Summary

• Add 2 HTTP-layer tests to packages/memtomem/tests/test_context_settings.py
  pinning the FastAPI route contract for the Settings Hooks sync surface
  (Phase D), satisfying ADR-0001 §5 c2 (round-trip) and §5 c4 (conflict
  path).

Why

PR-2 of RFC #761 — gap-fill against the §5 readiness contract that
landed in #769. The existing test_context_settings.py has 21 tests
covering merge / mtime / atomic-write at the Python helper layer, but
zero HTTP-layer tests against /api/context/settings/*. Production
UI depends on those routes, so a future refactor of settings_sync.py
could silently regress in-sync detection or the soft-abort response
shape without these tests catching it.

Tests added

test_sync_route_round_trip_unidirectional (§5 c2, unidirectional shape)

1. Write canonical .memtomem/settings.json with one
   (PostToolUse, "Write|Edit") hook.
2. Pre-populate target ~/.claude/settings.json with a user-authored
   (PreToolUse, "Bash") hook + non-hook keys (permissions, env).
3. POST /api/context/settings/sync {"allow_host_writes": true} →
   assert HTTP 200 + claude_settings.status == "ok".
4. GET /api/context/settings → assert status == "in_sync",
   canonical hook surfaces under hooks.synced.
5. Merge invariant: re-read target file from disk; assert user's
   (PreToolUse, "Bash") hook + permissions + env keys all
   survive verbatim (additive merge does not clobber).

Settings has no reverse-import API by design (additive merge cannot
distinguish canonical-authored from user-authored entries), so the
round-trip is shaped per §5 c2's unidirectional clause: write
canonical → apply via sync route → re-read via diff route + assert
target-side merge invariants.

test_resolve_route_returns_soft_abort_on_stale_mtime (§5 c4, soft-abort)

1. Setup canonical + target with conflicting (PostToolUse, "Write")
   rules.
2. Monkeypatch _safe_load_json so reading the target as a side
   effect bumps its mtime — simulates a concurrent writer landing
   between the route's mtime capture (line 239) and recheck
   (line 262).
3. POST /api/context/settings/resolve → assert HTTP 200 + body
   {"status": "aborted", "reason": <... modified by another process ...>}.

Mirrors the helper-level pattern from
TestClaudeSettingsMergeConcurrent.test_aborts_on_mtime_change
(line 243), but pins the route response contract — the helper test
covers the merge function's behavior, this test covers the FastAPI
surface that wires it up.

Scope notes

• No i18n JSON changes. Inventory confirmed existing
  settings.hooks.* keys (10 keys, en + ko both, lines 338–347 in
  web/static/locales/*.json) already cover the Settings overview
  card and section. PR-3's gate flip will use them as-is. The
  original RFC-761 plan's "add settings.ctx.settings_title" step
  was based on a wrong assumption — verified by re-reading
  context-gateway.js:88-95 (the card label uses
  t('settings.hooks.title'), not a separate ctx_settings.* key).
  §5 c3 (i18n parity, auto-discovered by test_i18n.py) is already
  satisfied.
• Tests use mode="dev" (currently required — settings_sync lives
  in _DEV_ONLY_ROUTERS). They remain correct after PR-3 moves the
  router to _PROD_ROUTERS (the router is mounted in either mode).

Out of scope

• PR-3 (the flip): router move + 2× STATE.uiMode === 'dev'
  gate removal in context-gateway.js + CHANGELOG entry. Ships
  after this PR lands.
• Phases B / C 409 conflict-path backfill: ADR-0001 §5 is
  prospective; gaps in already-shipped phases are tracked as hygiene
  follow-ups, not regressions.

Test plan

• uv run pytest packages/memtomem/tests/test_context_settings.py -v —
  37 tests pass (21 existing + 2 new + 14 unaffected).
• uv run pytest packages/memtomem/tests/test_i18n.py -v —
  16 tests pass; no JSON drift.
• uv run ruff check packages/memtomem/tests/test_context_settings.py —
  clean.
• uv run ruff format --check packages/memtomem/tests/test_context_settings.py —
  clean.

Refs: #761
Follows: #769

🤖 Generated with Claude Code