body["detail"], dict)
+            assert body["detail"].get("totp_required") is True
+
+    def test_wrong_totp_code_accumulates_throttle_failures(
+        self, totp_login_throttle_client
+    ) -> None:
+        """Wrong TOTP codes should eventually trigger 429."""
+        client, engine = totp_login_throttle_client
+        _enable_totp(client)
+
+        # Send N_FREE + 1 wrong-code requests — each should register a failure
+        for _ in range(N_FREE + 1):
+            client.post(
+                "/api/auth/login",
+                json={"username": "admin", "password": "test-password", "totp_code": "000000"},
+            )
+
+        # Next attempt (even with correct password + no code) should hit 429
+        resp = client.post(
+            "/api/auth/login",
+            json={"username": "admin", "password": "test-password"},
+        )
+        assert resp.status_code == 429, (
+            "After N_FREE+1 wrong TOTP codes the throttle should kick in"
+        )
+
+    def test_throttle_blocks_step_one_when_in_window(
+        self, totp_login_throttle_client
+    ) -> None:
+        """Once throttled (by wrong codes), even step-1 attempt is rejected 429."""
+        client, engine = totp_login_throttle_client
+        _enable_totp(client)
+
+        # Accumulate failures via wrong TOTP codes
+        for _ in range(N_FREE + 1):
+            client.post(
+                "/api/auth/login",
+                json={"username": "admin", "password": "test-password", "totp_code": "000000"},
+            )
+
+        # Throttle check is before password — even wrong password should get 429
+        resp = client.post(
+            "/api/auth/login",
+            json={"username": "admin", "password": "wrong"},
+        )
+        assert resp.status_code == 429

tests/test_api_totp.py

@@ -0,0 +1,529 @@
+"""Tests for M4-T05: TOTP service + setup/enable/disable/status API."""
+from __future__ import annotations
+
+import pyotp
+import pytest
+from fastapi.testclient import TestClient
+from sqlalchemy import create_engine, select
+from sqlalchemy.orm import Session
+
+from app.db import reset_db_caches
+from app.models.auth import AuthUser, RecoveryCode
+from app.services.auth import verify_password
+
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+
+def _login(client: TestClient) -> str:
+    """Log in and return the CSRF token from the response."""
+    resp = client.post(
+        "/api/auth/login",
+        json={"username": "admin", "password": "test-password"},
+    )
+    assert resp.status_code == 200, resp.text
+    return resp.json()["csrf_token"]
+
+
+def _setup(client: TestClient, csrf: str) -> dict:
+    resp = client.post(
+        "/api/auth/totp/setup",
+        headers={"X-CSRF-Token": csrf},
+    )
+    return resp
+
+
+def _enable(client: TestClient, csrf: str, code: str):  # noqa: ANN201
+    return client.post(
+        "/api/auth/totp/enable",
+        json={"code": code},
+        headers={"X-CSRF-Token": csrf},
+    )
+
+
+def _disable(client: TestClient, csrf: str, *, password: str | None = None, code: str | None = None):
+    body: dict = {}
+    if password is not None:
+        body["password"] = password
+    if code is not None:
+        body["code"] = code
+    return client.post(
+        "/api/auth/totp/disable",
+        json=body,
+        headers={"X-CSRF-Token": csrf},
+    )
+
+
+def _status(client: TestClient) -> dict:
+    resp = client.get("/api/auth/totp")
+    assert resp.status_code == 200, resp.text
+    return resp.json()
+
+
+# ---------------------------------------------------------------------------
+# Fixtures — reuse conftest app/client via the session-aware totp_client fixture
+# ---------------------------------------------------------------------------
+
+
+@pytest.fixture
+def totp_client(auth_database):
+    """TestClient + a direct SQLAlchemy engine so tests can inspect DB state."""
+    from app.main import create_app
+
+    app_url = auth_database["app_url"]
+    engine = create_engine(app_url, connect_args={"check_same_thread": False})
+    fastapi_app = create_app()
+    with TestClient(fastapi_app) as tc:
+        yield tc, engine
+    engine.dispose()
+    reset_db_caches()
+
+
+# ---------------------------------------------------------------------------
+# POST /api/auth/totp/setup
+# ---------------------------------------------------------------------------
+
+
+def test_setup_returns_secret_uri_and_recovery_codes(totp_client):
+    client, engine = totp_client
+    csrf = _login(client)
+
+    resp = _setup(client, csrf)
+
+    assert resp.status_code == 200, resp.text
+    body = resp.json()
+    assert "secret" in body
+    assert "otpauth_uri" in body
+    assert "recovery_codes" in body
+    assert len(body["recovery_codes"]) == 10
+    # Each code must match xxxx-xxxx pattern
+    for rc in body["recovery_codes"]:
+        parts = rc.split("-")
+        assert len(parts) == 2
+        assert all(len(p) == 4 for p in parts)
+    # URI must start with otpauth://totp/
+    assert body["otpauth_uri"].startswith("otpauth://totp/")
+    # Secret must be a valid base32 string (pyotp will accept it)
+    assert len(body["secret"]) >= 16
+
+
+def test_setup_does_not_enable_totp(totp_client):
+    """After setup, totp_enabled must still be False."""
+    client, engine = totp_client
+    csrf = _login(client)
+
+    _setup(client, csrf)
+
+    status_body = _status(client)
+    assert status_body["enabled"] is False
+
+
+def test_setup_persists_secret_and_code_hashes(totp_client):
+    """After setup, DB has totp_secret set and recovery code hashes stored."""
+    client, engine = totp_client
+    csrf = _login(client)
+
+    resp = _setup(client, csrf)
+    body = resp.json()
+
+    with Session(engine) as db:
+        user = db.scalar(select(AuthUser).where(AuthUser.username == "admin"))
+        assert user is not None
+        # Secret is stored
+        assert user.totp_secret == body["secret"]
+        # Enabled is still False
+        assert user.totp_enabled is False
+        # Recovery codes are stored as hashes
+        codes = db.execute(select(RecoveryCode).where(RecoveryCode.user_id == user.id)).scalars().all()
+        assert len(codes) == 10
+        # Hashes are not the same as plaintext
+        for i, rc in enumerate(codes):
+            assert rc.code_hash != body["recovery_codes"][i]
+            # But they must verify correctly with Argon2
+            assert verify_password(body["recovery_codes"][i], rc.code_hash)
+
+
+def test_setup_twice_replaces_pending_state(totp_client):
+    """Re-running setup clears old pending secret + codes and issues new ones."""
