Summary

• Sync All now classifies every non-ok Settings status, not
  just needs_confirmation (fix(web/ctx): Sync All silently skips Settings host-write target on default allow_host_writes=False #774 left two siblings on the table).
• New branches: error → error toast carrying the route's reason
  via toast.sync_failed; aborted → warning toast via
  settings.ctx.mtime_conflict. Both reuse i18n keys already used by
  the per-target Sync flow — zero new translations.
• Severity ladder: error > aborted > needs_confirmation >
  all-ok/skipped.

Why

Closes #799. generate_all_settings
(packages/memtomem/src/memtomem/context/settings.py:261) returns one
of five per-result statuses: ok / skipped / error /
needs_confirmation / aborted. PR #798 (#774) only added a branch
for needs_confirmation, so the underlying "resp.ok hides per-result
failure" pathology still applied to error and aborted:

• error — canonical or target file is malformed JSON. Route returns
  {"results": [{"name": "claude_settings", "status": "error", "reason": "/path/to/file is not valid JSON. Fix the file manually..."}]}. Pre-fix(web/ctx): Sync All toasts sync_success for non-ok Settings statuses (error, aborted) #799 JS toasted sync_success regardless.
• aborted — concurrent-write mtime conflict. Route returns
  {"results": [{"status": "aborted", ...}]}. Same lie.

The dedicated per-target Sync flow at
web/static/context-gateway.js:653, 946-955 already surfaces these
as mtime_conflict (warning) / request_failed (error). This PR
brings Sync All in line with the same shape.

Concrete edits

Total diff: 4 files, +92 / −15.

Pin shape

test_issue_799_sync_all_status_coverage asserts:

1. firstWithStatus('error') literal exists — pre-PR shape only had
   'needs_confirmation', so the assertion fails on the old code.
2. firstWithStatus('aborted') literal exists — same.
3. t('toast.sync_failed' exists in the file — already there in the
   catch block, so the assertion holds; the new error branch
   shares the same key, so a regression that drops the new branch
   leaves toast.sync_failed reachable only on transport-layer
   failure (no positive marker remains for the per-result error
   case — which is exactly what firstWithStatus('error') catches).
4. t('settings.ctx.mtime_conflict') exists — pre-PR the only
   reference was the per-target Sync handler, so this assertion
   formerly held; with the new branch it has a second reference
   and the assertion still holds. The firstWithStatus('aborted')
   marker provides the strict positive pin for the new site.
5. Inverted marker: showToast(t('settings.ctx.sync_success'))
   must remain in the file. A regression that collapses the new
   severity ladder into a single error toast (e.g. removes the else
   branch) drops this literal and fails the assertion.

Test plan

• uv run pytest -m "not ollama" packages/memtomem/tests/test_i18n.py packages/memtomem/tests/test_context_settings.py — 57/57 green
  (incl. the new test_issue_799_sync_all_status_coverage).
• uv run ruff check packages/memtomem/src && uv run ruff format --check packages/memtomem/src — clean. Tests file ruff-formatted
  in the same commit.
• Manual smoke: trigger Sync All against a project root whose
  canonical .memtomem/settings.json is malformed JSON (route
  returns status="error"); confirm an error-level toast surfaces
  the reason instead of sync_success.

Out of scope

• The dedicated Settings panel's Sync Now host-write confirmation
  flow (settings-hooks-watchdog.js) — Option 2 from fix(web/ctx): Sync All silently skips Settings host-write target on default allow_host_writes=False #774, larger UX
  cut.
• Per-generator naming in the toast text (e.g. "Sync All complete
  except claude_settings — ..."). The current ladder uses generic
  reasons; a future cut can name the generator if multi-generator
  settings sync becomes more common.

Refs: #774 (parent), PR #798 (needs_confirmation branch), #761 /
#771 (gate-flip that exposed the silent-skip class).

🤖 Generated with Claude Code