Repository navigation
chore(agents): adopt task-size workflow, CI guards and local review skills - #1
Merged
Merged
Conversation
…kills Adapt the Fintech Lab agent workflow (7f5e1c2) to StackPulse: - AGENTS.md: T0/T1/T2 task sizes, verification commands, project invariants, files/artifacts, skill routing and Git/PR rules; Eastbase overlays removed. CLAUDE.md stays a thin @AGENTS.md entry point. - docs/AGENT_START_HERE.md: reading matrix by task; ai-workflow.md and verification.md updated for provenance, local skills, guards and CI. - Skills: vendor vercel-react-best-practices and frontend-design byte-for-byte with manifest hashes; remove grill-me; add local codex-review, ship-pr and engineering-workflow (+ state helper and check script) with generalized wording. - Guards: review signals (T2/guard paths, node:test focus/skip rules) and append-only drizzle/ migration history, with regression tests. - CI: integrity on Ubuntu/Windows, PR guards, and lint, typecheck and offline tests (no database, providers or build). - .gitattributes (LF), ignore .tmp/ and .claude/worktrees/, PR template, pnpm test. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Scan every changed test file whole, so deleting an env-gate line
cannot leave an unconditional skip; detect { todo } options, quoted
keys, bracket access, shorthand and .skip()/.todo() calls; require
the gate to be `!<env var or identifier> &&`.
- List removed tests and assertions of deleted test files.
- T2: vote route only (not the read-only releases GET), digest
unsubscribe page, evaluate-skill-updates; guard: sync and workflow
integrity scripts.
- CI runs `node --test tests/*.test.mjs` directly so package.json
cannot narrow it unlabeled.
- Escape the LICENSE(.txt) check and test its failure path.
- Ignore .tmp/ and .claude/worktrees/ in ESLint and tsconfig.
- verification.md: build only against an isolated database.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Accept only the complete env gate (`!<var> && '<reason>'` then `,`, `}`, a comment or end of line) and check every skip/todo option on a line; detect `only:` split across lines. - List newly allowed skips found in changed test files' post-image, not only on added lines; CLI regression against a temporary repo. - T2: src/app/(auth)/ (sign-in passes the return path into OAuth). - engineering-workflow-state: re-verification after a correction may re-record verification evidence; reviewer reports stay immutable. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
github-actions
Bot
added
the
guard-change
This branch was successfully deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What and why
Adapts the Fintech Lab agent workflow (
daniel-tsx/fintech@7f5e1c2) to StackPulse:AGENTS.md, plus a reading matrix indocs/AGENT_START_HERE.md.codex-review,ship-prandengineering-workflowskills.drizzle/migrations.Skill changes: the Eastbase overlays are removed from routing,
frontend-designandvercel-react-best-practicesare vendored, andgrill-meis dropped. No application code changes.Review guide
scripts/check-pr-review-signals.mjs. The T2/guard path lists set the labels. The line scanner (env gate,allow-skip, whole-file pass over changed test files) can fail a PR..github/workflows/ci.yml. The job names become the required checks. Note the permissions, and that the app job runsnode --test tests/*.test.mjswith no database.AGENTS.md: the task-size table, invariants, and Git/PR rules. Local commits on task branches are now allowed; this replaces the old "No commits unless the user asks".scripts/engineering-workflow-state.mjs. One fix from the reference:verification.jsonmay be re-recorded during re-verification..agents/skill-sources.json..claude/skills/*, which are generated copies.docs/AGENT_START_HERE.md,docs/operations/*.mdanddocs/README.md..tmp/and.claude/worktrees/.pnpm/action-setupwith pnpm 11, and label sync.initial sync from a CRLF checkout creates full copies, adapts Grill Me and is repeatableto… creates identical full copies and is repeatable. The Grill Me adapter it asserted no longer exists. Sync fixtures now usegrillingandfrontend-design. No skips were added.Task size
CI and guard change, so it needs independent review. No T2 application area is touched and no application code changes.
Checks
Run locally on
c443f6d, in a fresh worktree with no.envand no DB/provider env.node scripts/sync-agent-skills.mjs --checknode scripts/check-engineering-workflow.mjspnpm test:agent-workflownode --test tests/*.test.mjs(same files aspnpm test)pnpm lint,pnpm typechecknext-env.d.ts, as in CInode scripts/check-pr-review-signals.mjs origin/main HEADnode scripts/check-migration-history.mjs origin/main HEADdrizzle/commits: no false positivesci.yml(js-yaml, PyYAML),bash -non everyrun:block, frontmatter of all 22SKILL.mdfilesgit diff --check origin/main HEAD; prettier on authored filespnpm build, SQL/HTTP database cases, live providers.envUnverified boundaries
ghlabel sync with the job token.docs/operations/ai-workflow.md.Review
3f783b5gpt-6.1-sol, high effort, read-only142e8cfgpt-6.1-sol, high effort, read-only (verification)c443f6dFixed with regression tests:
{ todo }option was not detected. Literal and suffixed gates (!false,|| true) and a second option on the same line passed.only:slipped through.package.jsoncould narrow CI's test run unlabeled; CI now owns the glob.src/app/(auth)/,src/app/digest/unsubscribe/andscripts/evaluate-skill-updates.ts. The read-only releases GET was a false positive..tmp/worktree was scanned by lint and tsc.LICENSE(\.txt)regex was unescaped and its failure path untested.verification.jsoncould not be refreshed during re-verification.pnpm build.Declined:
tests/release-votes-db.test.mjsandtests/release-pages-http.test.mjsare not guard paths. Weakening them still raises assertion-loss signals.guard-changelabel, as in the reference. Fixing that means also running the base copy in CI.Open, round 3 (P2, line-scanner limits; correction budget spent):
|| true,on the next line passes.only:andtruehides the focus.test.skip(…)are false positives.integration(...)wrapper; assertion-loss warnings still fire./security-review: not applicable, since no auth, ownership, cron-auth or secret code changed.Docs and temporary files
Docs updated:
AGENTS.md,CLAUDE.md,docs/AGENT_START_HERE.md,docs/README.md,docs/operations/ai-workflow.mdanddocs/operations/verification.md. No new docs. The ADR convention uses the existingdocs/architecture/decisions/.Temporary files, kept until merge in the task worktree's
.tmp/agent-workflow-v2/:review-prompt.md,review-2.md,review-3.md,pr.md. Round 1's report is summarized above.ADRs consulted: none applies.
🤖 Generated with Claude Code