+    client, engine = totp_client
+    csrf = _login(client)
+
+    first = _setup(client, csrf).json()
+    second = _setup(client, csrf).json()
+
+    # Different secret each time
+    assert first["secret"] != second["secret"]
+    # Different codes
+    assert first["recovery_codes"] != second["recovery_codes"]
+
+    # DB has exactly 10 codes (not 20)
+    with Session(engine) as db:
+        user = db.scalar(select(AuthUser).where(AuthUser.username == "admin"))
+        codes = db.execute(
+            select(RecoveryCode).where(RecoveryCode.user_id == user.id)
+        ).scalars().all()
+        assert len(codes) == 10
+
+
+# ---------------------------------------------------------------------------
+# POST /api/auth/totp/enable
+# ---------------------------------------------------------------------------
+
+
+def test_enable_with_wrong_code_fails(totp_client):
+    client, engine = totp_client
+    csrf = _login(client)
+
+    _setup(client, csrf)
+
+    resp = _enable(client, csrf, "000000")
+    assert resp.status_code == 400
+
+
+def test_enable_with_wrong_code_does_not_enable(totp_client):
+    client, engine = totp_client
+    csrf = _login(client)
+
+    _setup(client, csrf)
+    _enable(client, csrf, "000000")
+
+    assert _status(client)["enabled"] is False
+
+
+def test_enable_with_correct_code_succeeds(totp_client):
+    """Using pyotp.TOTP(secret).now() to generate the correct code."""
+    client, engine = totp_client
+    csrf = _login(client)
+
+    setup_body = _setup(client, csrf).json()
+    secret = setup_body["secret"]
+
+    correct_code = pyotp.TOTP(secret).now()
+    resp = _enable(client, csrf, correct_code)
+    assert resp.status_code == 204, resp.text
+
+
+def test_enable_with_correct_code_sets_totp_enabled(totp_client):
+    client, engine = totp_client
+    csrf = _login(client)
+
+    setup_body = _setup(client, csrf).json()
+    secret = setup_body["secret"]
+
+    _enable(client, csrf, pyotp.TOTP(secret).now())
+
+    assert _status(client)["enabled"] is True
+
+
+def test_enable_persists_recovery_code_hashes(totp_client):
+    """Recovery codes are already hashed in DB after enable (not plaintext)."""
+    client, engine = totp_client
+    csrf = _login(client)
+
+    setup_body = _setup(client, csrf).json()
+    secret = setup_body["secret"]
+
+    _enable(client, csrf, pyotp.TOTP(secret).now())
+
+    with Session(engine) as db:
+        user = db.scalar(select(AuthUser).where(AuthUser.username == "admin"))
+        codes = db.execute(select(RecoveryCode).where(RecoveryCode.user_id == user.id)).scalars().all()
+        assert len(codes) == 10
+        for i, rc in enumerate(codes):
+            plaintext = setup_body["recovery_codes"][i]
+            # code_hash must NOT be the plaintext
+            assert rc.code_hash != plaintext
+            # But verify_password must return True (Argon2 hash matches)
+            assert verify_password(plaintext, rc.code_hash)
+
+
+# ---------------------------------------------------------------------------
+# POST /api/auth/totp/disable
+# ---------------------------------------------------------------------------
+
+
+def _do_full_enable(client: TestClient, csrf: str) -> str:
+    """Perform setup + enable and return the secret."""
+    setup_body = _setup(client, csrf).json()
+    secret = setup_body["secret"]
+    _enable(client, csrf, pyotp.TOTP(secret).now())
+    return secret
+
+
+def test_disable_with_password_succeeds(totp_client):
+    client, engine = totp_client
+    csrf = _login(client)
+
+    _do_full_enable(client, csrf)
+    assert _status(client)["enabled"] is True
+
+    resp = _disable(client, csrf, password="test-password")
+    assert resp.status_code == 204, resp.text
+
+
+def test_disable_clears_enabled_flag(totp_client):
+    client, engine = totp_client
+    csrf = _login(client)
+
+    _do_full_enable(client, csrf)
+    _disable(client, csrf, password="test-password")
+
+    assert _status(client)["enabled"] is False
+
+
+def test_disable_clears_secret_and_recovery_codes(totp_client):
+    client, engine = totp_client
+    csrf = _login(client)
+
+    _do_full_enable(client, csrf)
+    _disable(client, csrf, password="test-password")
+
+    with Session(engine) as db:
+        user = db.scalar(select(AuthUser).where(AuthUser.username == "admin"))
+        assert user is not None
+        assert user.totp_secret is None
+        assert user.totp_enabled is False
+        codes = db.execute(select(RecoveryCode).where(RecoveryCode.user_id == user.id)).scalars().all()
+        assert len(codes) == 0
+
+
+def test_disable_with_totp_code_succeeds(totp_client):
+    client, engine = totp_client
+    csrf = _login(client)
+
+    secret = _do_full_enable(client, csrf)
+    current_code = pyotp.TOTP(secret).now()
+
+    resp = _disable(client, csrf, code=current_code)
+    assert resp.status_code == 204, resp.text
+    assert _status(client)["enabled"] is False
+
+
+def test_disable_with_wrong_password_fails(totp_client):
+    client, engine = totp_client
+    csrf = _login(client)
+
+    _do_full_enable(client, csrf)
+
+    resp = _disable(client, csrf, password="wrong-password")
+    assert resp.status_code == 400
+    assert _status(client)["enabled"] is True
+
+
+def test_disable_with_no_credential_fails(totp_client):
+    client, engine = totp_client
+    csrf = _login(client)
+
+    _do_full_enable(client, csrf)
+
+    resp = _disable(client, csrf)  # neither password nor code
+    assert resp.status_code == 400
+
+
+# ---------------------------------------------------------------------------
+# GET /api/auth/totp — status endpoint
+# ---------------------------------------------------------------------------
+
+
+def test_status_does_not_return_secret(totp_client):
+    client, engine = totp_client
+    csrf = _login(client)
+
+    setup_body = _setup(client, csrf).json()
+    secret = setup_body["secret"]
+    _enable(client, csrf, pyotp.TOTP(secret).now())
+
+    status_resp = client.get("/api/auth/totp")
+    assert status_resp.status_code == 200
+    body = status_resp.json()
+    # Must contain ONLY enabled
+    assert set(body.keys()) == {"enabled"}
+    assert "secret" not in str(body)
+    assert "recovery_codes" not in str(body)
+    assert "otpauth" not in str(body)
+
+
+def test_status_initially_disabled(totp_client):
+    client, _ = totp_client
+    _login(client)
+
+    body = _status(client)
+    assert body["enabled"] is False
+
+
+def test_status_after_enable(totp_client):
+    client, _ = totp_client
+    csrf = _login(client)
+
+    setup_body = _setup(client, csrf).json()
+    _enable(client, csrf, pyotp.TOTP(setup_body["secret"]).now())
+
+    body = _status(client)
+    assert body["enabled"] is True
+
+
+# ---------------------------------------------------------------------------
+# Security: unauthenticated / missing CSRF
+# ---------------------------------------------------------------------------
+
+
+def test_setup_unauthenticated_returns_401(client: TestClient):
+    resp = client.post("/api/auth/totp/setup", headers={"X-CSRF-Token": "x"})
+    assert resp.status_code == 401
+
+
+def test_setup_missing_csrf_returns_403(totp_client):
+    client, _ = totp_client
+    _login(client)
+    # No X-CSRF-Token header
+    resp = client.post("/api/auth/totp/setup")
+    assert resp.status_code == 403
+
+
+def test_enable_unauthenticated_returns_401(client: TestClient):
+    resp = client.post(
+        "/api/auth/totp/enable",
+        json={"code": "123456"},
+        headers={"X-CSRF-Token": "x"},
+    )
+    assert resp.status_code == 401
+
+
+def test_enable_missing_csrf_returns_403(totp_client):
+    client, _ = totp_client
+    _login(client)
+    resp = client.post("/api/auth/totp/enable", json={"code": "123456"})
+    assert resp.status_code == 403
+
+
+def test_disable_unauthenticated_returns_401(client: TestClient):
+    resp = client.post(
+        "/api/auth/totp/disable",
+        json={"password": "test-password"},
+        headers={"X-CSRF-Token": "x"},
+    )
+    assert resp.status_code == 401
+
+
+def test_disable_missing_csrf_returns_403(totp_client):
+    client, _ = totp_client
+    _login(client)
+    resp = client.post("/api/auth/totp/disable", json={"password": "test-password"})
+    assert resp.status_code == 403
+
+
+def test_status_unauthenticated_returns_401(client: TestClient):
+    resp = client.get("/api/auth/totp")
+    assert resp.status_code == 401
+
+
+# ---------------------------------------------------------------------------
+# M4-T06 — verify_recovery_code service-level tests
+# ---------------------------------------------------------------------------
+
+
+def _full_setup_and_enable(client: TestClient) -> tuple[str, list[str]]:
+    """Log in, run setup, enable TOTP, return (secret, plaintext_recovery_codes)."""
