Commit Graph
6 Commits
Author SHA1 Message Date
Hermes 80a82c4cae refactor(persistence): canonical atomic file writer + notifications.json crash-safety (DC-099) [glm-grade=B]
CI / Test & Lint (push) Canceled after 0s
CI / Security audit (push) Canceled after 0s
- src/utils/atomic-write.js: single shared tmp+fsync+rename writer
  (exclusive-create 0600, unique tmp names, cleanup-on-failure,
  best-effort parent-dir fsync after rename for swap durability)
- notification-manager: both write paths (load-time canonicalization
  write-back + saveConfig) converted from plain writeFileSync — a
  crash mid-write can no longer truncate notifications.json
- DC-097/098 test seams migrated to the atomic path; new suite pins
  syscall discipline (order, wx flags, tmp naming, error cleanup,
  dir-fsync swallow)
- 121 suites / 2768 tests green
- Judge: GLM-5.3 cold read grade B/ship (deleg_23f7abad); polish items
  folded: dir-fsync added, header copy-count corrected. Remaining:
  migrate invite/user/share-store private _atomicWriteJSON copies as
  they are touched (queued).

URN: pending (recorded post-commit)
2026-08-22 23:28:36 -07:00
Hermes 3ccd00d1a1 fix(notifications): persist canonicalized config on load — legacy keys no longer stale on disk (DC-097) [glm-grade=B]
CI / Test & Lint (push) Canceled after 0s
CI / Security audit (push) Canceled after 0s
_loadConfig canonicalized legacy spellings in memory only (DC-092); the
on-disk notifications.json kept email user/pass, camelCase event keys and
string secure until the next explicit UI save — i.e. forever on installs
that never open the settings page.

- _persistCanonicalForm(): after the defaults merge, re-serialize and write
  back only when the bytes differ; idempotent on subsequent loads.
- Best-effort: write failures (read-only mount, EACCES) warn and continue —
  the in-memory config is already correct; constructor never throws.
- No secrets in new log lines; JSON.stringify(this.config) same as saveConfig.

Judge notes (non-blocking, GLM-5.3 cold read): unknown top-level keys are
now dropped from disk at boot (pre-existing merge-drop semantics, previously
deferred to next UI save); write is non-atomic, matching saveConfig.
Verdict: urn:ump:rchawiu427idev5mrettlxkyw2rcnwqpegiolqmzbc5ygc277u5q

Tests: +5 (__tests__/notification-config-writeback-dc097.test.js) —
canonical rewrite, idempotence, clean-file-untouched, EACCES no-throw,
fresh-install no-write. Full suite 119 suites / 2740 tests green.
2026-08-22 21:52:19 -07:00
Hermes 97672f7e74 [glm-grade=A] fix(notifications): un-gate 7 dead emitters + repair legacy 4-arg send shape (DC-094)
CI / Test & Lint (push) Canceled after 0s
CI / Security audit (push) Canceled after 0s
Two silent-death defects in one class:

1. Seven emitters absent from DEFAULT events, so the send() gate
   (config.events[canonical] !== true) dropped them on every install:
   ssl-cert-expiry, dns-propagation, drift-detected,
   dependency-restart-complete/-failed, recipeRemoved, workflow.
   All now default ON; dependency-restart spellings fold onto one
   canonical toggle; recipeRemoved aliases to recipe-removed in both
   the manager and route alias maps.

2. Nine call sites used a legacy 4-arg send(event, title, message, type)
   against the 3-arg signature: the message string landed in the type
   slot (Discord embed color fell back to info-blue, history.type wrong)
   and providers received the TITLE as the body - deploy-failure
   notifications carried no error text at all. Fixed at source (9 sites)
   plus a type-guarded shim in send() for external legacy callers.

Also: explicit data.title now flows to ntfy Title header, email subject,
Discord embed title, and history; settings UI gains 9 event toggles
(separate Backup Complete/Failed) with defaults-on semantics.

Tests: +24 (new DC-094 suite: defaults, gate pass-through, alias folding
send-time and load-time, stored-config inheritance, shim body/title/
color/subject/history, type-guard, 3-arg no-regression); 4 assertions in
bundled-workflows-health-check updated from the old 4-arg mock contract
to the canonical shape (same behavior asserted). Full suite 116 suites /
2706 tests green.

Judge: GLM-5.3 cold read via delegate_task (deleg_8a7cedd0), grade A,
zero blockers; 2 polish items (Backups toggle conflation, shim
type-guard) folded into this commit. Verdict URN:
urn:ump:uyipjjwdjqjy3alvceqxvlucrymd7hnsben5udpp2bqycoh3l2la
2026-08-22 19:42:25 -07:00
Hermes 7e4ee60dcf [glm-grade=B] fix(notifications): repair UI/API contract drift — SMTP auth, event gate, test button (DC-092)
CI / Test & Lint (push) Canceled after 0s
CI / Security audit (push) Canceled after 0s
Three user-facing notification features were silently dead from contract
drift between the settings UI and the backend:

1. SMTP auth never applied: the UI sent email.user/email.pass while the
   manager read username/password. Route now normalizes aliases; legacy
   config files canonicalize at load.
2. Event toggles were cosmetic: UI sent camelCase keys (containerDown),
   the send() gate read kebab-case ('container-down'). EVENT_ALIASES now
   folds every known spelling (manager gate, route store, legacy files);
   UI sends and reads canonical keys.
3. Deploy/auto-restart notifications always dropped: deploy-success/
   deploy-failed/auto-restart were missing from DEFAULT events, and
   send('test') was itself gated -> the Test button was a no-op. Defaults
   added; 'test' bypasses the gate.

