- dev-bypass path: DB lookup for users.isAdmin (T-10-05 bypass skips OIDC not DB)
- OIDC path: same resolveAdminAndSetupStatus helper after upsertUser
- needsProviderSetup: true when no member_credentials row, false when one exists
- no /api/me/credential POST added here (Plan 03)
- dev-bypass path: isAdmin from DB (not hardcoded), needsProviderSetup from member_credentials
- needsProviderSetup=true when no member_credentials row exists
- needsProviderSetup=false when member_credentials row exists
- zero-admin COUNT check before INSERT: first user gets is_admin=true
- subsequent users (admin already exists) get is_admin=false
- existing-user early-return path unchanged (is_admin not modified)
- Phase-12 hook comment: tighten to first login after app_config.setup_complete
- adds 'import { sql }' from drizzle-orm
- first user with zero admins → is_admin=true in INSERT values
- subsequent user with admin present → is_admin=false in INSERT values
- existing user re-upsert → is_admin unchanged (early-return path, no insert)
- update existing color tests to accommodate new 4-select flow order
- reads users.isAdmin from DB (never trusts context user's isAdmin claim)
- 403 with { error: 'Forbidden' } for non-admins and missing user
- side-effect import of devBypass.js for ContextVariableMap augmentation
- bypass path skips OIDC only, not the DB check (T-10-04/T-10-05)
- 403 for non-admin user (is_admin=false in DB)
- next() called for admin user (is_admin=true in DB)
- 403 when no user on context (no DB query)
- 403 when context user spoofs isAdmin=true but DB has is_admin=false (T-10-04)
- INSERT INTO users (id=1, is_admin=true) ON DUPLICATE KEY UPDATE is_admin=true (idempotent)
- Supplies placeholder non-null oidc_iss='dev-bypass', oidc_sub='dev-user-1', color='#4A90D9'
- requireAdmin (Plan 02) does a DB lookup for the bypass user; without this seed it would 403
- Existing calendar/event/list seeds unchanged (INSERT IGNORE INTO calendars, Seeded Test Event)
- Generated via drizzle-kit generate from updated schema.ts
- SQL is additive-only: CREATE TABLE app_config + ALTER TABLE ADD COLUMN (3x) + ADD CONSTRAINT UNIQUE
- Applied to live dev MariaDB via direct SQL execution (drizzle-kit migrate journal hash mismatch
with legacy migration tracking; DDL applied + hash recorded in __drizzle_migrations manually)
- Verified: MIGRATION OK via live DB SHOW COLUMNS / SHOW TABLES query
- No DROP/TRUNCATE statements in generated SQL (grep returns 0)
dorny/paths-filter combines patterns with Array.some, and picomatch compiles
'!.gitea/**' as 'any path not under .gitea'. The '- !.gitea/**' line (from
quick task 260613-dmw) therefore matched every non-.gitea file — including
.planning/** and *.md — so code=true for doc-only PRs and the heavy api/harness
jobs ran anyway. It also never excluded .gitea (the **/*.yml glob already
matched workflow files), regressing the Phase 15 doc-only skip.
Switch to predicate-quantifier 'every' with negation-only globs: a file counts
as code only if it is outside .gitea/, outside .planning/, and not Markdown.
Verified with picomatch against representative file sets.
Add paths-ignore ['.gitea/**', '.planning/**'] to publish.yml's push
trigger. Doc/CI-only pushes to main (planning docs push straight to main;
workflow edits merge via PR) produce a byte-identical image (.dockerignore
already excludes .planning), so the build + :latest re-push was wasted.
paths-ignore skips only when every changed file matches; mixed code+docs
pushes still publish.
Workflow-only edits shouldn't pay for the multi-minute MariaDB/integration/harness
suite. A negation '- !.gitea/**' (ordered after the yml/yaml globs) makes CI-config-
only PRs resolve code=false so api/harness skip like docs. fast-checks (format:check
still validates the workflow YAML) + gate continue to gate them. PRs that also touch
app code or lockfile still resolve code=true and run the full suite.
The AUDIT-ADVISORY tier cross-checks pnpm outdated (direct deps) against
advisory module_names (mostly transitive), so it almost never fires.
Relabel to OUTDATED-WITH-ADVISORY and document the limitation in the
header, header text, and inline comment; point to check-audit.mjs as the
authoritative advisory gate. No risky full-tree rewrite.
WR-02: capture docker run exit directly (not the piped head exit) so a
chatty booting image can't SIGPIPE to 141 and false-PASS; require the
FATAL guard marker in output as a positive assertion.
WR-05: strip comment lines and use anchored fixed-string (grep -qF)
matching so a commented-out rule can't satisfy the hygiene check and
patterns aren't treated as regexes.
Bind github.base_ref/base.sha/head.sha through env: and reference quoted
shell vars (no inline ${{ }} in run:) to close the script-injection vector.
Add a HEAD_SHA rev-parse fallback mirroring BASE_SHA and echo the final
${BASE_SHA}..${HEAD_SHA} range before gitleaks.
Add isWaived() predicate: a waiver with a past 'expires' date is treated
as absent so the High/Critical advisory re-blocks. Applied in both
selectBlocking and partitionAdvisories. Add expired-waiver unit tests.
isMain now compares fully-resolved real paths (mirrors index.ts).