+    csrf = _login(client)
+    setup_body = _setup(client, csrf).json()
+    secret = setup_body["secret"]
+    recovery_codes = setup_body["recovery_codes"]
+    _enable(client, csrf, pyotp.TOTP(secret).now())
+    return secret, recovery_codes
+
+
+class TestVerifyRecoveryCodeService:
+    """Unit-level checks for totp_service.verify_recovery_code (used in T06 login)."""
+
+    def test_verify_recovery_code_returns_true_on_valid_code(self, totp_client):
+        """verify_recovery_code returns True for an unused code."""
+        from app.models.auth import AuthUser
+        from sqlalchemy import select
+        from app.services import totp as totp_svc
+
+        client, engine = totp_client
+        _, recovery_codes = _full_setup_and_enable(client)
+        code = recovery_codes[0]
+
+        with Session(engine) as db:
+            user = db.scalar(select(AuthUser).where(AuthUser.username == "admin"))
+            result = totp_svc.verify_recovery_code(db, user=user, code=code)
+            assert result is True
+
+    def test_verify_recovery_code_returns_false_on_used_code(self, totp_client):
+        """verify_recovery_code returns False when the code was already consumed."""
+        from app.models.auth import AuthUser
+        from sqlalchemy import select
+        from app.services import totp as totp_svc
+
+        client, engine = totp_client
+        _, recovery_codes = _full_setup_and_enable(client)
+        code = recovery_codes[0]
+
+        with Session(engine) as db:
+            user = db.scalar(select(AuthUser).where(AuthUser.username == "admin"))
+            # First call consumes
+            assert totp_svc.verify_recovery_code(db, user=user, code=code) is True
+            # Reload user after commit
+            db.expire_all()
+            user = db.scalar(select(AuthUser).where(AuthUser.username == "admin"))
+            # Second call must fail
+            assert totp_svc.verify_recovery_code(db, user=user, code=code) is False
+
+    def test_verify_recovery_code_returns_false_on_invalid_code(self, totp_client):
+        """verify_recovery_code returns False for a completely wrong code."""
+        from app.models.auth import AuthUser
+        from sqlalchemy import select
+        from app.services import totp as totp_svc
+
+        client, engine = totp_client
+        _full_setup_and_enable(client)
+
+        with Session(engine) as db:
+            user = db.scalar(select(AuthUser).where(AuthUser.username == "admin"))
+            result = totp_svc.verify_recovery_code(db, user=user, code="xxxx-xxxx")
+            assert result is False
+
+    def test_verify_totp_code_correct(self, totp_client):
+        """verify_totp_code returns True for the current TOTP code."""
+        from app.models.auth import AuthUser
+        from sqlalchemy import select
+        from app.services import totp as totp_svc
+
+        client, engine = totp_client
+        secret, _ = _full_setup_and_enable(client)
+
+        with Session(engine) as db:
+            user = db.scalar(select(AuthUser).where(AuthUser.username == "admin"))
+            result = totp_svc.verify_totp_code(user, pyotp.TOTP(secret).now())
+            assert result is True
+
+    def test_verify_totp_code_wrong(self, totp_client):
+        """verify_totp_code returns False for a wrong code."""
+        from app.models.auth import AuthUser
+        from sqlalchemy import select
+        from app.services import totp as totp_svc
+
+        client, engine = totp_client
+        _full_setup_and_enable(client)
+
+        with Session(engine) as db:
+            user = db.scalar(select(AuthUser).where(AuthUser.username == "admin"))
+            result = totp_svc.verify_totp_code(user, "000000")
+            assert result is False
+
+    def test_verify_totp_code_no_secret(self, totp_client):
+        """verify_totp_code returns False if user has no secret (TOTP not set up)."""
+        from app.models.auth import AuthUser
+        from sqlalchemy import select
+        from app.services import totp as totp_svc
+
+        client, engine = totp_client
+
+        with Session(engine) as db:
+            user = db.scalar(select(AuthUser).where(AuthUser.username == "admin"))
+            assert user.totp_secret is None
+            result = totp_svc.verify_totp_code(user, "123456")
+            assert result is False

tests/test_auth_throttle_model.py

@@ -0,0 +1,200 @@
+"""Tests for M4-T01: LoginThrottle model and auth_login_throttle migration."""
+from __future__ import annotations
+
+from datetime import datetime, timezone
+from pathlib import Path
+
+import pytest
+from alembic import command
+from alembic.config import Config
+from sqlalchemy import create_engine, inspect
+from sqlalchemy.exc import IntegrityError
+from sqlalchemy.orm import Session
+
+from app.models.auth_throttle import LoginThrottle
+
+
+def _make_app_alembic_config(database_url: str) -> Config:
+    config = Config("alembic_app.ini")
+    config.set_main_option("sqlalchemy.url", database_url)
+    return config
+
+
+# ---------------------------------------------------------------------------
+# Fixtures
+# ---------------------------------------------------------------------------
+
+
+@pytest.fixture()
+def throttle_db(tmp_path: Path):
+    """Temporary SQLite DB upgraded to the current Alembic head."""