Also: strict boolean contract (string 'false' for secure/enabled rejected
— previously coerced truthy, silently forcing TLS), SMTP port bounds,
non-destructive credential merge (blank password no longer clobbers the
stored one), GET /config returns port/secure/to/username + hasPassword
(password never returned), full form prefill + keep-hint placeholder.

Verified live pre-fix: send('test')/'deploymentSuccess'/'auto-restart'
all returned 'not enabled'. Post-fix: +20 tests (route + manager),
full suite 2641/2641 (112 suites).

Judge: GLM-5.3 cold read via delegate_task (deleg_ad765d63), grade B,
ship, zero blockers; judge independently ran the DC-092 suites (38/38).
Polish #1 (canonical event for provider titles) folded in. Remaining
emitters with the same gate-miss class (recipeRemoved, workflow,
ssl-cert-expiry, dns-propagation, drift-detected, dependency-restart-*)
noted for a follow-up tick.
2026-08-22 18:17:14 -07:00
Krystie e99413150e [glm-grade=B] fix: dead dashboard WS, corrupted error.log, auth-polling storm, re-auth freeze + CVE bumps
CI / Test & Lint (push) Canceled after 0s
CI / Security audit (push) Canceled after 0s
Adversarial audit 2026-08-16 (GLM-5.3 delegate, 2 rounds, 141 tool calls):

P0-1: Dashboard WebSocket (/api/v1/ws) dead on EVERY boot since DC-076.
server.js passed module exports (DependencyManager class, {AutoRestartManager}
namespace, SSLMonitor class) instead of createApp()'s live instances — first
.on() threw ERR_INVALID_ARG_TYPE, catch swallowed it. Fix: app.locals.ctx
exposed in src/app.js; server.js passes all 8 real EventEmitter instances.

P0-2: error.log corrupted since 2026-07-14. errorMiddleware called
logError(FILE, SIZE, path, err, meta) — 5 args into a 3-arg wrapper —
logging 'Error: 5242880' garbage every ~60s and DISCARDING the real error
object. Fix: correct 3-arg call + legacy-shape guard in logErrorWrapper +
~74 log.error sites swept to pass real error objects (AST-verified scope-
safe 71/71, 29/29 modules load clean).

P0-3: auth-polling storm (stranded grade=B commit never landed in prod):
401/403 behind TOTP gate hammered /api/v1/services/status + SSE reconnect
every 2-8s, with misleading direct-probe fallback marking services 'up'.
Fix landed + B-round MEDIUM follow-up: TOTP re-auth success now clears
_dcAuthLost, resumes SSE (new _sseResume clears the latch), and refreshes.

Also: eslintignore static-sites/ (33→0 errors); nodemailer 8→9.0.5 and
sharp 0.33→0.35.3 (3 high CVEs killed; jest green on new majors);
dockerode@5/uuid deferred (semver-major, Docker API surface).

Verification: 80/80 suites, 1837/1837 tests; ESLint 0 errors/743 warnings;
node --check all changed files; bundles rebuilt + SW cache bumped.

Judges: Codex quota-dead until Aug 19 (verified live) — GLM adversarial
delegate per operator directive 2026-08-07. Round 1: 98-call mechanical
verification (timed out pre-verdict). Round 2 (this grade): B, one MEDIUM
(re-auth freeze) — fixed in this commit as prescribed.
2026-08-16 04:18:07 -07:00
Hermes 7bc2a207f3 DC-005: Fix all 138 broken test paths after src/ refactor
After the DC-005 module reorganization (41 files moved into src/ subdirs),
138 test suites failed because the refactor script's path-rewrite logic
missed three categories:

1. Files inside src/ doing 'require("./src/...")' — should be 'require("../...")'
2. Files in src/X/Y/ doing 'require("../../../src/...")' — should be 'require("../../...")'
3. Test files in __tests__/ with leftover 'require("../../../src/...")' paths

Root cause: the original refactor script ran before all files were moved,
so it computed relative paths against stale filesystem state.

Result:
- 30/30 test suites pass
- 879/879 tests pass (was: 18/30 suites, 614/687 tests)

Also fixed:
- routes/apps/restore.js: wrong responses import path
- routes/*/*.js: '../../src/utilities/X' → '../src/utilities/X' (depth 2 routes)
2026-06-13 12:16:56 -07:00