- Static assertions step: verifies .dockerignore exists, covers all forbidden
patterns (.env, node_modules, apps/api/scripts, .git, .planning, apps/api/tests,
apps/pwa/e2e), and --target production is still pinned in publish.yml (T-16-19)
- Boot-smoke step: runs freshly-built sha_tag image with NODE_ENV=production +
DEV_AUTH_BYPASS=true; fails if image starts (EXIT==0) or times out (EXIT==124),
passes on any other non-zero exit — proves D-08 guard fires in shipped image (T-16-18)
- Both steps ordered after build and before push; a hygiene failure blocks publish (T-16-20)
- Separate 'Build and push' into 'Build production image' (build only) and 'Push image' steps
- Creates insertion point between build and push for hygiene assertions (Task 2)
- Preserves --target production, Dockerfile path, both tags, and root context
- WR-04 immutable-tag-first push ordering preserved in push step
- gate needs: expanded to include security
- individual needs.security.result check added (requires success, not success-or-skipped)
- security not folded into api/harness for loop — gitleaks always runs so it cannot be skipped
- follows Gitea #31007 individual needs.X.result pattern
- 613 commits scanned, ~23 MB of git history
- Zero findings after allowlisting crypto.test.ts AES test fixture
- Human-verified: single baseline finding (TEST_KEY) was a Vitest fixture, not a real credential
- Baseline is empty JSON array [] — PR-diff scans in 16-05 start from clean known state
- Add 4th [[allowlists]] block for apps/api/tests/broker/crypto.test.ts
- Human-verified: TEST_KEY in Vitest beforeAll is a synthetic AES-256-GCM fixture, not a real credential
- Original 3 allowlists (VAPID fixture, .env.example, .env.spike) intact
- Disable detect-object-injection globally in eslint.config.js: all hits were
numeric loop array indices (ranks[i]) — not user-controlled keys; zod guards
real API input boundaries; justification comment added (T-16-09)
- Add inline eslint-disable for detect-non-literal-fs-filename at 2 sites:
- apps/api/src/index.ts: realpathSync(process.argv[1]) — runtime entry path, not user input
- apps/api/tests/broker/expand.test.ts: readFileSync of test fixture path — test-controlled
- pnpm lint exits 0 across both apps with --max-warnings 0
- 14 of 15 security rules remain active at error; no blanket file disables
- Install eslint-plugin-security@3.0.1 to root devDependencies (pinned exact version; 3.0.1 chosen over 4.0.1 for bake time per RESEARCH)
- Add pluginSecurity import to eslint.config.js
- Insert security config block (files: apps/**/*.{ts,tsx}) with configs.recommended spread, placed before prettierConfig which stays last
- ESLint version unchanged at 9.39.4 (D-13-ESLint-PIN)
- Classifies outdated packages into four tiers: AUDIT-ADVISORY, MAJOR-BEHIND-INTENTIONAL,
MAJOR-BEHIND-UNPINNED, ROUTINE-DRIFT
- Reads outdated-pins.json for intentional pin reasons (eslint, @eslint/js, zod, @types/node)
- Cross-checks pnpm audit --json to flag pinned versions with active advisories
- Always exits 0 — never gates the build (D-06)
- Exports selectBlocking() and partitionAdvisories() as pure functions for unit testing
- Main body guarded by import.meta.url check (only runs when invoked directly)
- Uses pnpm audit --json (no --audit-level — Pitfall 1 honored)
- Exits 1 on unwaived High/Critical; exits 0 with advisory report on moderate/low
- All 5 unit tests pass (node --test)
- Tests four behavioral cases: unwaived High → blocking, waived High → not blocking,
moderate/low only → not blocking, no advisories → not blocking
- Uses node:test + node:assert (no extra dependencies)
- Fails at RED: check-audit.mjs does not yet exist
- audit-allowlist.json: GHSA-gv7w-rqvm-qjhr waiver (esbuild dev transitive via drizzle-kit/vitest/vite; not in production runtime)
- outdated-pins.json: intentional pin reasons for eslint, @eslint/js, zod, @types/node
- Add ENV NODE_ENV=production in production stage after WORKDIR /app/apps/api
- Engages the devBypass.ts hard guard so DEV_AUTH_BYPASS can never inject
DEV_USER in the shipped image, even if accidentally set (D-07)
- No other stage (base/builder/dev/pwa-builder) is affected
- Create apps/api/src/lib/bootGuards.ts with assertNotDevBypassInProduction()
- Guard exits non-zero when NODE_ENV=production AND DEV_AUTH_BYPASS=true (D-08)
- Wire import + call as first statement in isMainModule() block in index.ts
- 3/3 unit tests pass, typecheck green
- Three test cases: prod+bypass=exit(1), dev+bypass=no-exit, prod+unset=no-exit
- Fails with Cannot find module (src/lib/bootGuards.ts absent) — RED confirmed
- extend outboxWorker import to include initOutboxTrigger alongside startOutboxWorker
- call initOutboxTrigger() immediately after startOutboxWorker() inside isMainModule() guard
- gated by isMainModule() so tests importing app never register the drain listener (Pitfall 4)
- import signalOutboxDrain from '../lib/outboxTrigger.js'
- Site 1 (POST /create): fire-and-forget signal after outbox insert, before 202
- Site 2 (PATCH edit-as-move): signal after await db.transaction() resolves, not inside callback (D-03)
- Site 3 (PATCH same-calendar update): signal after outbox insert, before 202
- Site 4 (DELETE): signal after outbox insert, before 202
- Import scheduleOutboxDrain (not yet exported — causes RED)
- Import signalOutboxDrain from outboxTrigger.ts
- Add describe block 'scheduleOutboxDrain — trigger wiring (D-09)' with 3 tests:
Test A SC-1: signalOutboxDrain() fires drain promptly without timer advance
Test B D-05: two mid-drain signals collapse to exactly one trailing re-drain
Test C D-07: concurrent scheduleOutboxDrain() calls dispatch exactly once via isDraining guard
- 27 pre-existing tests unmodified and passing; 3 new tests failing (RED)
Lucas Berger