+    db_path = tmp_path / "throttle_test.db"
+    db_url = f"sqlite:///{db_path}"
+    alembic_cfg = _make_app_alembic_config(db_url)
+    command.upgrade(alembic_cfg, "head")
+    engine = create_engine(db_url, connect_args={"check_same_thread": False})
+    yield engine
+    engine.dispose()
+
+
+# ---------------------------------------------------------------------------
+# Migration shape tests
+# ---------------------------------------------------------------------------
+
+
+def test_table_exists_after_upgrade(throttle_db):
+    """auth_login_throttle table must be present after upgrade to head."""
+    inspector = inspect(throttle_db)
+    assert "auth_login_throttle" in inspector.get_table_names()
+
+
+def test_table_has_expected_columns(throttle_db):
+    """All required columns must be present with correct nullability."""
+    inspector = inspect(throttle_db)
+    columns = {col["name"]: col for col in inspector.get_columns("auth_login_throttle")}
+
+    expected_non_nullable = {"id", "key", "scope", "failures", "first_failed_at", "last_failed_at"}
+    expected_nullable = {"next_allowed_at"}
+
+    for col_name in expected_non_nullable:
+        assert col_name in columns, f"Missing column: {col_name}"
+        assert not columns[col_name]["nullable"], f"Column {col_name} should be NOT NULL"
+
+    for col_name in expected_nullable:
+        assert col_name in columns, f"Missing column: {col_name}"
+        assert columns[col_name]["nullable"], f"Column {col_name} should be nullable"
+
+
+def test_unique_constraint_scope_key(throttle_db):
+    """(scope, key) pair must be unique — inserting a duplicate must raise."""
+    now = datetime.now(tz=timezone.utc)
+    with Session(throttle_db) as session:
+        session.add(
+            LoginThrottle(
+                key="192.168.1.1",
+                scope="ip",
+                failures=1,
+                first_failed_at=now,
+                last_failed_at=now,
+            )
+        )
+        session.commit()
+
+    with Session(throttle_db) as session:
+        session.add(
+            LoginThrottle(
+                key="192.168.1.1",
+                scope="ip",
+                failures=2,
+                first_failed_at=now,
+                last_failed_at=now,
+            )
+        )
+        with pytest.raises(IntegrityError):
+            session.commit()
+
+
+def test_same_key_different_scope_allowed(throttle_db):
+    """Same key string but different scopes are independent rows — both should insert."""
+    now = datetime.now(tz=timezone.utc)
+    with Session(throttle_db) as session:
+        session.add(
+            LoginThrottle(
+                key="admin",
+                scope="ip",
+                failures=1,
+                first_failed_at=now,
+                last_failed_at=now,
+            )
+        )
+        session.add(
+            LoginThrottle(
+                key="admin",
+                scope="user",
+                failures=3,
+                first_failed_at=now,
+                last_failed_at=now,
+            )
+        )
+        session.commit()  # must not raise
+
+    with Session(throttle_db) as session:
+        rows = session.query(LoginThrottle).filter_by(key="admin").all()
+        assert len(rows) == 2
+        scopes = {r.scope for r in rows}
+        assert scopes == {"ip", "user"}
+
+
+def test_next_allowed_at_nullable(throttle_db):
+    """next_allowed_at is nullable — a row without it must persist fine."""
+    now = datetime.now(tz=timezone.utc)
+    with Session(throttle_db) as session:
+        row = LoginThrottle(
+            key="10.0.0.1",
+            scope="ip",
+            failures=1,
+            first_failed_at=now,
+            last_failed_at=now,
+            next_allowed_at=None,
+        )
+        session.add(row)
+        session.commit()
+        session.refresh(row)
+        assert row.next_allowed_at is None
+
+
+def test_next_allowed_at_persists_when_set(throttle_db):
+    """next_allowed_at can be set and retrieved correctly."""
+    now = datetime.now(tz=timezone.utc)
+    with Session(throttle_db) as session:
+        row = LoginThrottle(
+            key="10.0.0.2",
+            scope="ip",
+            failures=5,
+            first_failed_at=now,
+            last_failed_at=now,
+            next_allowed_at=now,
+        )
+        session.add(row)
+        session.commit()
+        row_id = row.id
+
+    with Session(throttle_db) as session:
+        fetched = session.get(LoginThrottle, row_id)
+        assert fetched is not None
+        assert fetched.next_allowed_at is not None
+
+
+# ---------------------------------------------------------------------------
+# Downgrade / upgrade reversibility test
+# ---------------------------------------------------------------------------
+
+
+def test_downgrade_removes_table(tmp_path: Path):
+    """Downgrading the auth_login_throttle revision must cleanly drop the table.
+
+    We upgrade to the specific throttle revision (not necessarily head, since later
+    revisions may have been added), then downgrade one step to verify the table is gone.
+    """
+    db_path = tmp_path / "down_test.db"
+    db_url = f"sqlite:///{db_path}"
+    alembic_cfg = _make_app_alembic_config(db_url)
+    throttle_revision = "20260621_07_auth_login_throttle"
+
+    command.upgrade(alembic_cfg, throttle_revision)
+    engine = create_engine(db_url, connect_args={"check_same_thread": False})
+    assert "auth_login_throttle" in inspect(engine).get_table_names()
+    engine.dispose()
+
+    command.downgrade(alembic_cfg, "-1")
+    engine = create_engine(db_url, connect_args={"check_same_thread": False})
+    assert "auth_login_throttle" not in inspect(engine).get_table_names()
+    engine.dispose()
+
+
+def test_index_exists_on_scope_key(throttle_db):
+    """ix_auth_login_throttle_scope_key composite index must exist."""
+    inspector = inspect(throttle_db)
+    indexes = inspector.get_indexes("auth_login_throttle")
+    index_names = {idx["name"] for idx in indexes}
+    assert "ix_auth_login_throttle_scope_key" in index_names

tests/test_login_throttle.py

@@ -0,0 +1,417 @@
+"""Tests for M4-T02: login throttle service and endpoint integration."""
+from __future__ import annotations
+
+from datetime import UTC, datetime, timedelta
+from pathlib import Path
+
+import pytest
+from alembic import command
+from alembic.config import Config
+from fastapi.testclient import TestClient
+from sqlalchemy import create_engine
+from sqlalchemy.orm import sessionmaker, Session
+
+from app.db import reset_db_caches
+from app.config import get_settings
+from app.main import create_app
+from app.models.auth_throttle import LoginThrottle
+from app.services import login_throttle as svc
+from app.services.login_throttle import (
+    N_FREE,
+    BACKOFF_BASE_S,
+    BACKOFF_CAP_S,
+    _compute_next_allowed_at,
+)
+
+
+# ---------------------------------------------------------------------------
+# Fixtures — throttle-aware variants (throttle enabled)
+# ---------------------------------------------------------------------------
+
+
+def _make_app_alembic_config(database_url: str) -> Config:
+    config = Config("alembic_app.ini")
+    config.set_main_option("sqlalchemy.url", database_url)
+    return config
+
+
+@pytest.fixture()
+def throttle_db(tmp_path: Path, monkeypatch: pytest.MonkeyPatch):
+    """Fresh migrated DB with throttle enabled (AUTH_LOGIN_THROTTLE_ENABLED=true)."""
+    app_database_path = tmp_path / "throttle_test.db"
+    app_database_url = f"sqlite:///{app_database_path}"
+
+    monkeypatch.setenv("APP_DATABASE_URL", app_database_url)
+    monkeypatch.setenv("AUTH_BOOTSTRAP_USERNAME", "admin")
+    monkeypatch.setenv("AUTH_BOOTSTRAP_PASSWORD", "test-password")
+    monkeypatch.setenv("AUTH_COOKIE_SECURE_OVERRIDE", "false")
+    monkeypatch.setenv("AUTH_LOGIN_THROTTLE_ENABLED", "true")
+    get_settings.cache_clear()
+    reset_db_caches()
+
+    command.upgrade(_make_app_alembic_config(app_database_url), "head")
+    reset_db_caches()
+
+    yield app_database_url
+
+    get_settings.cache_clear()
+    reset_db_caches()
+
+
+@pytest.fixture()
+def throttle_session(throttle_db: str) -> Session:
+    """SQLAlchemy session for the throttle-enabled test DB."""
+    engine = create_engine(throttle_db, connect_args={"check_same_thread": False})
+    SessionLocal = sessionmaker(bind=engine, autoflush=False, autocommit=False, class_=Session)
+    session = SessionLocal()
+    try:
+        yield session
+    finally:
+        session.close()
+        engine.dispose()
+
+
+@pytest.fixture()
+def throttle_client(throttle_db: str):
+    """TestClient wired to a throttle-enabled DB."""
+    app = create_app()
+    with TestClient(app) as client:
+        yield client
+
+
+# ---------------------------------------------------------------------------
+# Unit tests — service internals
+# ---------------------------------------------------------------------------
+
+
+class TestComputeNextAllowedAt:
+    """Verify the exponential formula independently of DB I/O."""
+
+    def test_no_delay_for_first_n_free_failures(self):
+        now = datetime.now(UTC)
+        for f in range(1, N_FREE + 1):
+            result = _compute_next_allowed_at(f, now)
+            assert result is None, f"Expected None for failures={f}"
+
+    def test_delay_starts_after_n_free(self):
+        now = datetime.now(UTC)
+        result = _compute_next_allowed_at(N_FREE + 1, now)
+        assert result is not None
+        expected_wait = BACKOFF_BASE_S * (2 ** (N_FREE + 1 - N_FREE))  # base * 2^1
+        assert abs((result - now).total_seconds() - expected_wait) < 0.1
+
+    def test_exponential_growth(self):
+        now = datetime.now(UTC)
+        prev_wait = 0.0
+        for failures in range(N_FREE + 1, N_FREE + 8):
+            result = _compute_next_allowed_at(failures, now)
+            assert result is not None
+            wait = (result - now).total_seconds()
+            if prev_wait > 0:
+                # Each step should double (until cap)
+                if wait < BACKOFF_CAP_S:
+                    assert abs(wait - prev_wait * 2) < 0.1, (
+                        f"Expected doubling at failures={failures}: {prev_wait} → {wait}"
+                    )
+            prev_wait = wait
+
+    def test_cap_is_respected(self):
+        now = datetime.now(UTC)
+        # Use a very high failure count to guarantee we're past the cap threshold.
+        result = _compute_next_allowed_at(N_FREE + 20, now)
+        assert result is not None
+        wait = (result - now).total_seconds()
+        assert wait <= BACKOFF_CAP_S + 0.1
+
+
+class TestRegisterFailureAndCheck:
+    """Service functions against an in-memory DB."""
+
+    def test_single_failure_within_free_window_no_wait(self, throttle_session: Session):
+        svc.register_failure(throttle_session, ip="1.2.3.4", username="alice")
+        wait = svc.check_and_get_wait(throttle_session, ip="1.2.3.4", username="alice")
+        assert wait == 0, "First failure should not produce any wait"
+
+    def test_free_failures_all_zero_wait(self, throttle_session: Session):
+        for _ in range(N_FREE):
+            svc.register_failure(throttle_session, ip="10.0.0.1", username="bob")
+        wait = svc.check_and_get_wait(throttle_session, ip="10.0.0.1", username="bob")
+        assert wait == 0
+
+    def test_fourth_failure_produces_wait(self, throttle_session: Session):
+        """failures = N_FREE + 1 should produce a non-zero wait."""
+        for _ in range(N_FREE + 1):
+            svc.register_failure(throttle_session, ip="10.0.0.2", username="carol")
+        wait = svc.check_and_get_wait(throttle_session, ip="10.0.0.2", username="carol")
+        assert wait > 0
+
+    def test_wait_grows_with_more_failures(self, throttle_session: Session):
+        ip, user = "10.0.0.3", "dave"
+        # Accumulate past the free window
+        for _ in range(N_FREE + 1):
+            svc.register_failure(throttle_session, ip=ip, username=user)
+        wait1 = svc.check_and_get_wait(throttle_session, ip=ip, username=user)
+
+        svc.register_failure(throttle_session, ip=ip, username=user)
+        wait2 = svc.check_and_get_wait(throttle_session, ip=ip, username=user)
+
+        assert wait2 > wait1, "Wait should increase with more failures"
+
+    def test_wait_is_capped(self, throttle_session: Session):
+        ip, user = "10.0.0.4", "eve"
+        # Simulate a high failure count by calling register_failure many times.
+        for _ in range(N_FREE + 15):
+            svc.register_failure(throttle_session, ip=ip, username=user)
+        wait = svc.check_and_get_wait(throttle_session, ip=ip, username=user)
+        assert wait <= BACKOFF_CAP_S, f"Wait {wait} exceeds cap {BACKOFF_CAP_S}"
+
+    def test_clear_removes_rows(self, throttle_session: Session):
+        ip, user = "10.0.0.5", "frank"
+        for _ in range(N_FREE + 2):
+            svc.register_failure(throttle_session, ip=ip, username=user)
+        assert svc.check_and_get_wait(throttle_session, ip=ip, username=user) > 0
+
+        svc.clear(throttle_session, ip=ip, username=user)
+        assert svc.check_and_get_wait(throttle_session, ip=ip, username=user) == 0
+
+    def test_clear_is_idempotent(self, throttle_session: Session):
+        """Calling clear when no rows exist must not raise."""
+        svc.clear(throttle_session, ip="192.168.1.1", username="nobody")
+        assert svc.check_and_get_wait(throttle_session, ip="192.168.1.1", username="nobody") == 0
+
+    def test_dual_key_max_is_used(self, throttle_session: Session):
+        """If only the username key is in backoff, check returns its wait time."""
+        # Register enough failures for username but use a fresh IP.
+        user = "grace"
+        for _ in range(N_FREE + 1):
+            svc.register_failure(throttle_session, ip="172.16.0.1", username=user)
+
+        # Fresh IP, same username — username key should dominate.
+        wait = svc.check_and_get_wait(throttle_session, ip="172.16.0.2", username=user)
+        assert wait > 0, "Username key should dominate even with a fresh IP"
+
+    def test_expired_window_is_zero(self, throttle_session: Session):
+        """A row whose next_allowed_at is in the past should yield 0."""
+        now = datetime.now(UTC)
+        past = now - timedelta(seconds=10)
+        row = LoginThrottle(
+            scope="ip",
+            key="192.168.99.1",
+            failures=N_FREE + 2,
+            first_failed_at=past,
+            last_failed_at=past,
+            next_allowed_at=past,
+        )
+        throttle_session.add(row)
+        throttle_session.commit()
+
+        wait = svc.check_and_get_wait(throttle_session, ip="192.168.99.1", username="nobody")
+        assert wait == 0
+
+
+# ---------------------------------------------------------------------------
+# Integration tests — POST /api/auth/login endpoint
+# ---------------------------------------------------------------------------
+
+
+def _bad_login(client: TestClient, *, username: str = "admin", password: str = "wrong") -> int:
+    resp = client.post("/api/auth/login", json={"username": username, "password": password})
+    return resp.status_code
+
+
+def _good_login(client: TestClient) -> int:
+    resp = client.post(
+        "/api/auth/login", json={"username": "admin", "password": "test-password"}
+    )
+    return resp.status_code
+
+
+class TestLoginEndpointThrottle:
+    def test_first_few_failures_return_401(self, throttle_client: TestClient):
+        """First N_FREE failures should still return 401, not 429."""
+        for _ in range(N_FREE):
+            code = _bad_login(throttle_client)
+            assert code == 401, f"Expected 401 within free window, got {code}"
+
+    def test_429_after_enough_failures(self, throttle_client: TestClient):
+        """After N_FREE+1 failures the endpoint must return 429."""
+        for _ in range(N_FREE + 1):
+            _bad_login(throttle_client)
+        resp = throttle_client.post(
+            "/api/auth/login", json={"username": "admin", "password": "wrong"}
+        )
+        assert resp.status_code == 429
+
+    def test_429_includes_retry_after_header(self, throttle_client: TestClient):
+        """429 response must carry a Retry-After header with a positive integer."""
+        for _ in range(N_FREE + 1):
+            _bad_login(throttle_client)
+        resp = throttle_client.post(
+            "/api/auth/login", json={"username": "admin", "password": "wrong"}
+        )
+        assert resp.status_code == 429
+        retry_after = resp.headers.get("retry-after")
+        assert retry_after is not None, "Retry-After header missing"
+        assert int(retry_after) > 0, "Retry-After must be positive"
+
+    def test_correct_password_blocked_in_window(self, throttle_client: TestClient):
+        """Even a correct password must be rejected (429) while inside backoff window."""
+        for _ in range(N_FREE + 1):
+            _bad_login(throttle_client)
+        resp = throttle_client.post(
+            "/api/auth/login", json={"username": "admin", "password": "test-password"}
+        )
+        assert resp.status_code == 429, (
+            "Correct password should still be blocked while in backoff window"
+        )
+
+    def test_success_clears_throttle(self, throttle_client: TestClient):
+        """A successful login must clear the backoff so the next attempt is free."""
+        # Register some failures (but not enough to trigger 429).
+        for _ in range(N_FREE):
+            _bad_login(throttle_client)
+
+        # Successful login clears state.
+        assert _good_login(throttle_client) == 200
+
+        # Subsequent bad attempts should be counted fresh (no residual backoff).
+        code = _bad_login(throttle_client)
+        assert code == 401, "After clear, first new failure should be 401 not 429"
+
+    def test_401_uses_generic_message(self, throttle_client: TestClient):
+        """401 detail must be generic — must not reveal whether the user exists."""
+        resp = throttle_client.post(
+            "/api/auth/login", json={"username": "nonexistent", "password": "wrong"}
+        )
+        assert resp.status_code == 401
+        detail = resp.json().get("detail", "")
+        assert "invalid username or password" == detail
+        assert "user" not in detail.lower() or "username" in detail.lower()
+        # Must NOT say things like "user not found" or "user does not exist"
+        assert "not found" not in detail
+        assert "does not exist" not in detail
+
+    def test_unknown_user_401_has_same_message(self, throttle_client: TestClient):
+        """Unknown username and wrong password for known user give identical 401 detail."""
+        resp_known = throttle_client.post(
+            "/api/auth/login", json={"username": "admin", "password": "wrong"}
+        )
+        resp_unknown = throttle_client.post(
+            "/api/auth/login", json={"username": "nobody", "password": "wrong"}
+        )
+        assert resp_known.status_code == 401
+        assert resp_unknown.status_code == 401
+        assert resp_known.json()["detail"] == resp_unknown.json()["detail"]
+
+    def test_retry_after_grows_with_more_failures(self, throttle_client: TestClient):
+        """Retry-After should be strictly larger after more failures (exponential)."""
+        # Accumulate past the free window.
+        for _ in range(N_FREE + 1):
+            _bad_login(throttle_client)
+        resp1 = throttle_client.post(
+            "/api/auth/login", json={"username": "admin", "password": "wrong"}
+        )
+        assert resp1.status_code == 429
+        wait1 = int(resp1.headers["retry-after"])
+
+        # One more failure → longer wait.
+        _bad_login(throttle_client)
+        resp2 = throttle_client.post(
+            "/api/auth/login", json={"username": "admin", "password": "wrong"}
+        )
+        assert resp2.status_code == 429
+        wait2 = int(resp2.headers["retry-after"])
+
+        assert wait2 >= wait1, "Retry-After should not decrease with more failures"
+
+    def test_xff_header_used_when_trust_enabled(
+        self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+    ):
+        """When AUTH_TRUST_FORWARDED_FOR=true the leftmost XFF IP is used as the IP key."""
+        app_database_path = tmp_path / "xff_test.db"
+        app_database_url = f"sqlite:///{app_database_path}"
+
+        monkeypatch.setenv("APP_DATABASE_URL", app_database_url)
+        monkeypatch.setenv("AUTH_BOOTSTRAP_USERNAME", "admin")
+        monkeypatch.setenv("AUTH_BOOTSTRAP_PASSWORD", "test-password")
+        monkeypatch.setenv("AUTH_COOKIE_SECURE_OVERRIDE", "false")
+        monkeypatch.setenv("AUTH_LOGIN_THROTTLE_ENABLED", "true")
+        monkeypatch.setenv("AUTH_TRUST_FORWARDED_FOR", "true")
+        get_settings.cache_clear()
+        reset_db_caches()
+
+        command.upgrade(_make_app_alembic_config(app_database_url), "head")
+        reset_db_caches()
+
+        app = create_app()
+        with TestClient(app) as client:
+            # Accumulate failures from XFF IP "5.5.5.5".
+            for _ in range(N_FREE + 1):
+                client.post(
+                    "/api/auth/login",
+                    json={"username": "admin", "password": "wrong"},
+                    headers={"X-Forwarded-For": "5.5.5.5, 10.0.0.1"},
+                )
+
+            # Same XFF IP should now be in backoff.
+            resp = client.post(
+                "/api/auth/login",
+                json={"username": "admin", "password": "wrong"},
+                headers={"X-Forwarded-For": "5.5.5.5, 10.0.0.1"},
+            )
+            assert resp.status_code == 429
+
+            # Different XFF IP should NOT be in backoff (username might be, so use fresh one).
+            resp2 = client.post(
+                "/api/auth/login",
+                json={"username": "admin", "password": "wrong"},
+                headers={"X-Forwarded-For": "6.6.6.6"},
+            )
+            # This IP has no failures, but username "admin" DOES have N_FREE+1 failures.
+            # The username key is in backoff, so this should still 429.
+            # To cleanly test IP isolation we'd need a different username — but single admin
+            # means we can only confirm the XFF IP is being read (not defaulting to socket).
+            # The important thing: the request was processed with XFF logic (no crash).
+            assert resp2.status_code in (401, 429)
+
+        get_settings.cache_clear()
+        reset_db_caches()
+
+
+# ---------------------------------------------------------------------------
+# Throttle disabled — no-op behaviour
+# ---------------------------------------------------------------------------
+
+
+class TestThrottleDisabled:
+    """When AUTH_LOGIN_THROTTLE_ENABLED=false the endpoint behaves exactly as before."""
+
+    def test_many_failures_still_return_401_not_429(self, client: TestClient):
+        """With throttle off, no number of failures should produce 429."""
+        # The default `client` fixture has throttle disabled (AUTH_LOGIN_THROTTLE_ENABLED=false).
+        for _ in range(N_FREE + 5):
+            resp = client.post(
+                "/api/auth/login", json={"username": "admin", "password": "wrong"}
+            )
+            assert resp.status_code == 401, f"Expected 401 with throttle off, got {resp.status_code}"
+
+    def test_no_throttle_rows_written_when_disabled(
+        self, auth_database, monkeypatch: pytest.MonkeyPatch
+    ):
+        """With throttle disabled, no LoginThrottle rows are created."""
+        db_url = auth_database["app_url"]
+        app = create_app()
+        with TestClient(app) as client:
+            for _ in range(N_FREE + 2):
+                client.post(
+                    "/api/auth/login", json={"username": "admin", "password": "wrong"}
+                )
+
+        engine = create_engine(db_url, connect_args={"check_same_thread": False})
+        SessionLocal = sessionmaker(
+            bind=engine, autoflush=False, autocommit=False, class_=Session
+        )
+        with SessionLocal() as session:
+            count = session.query(LoginThrottle).count()
+        engine.dispose()
+        assert count == 0, f"Expected 0 throttle rows with throttle disabled, found {count}"

tests/test_totp_models.py

@@ -0,0 +1,280 @@
+"""Tests for M4-T04: AuthUser TOTP fields and auth_recovery_code migration."""
+from __future__ import annotations
+
+from datetime import datetime, timezone
+from pathlib import Path
+
+import pytest
+from alembic import command
+from alembic.config import Config
+from sqlalchemy import create_engine, inspect, text
+from sqlalchemy.orm import Session
+
+from app.models.auth import AuthUser, RecoveryCode
+
+
+def _make_app_alembic_config(database_url: str) -> Config:
+    config = Config("alembic_app.ini")
+    config.set_main_option("sqlalchemy.url", database_url)
+    return config
+
+
+# ---------------------------------------------------------------------------
+# Fixtures
+# ---------------------------------------------------------------------------
+
+
+@pytest.fixture()
+def totp_db(tmp_path: Path):
+    """Temporary SQLite DB upgraded to the current Alembic head."""
+    db_path = tmp_path / "totp_test.db"
+    db_url = f"sqlite:///{db_path}"
+    alembic_cfg = _make_app_alembic_config(db_url)
+    command.upgrade(alembic_cfg, "head")
+    engine = create_engine(db_url, connect_args={"check_same_thread": False})
+    yield engine
+    engine.dispose()
+
+
+# ---------------------------------------------------------------------------
+# Migration shape tests
+# ---------------------------------------------------------------------------
+
+
+def test_auth_users_has_totp_columns(totp_db):
+    """After upgrade to head, auth_users must have totp_secret and totp_enabled columns."""
+    inspector = inspect(totp_db)
+    columns = {col["name"]: col for col in inspector.get_columns("auth_users")}
+
+    assert "totp_secret" in columns, "totp_secret column missing from auth_users"
+    assert "totp_enabled" in columns, "totp_enabled column missing from auth_users"
+
+    # totp_secret must be nullable
+    assert columns["totp_secret"]["nullable"], "totp_secret should be nullable"
+    # totp_enabled must be NOT NULL
+    assert not columns["totp_enabled"]["nullable"], "totp_enabled should be NOT NULL"
+
+
+def test_auth_recovery_code_table_exists(totp_db):
+    """auth_recovery_code table must exist after upgrade to head."""
+    inspector = inspect(totp_db)
+    assert "auth_recovery_code" in inspector.get_table_names()
+
+
+def test_auth_recovery_code_has_expected_columns(totp_db):
+    """auth_recovery_code must have id, user_id, code_hash, used_at with correct nullability."""
+    inspector = inspect(totp_db)
+    columns = {col["name"]: col for col in inspector.get_columns("auth_recovery_code")}
+
+    expected_non_nullable = {"id", "user_id", "code_hash"}
+    expected_nullable = {"used_at"}
+
+    for col_name in expected_non_nullable:
+        assert col_name in columns, f"Missing column: {col_name}"
+        assert not columns[col_name]["nullable"], f"Column {col_name} should be NOT NULL"
+
+    for col_name in expected_nullable:
+        assert col_name in columns, f"Missing column: {col_name}"
+        assert columns[col_name]["nullable"], f"Column {col_name} should be nullable"
+
+
+def test_auth_recovery_code_has_fk_to_auth_users(totp_db):
+    """auth_recovery_code.user_id must have a FK constraint pointing to auth_users.id."""
+    inspector = inspect(totp_db)
+    fks = inspector.get_foreign_keys("auth_recovery_code")
+    assert len(fks) == 1, f"Expected 1 FK on auth_recovery_code, got {len(fks)}"
+    fk = fks[0]
+    assert fk["referred_table"] == "auth_users"
+    assert "user_id" in fk["constrained_columns"]
+    assert "id" in fk["referred_columns"]
+
+
+def test_auth_recovery_code_user_id_index_exists(totp_db):
+    """ix_auth_recovery_code_user_id index must exist."""
+    inspector = inspect(totp_db)
+    indexes = inspector.get_indexes("auth_recovery_code")
+    index_names = {idx["name"] for idx in indexes}
+    assert "ix_auth_recovery_code_user_id" in index_names
+
+
+# ---------------------------------------------------------------------------
+# Default value tests
+# ---------------------------------------------------------------------------
+
+
+def test_totp_enabled_defaults_to_false(totp_db):
+    """A newly inserted AuthUser must have totp_enabled=False and totp_secret=None."""
+    now = datetime.now(tz=timezone.utc)
+    with Session(totp_db) as session:
+        user = AuthUser(
+            username="testuser",
+            password_hash="$argon2id$v=19$m=65536,t=3,p=4$fakehash",
+            is_active=True,
+            force_password_change=False,
+            created_at=now,
+        )
+        session.add(user)
+        session.commit()
+        user_id = user.id
+
+    with Session(totp_db) as session:
+        fetched = session.get(AuthUser, user_id)
+        assert fetched is not None
+        assert fetched.totp_enabled is False, "totp_enabled should default to False"
+        assert fetched.totp_secret is None, "totp_secret should default to None"
+
+
+def test_totp_secret_can_be_set(totp_db):
+    """totp_secret can be set to a base32 string and retrieved correctly."""
+    now = datetime.now(tz=timezone.utc)
+    secret = "JBSWY3DPEHPK3PXP"
+    with Session(totp_db) as session:
+        user = AuthUser(
+            username="totp_user",
+            password_hash="$argon2id$v=19$m=65536,t=3,p=4$fakehash",
+            is_active=True,
+            force_password_change=False,
+            created_at=now,
+            totp_secret=secret,
+            totp_enabled=True,
+        )
+        session.add(user)
+        session.commit()
+        user_id = user.id
+
+    with Session(totp_db) as session:
+        fetched = session.get(AuthUser, user_id)
+        assert fetched is not None
+        assert fetched.totp_secret == secret
+        assert fetched.totp_enabled is True
+
+
+def test_recovery_code_insert_and_used_at_nullable(totp_db):
+    """A RecoveryCode can be inserted with used_at=None and later marked consumed."""
+    now = datetime.now(tz=timezone.utc)
+    with Session(totp_db) as session:
+        user = AuthUser(
+            username="rc_user",
+            password_hash="$argon2id$v=19$m=65536,t=3,p=4$fakehash",
+            is_active=True,
+            force_password_change=False,
+            created_at=now,
+        )
+        session.add(user)
+        session.flush()
+
+        code = RecoveryCode(
+            user_id=user.id,
+            code_hash="$argon2id$v=19$m=65536,t=3,p=4$rcfakehash",
+            used_at=None,
+        )
+        session.add(code)
+        session.commit()
+        code_id = code.id
+
+    with Session(totp_db) as session:
+        fetched = session.get(RecoveryCode, code_id)
+        assert fetched is not None
+        assert fetched.used_at is None
+
+        # Mark as consumed
+        fetched.used_at = now
+        session.commit()
+
+    with Session(totp_db) as session:
+        fetched = session.get(RecoveryCode, code_id)
+        assert fetched is not None
+        assert fetched.used_at is not None
+
+
+# ---------------------------------------------------------------------------
+# Existing user migration test (not-break-login guarantee)
+# ---------------------------------------------------------------------------
+
+
+def test_existing_user_defaults_to_totp_disabled_after_upgrade(tmp_path: Path):
+    """An existing user row inserted before the TOTP migration must have totp_enabled=0 after upgrade.
+
+    Simulates the real-world scenario: a pre-existing row in auth_users gets migrated
+    to the TOTP revision and the server_default ensures totp_enabled=false (0) for it.
+    """
+    db_path = tmp_path / "existing_user_test.db"
+    db_url = f"sqlite:///{db_path}"
+    alembic_cfg = _make_app_alembic_config(db_url)
+
+    # Upgrade to the TOTP revision's predecessor (the throttle revision).
+    command.upgrade(alembic_cfg, "20260621_07_auth_login_throttle")
+
+    # Insert a user row directly (simulates a user that existed before the TOTP migration).
+    engine = create_engine(db_url, connect_args={"check_same_thread": False})
+    with engine.connect() as conn:
+        conn.execute(
+            text(
+                "INSERT INTO auth_users (username, password_hash, is_active, force_password_change, created_at)"
+                " VALUES ('legacy_admin', 'hash', 1, 0, '2026-01-01T00:00:00+00:00')"
+            )
+        )
+        conn.commit()
+        row = conn.execute(
+            text("SELECT id FROM auth_users WHERE username = 'legacy_admin'")
+        ).fetchone()
+        legacy_user_id = row[0]
+    engine.dispose()
+
+    # Now upgrade to head (applies the TOTP migration).
+    command.upgrade(alembic_cfg, "head")
+
+    engine = create_engine(db_url, connect_args={"check_same_thread": False})
+    with engine.connect() as conn:
+        row = conn.execute(
+            text(
+                "SELECT totp_enabled, totp_secret FROM auth_users WHERE id = :uid"
+            ),
+            {"uid": legacy_user_id},
+        ).fetchone()
+    engine.dispose()
+
+    assert row is not None
+    # totp_enabled should be 0 (false) for the legacy row — server_default="0" ensures this.
+    assert row[0] == 0, f"Legacy user should have totp_enabled=0 after migration, got {row[0]}"
+    # totp_secret should be NULL for the legacy row.
+    assert row[1] is None, f"Legacy user should have totp_secret=NULL after migration, got {row[1]}"
+
+
+# ---------------------------------------------------------------------------
+# Downgrade / upgrade reversibility tests
+# ---------------------------------------------------------------------------
+
+
+def test_downgrade_removes_totp_columns_and_table(tmp_path: Path):
+    """downgrade -1 from head must drop auth_recovery_code and remove TOTP columns from auth_users."""
+    db_path = tmp_path / "down_totp_test.db"
+    db_url = f"sqlite:///{db_path}"
+    alembic_cfg = _make_app_alembic_config(db_url)
+
+    command.upgrade(alembic_cfg, "head")
+
+    engine = create_engine(db_url, connect_args={"check_same_thread": False})
+    inspector = inspect(engine)
+    assert "auth_recovery_code" in inspector.get_table_names()
+    auth_user_cols = {col["name"] for col in inspector.get_columns("auth_users")}
+    assert "totp_secret" in auth_user_cols
+    assert "totp_enabled" in auth_user_cols
+    engine.dispose()
+
+    # Downgrade one step (removes TOTP migration).
+    command.downgrade(alembic_cfg, "-1")
+
+    engine = create_engine(db_url, connect_args={"check_same_thread": False})
+    inspector = inspect(engine)
+    assert "auth_recovery_code" not in inspector.get_table_names(), (
+        "auth_recovery_code table should be gone after downgrade"
+    )
+    auth_user_cols_after = {col["name"] for col in inspector.get_columns("auth_users")}
+    assert "totp_secret" not in auth_user_cols_after, (
+        "totp_secret column should be removed after downgrade"
+    )
+    assert "totp_enabled" not in auth_user_cols_after, (
+        "totp_enabled column should be removed after downgrade"
+    )
+    engine.dispose()