agentleFS
strelov1strelov1/freehire/.agents/skills/verify-freehire/SKILL.md

verify-freehire

Skill3.6k starsChanged yesterdayMITView on GitHub

Launch an isolated freehire stack from this checkout with fixture or live ingestion and drive it like a user with Playwright to prove a change works. Use before calling a change done, when checking a pull request is ready to merge, or when asked to run, see or screenshot freehire locally. Works on the laptop (mise + Docker) and in the SuperPlane coder runner.

Adopted by

Steady: 0 new projects took it in the last 4 weeks, none the 4 weeks before.thin data

Week of Sep 20: 0 new projectsWeek of Sep 27: 0 new projectsWeek of Oct 4: 0 new projectsWeek of Oct 11: 0 new projectsWeek of Oct 18: 0 new projectsWeek of Oct 25: 0 new projectsWeek of Nov 1: 0 new projectsWeek of Nov 8: 0 new projects
Sep 2026Oct 2026Nov 2026
New projects taking a version of this file each week, over the 8 weeks our latest run read: only 0 in all, so this is a hint, not a trend.

Who's talking about strelov1/freehire

The biggest 5 of 35 mentions linking to the repository that our crawl found. Each line is their own headline.

See all 35

Talked about on GitHub and Bluesky (35 times).

On GitHub · 34

29 more on GitHub

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 1 is distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Run .agents/skills/verify-freehire/verify.sh preflight to check local tools.

    line 25
  2. Very specific:

    Not running, on purpose: Redis (the rate limiter fails open, the log shows connection refused, harmless), mail (no SES: sign-up works, …

    line 64
  3. Very specific:

    Every line must say ok. It checks: the server was built from the current HEAD and the Go build source snapshot (including dirty/untracked …

    line 78
  4. Distinctive:

    Check side effects next to what is visible: the row in Postgres, the document in search.

    line 131

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

verify-freehire

Proves a change in the real app: the API built from THIS checkout, the web dev server in front of it, postings through the real ingest path, and a real browser. Tests passing is not this; this is the user path.

Everything lives in this folder. Run commands from the repository root. The helpers require Python 3 to fingerprint source files.

HelperWhat it does
.agents/skills/verify-freehire/stack.sh upstart everything, seed, wait until ready, print doctor and the env
.agents/skills/verify-freehire/stack.sh doctorread-only health check of the running instance
.agents/skills/verify-freehire/stack.sh envthe exports a drive needs: eval "$(.agents/skills/verify-freehire/stack.sh env)"
.agents/skills/verify-freehire/run-journey.sh <spec>run one Playwright journey against the instance, evidence included
.agents/skills/verify-freehire/stack.sh downstop what up started, drop its database; evidence stays

Foreground lifecycle (recommended)

Run .agents/skills/verify-freehire/verify.sh preflight to check local tools. Run verify.sh .agents/skills/verify-freehire/journeys/jobs-search.spec.ts for a complete foreground lifecycle: startup, doctor, journey, evidence and teardown. Pass additional spec paths to run them in sequence. Keep this command in one live exec session until it exits; detached Vite can disappear when the launching session ends. EXIT, INT and TERM trigger cleanup, retaining startup output and service logs in a unique evidence directory. SIGKILL cannot run cleanup; use stack.sh down after inspecting diagnostics. An existing state directory is refused rather than taken over.

Launch

.agents/skills/verify-freehire/stack.sh up      # ~1 min warm, longer on a first pnpm/go build

What up does, in order:

  1. Postgres. Uses TESTDB_ADMIN_URL (mise sets it on the laptop; the coder runner's image sets it and starts Postgres itself). On the laptop, if nothing answers, it runs docker compose up -d db. Creates database verify_<api port> and runs cmd/migrate.
  2. Meilisearch, private to this instance on api port + 2: a meilisearch binary on PATH if there is one, else a container verify-freehire-meili-<port> from the image docker-compose.yml pins. Neither: it warns and /jobs will answer 500.
  3. Seed. cmd/reindex creates the empty index. By default a three-posting fixture substitutes the external Greenhouse response and runs the real adapter, pipeline, database store and search enqueue path. cmd/search-drain indexes it. Set VERIFY_SEED_MODE=live to crawl a real board with cmd/add-board and cmd/ingest; VERIFY_SEED_PROVIDER / VERIFY_SEED_BOARD choose it. VERIFY_SEED_BOARD= skips seeding. Fixture writes require a loopback verify_ database; no model or external board request is needed.
  4. API: go build ./cmd/server, started on the api port with JWT_SECRET, FRONTEND_ORIGIN and MEILI_* set. Ready = /health says "database":"up".
  5. Web: pnpm install in design-system/ and web/, then vite dev on api port + 1, proxying /api to the API. Ready = GET /jobs answers 200 (the first SSR compiles).

Ports: the api port is $VERIFY_API_PORT, else mise's per-worktree $PORT, else 18080. State (pids, logs, the server binary) is in /tmp/verify-freehire-<api port>. Logs to read when something is off: server.log, web.log, seed.log, migrate.log there.

Not running, on purpose: Redis (the rate limiter fails open, the log shows connection refused, harmless), mail (no SES: sign-up works, email verification codes never arrive), OAuth providers, payments, LLM features (no LLM_*: AI routes refuse).

Isolation. Two instances can run side by side if their api ports differ (each worktree gets its own $PORT from mise). up refuses to start over a live instance in the same state dir. Do not point a drive at someone else's instance.

Doctor

.agents/skills/verify-freehire/stack.sh doctor

Every line must say ok. It checks: the server was built from the current HEAD and the Go build source snapshot (including dirty/untracked files and embedded assets under cmd/ and internal/, vendor files and module/workspace files; down + up after changing those inputs), the API and web processes are the ones up started, /health sees the database, Meilisearch is healthy, search holds seeded postings, /jobs renders. The journey helper runs it before every journey and refuses an unhealthy or stale stack.

Drive

Playwright from web/node_modules with its pinned Chromium (the runner has it in /opt/ms-playwright; on a laptop without it: cd web && pnpm exec playwright install chromium).

.agents/skills/verify-freehire/run-journey.sh .agents/skills/verify-freehire/journeys/jobs-search.spec.ts

The journeys in journeys/ are the mapped features (see features/README.md). For a change no journey covers, write a spec of your own the same way (any path; the helper copies it into a unique directory under web/e2e/__verify__/ and deletes only that directory after) and run it with the same helper. If it is a real regression test, also add it under web/e2e/ in the PR.

Rules that came from failed runs:

  • Wait for hydration. The page is server-rendered; typing or submitting before Svelte hydrates is lost (a form submits as a native GET). await page.waitForLoadState('networkidle') after every goto that is followed by input.
  • Cold routes are slow. vite dev compiles each route on its first visit (5-15s), so a journey uses expect.configure({ timeout: 30_000 }) and test.setTimeout(120_000), as the shipped ones do. up warms /jobs and one job page.
  • Stable handles: getByTestId('job-card'), roles with names (combobox 'Search jobs', button 'Create your account'), getByLabel('Email'). When a locator misses, read the page snapshot in the failure's error-context.md; it lists every role and name.
  • Some lists open a posting in a new tab: wait for context.waitForEvent('page') only when the link has target="_blank" (the feed does not).
  • API checks go through the web origin (page.request.get('/api/v1/...')) or straight to $VERIFY_API_URL with curl.
  • Database side effects: psql "$DATABASE_URL" after eval "$(stack.sh env)".

Evidence

Each journey prints its unique evidence directory under $VERIFY_EVIDENCE (default /tmp/verify-freehire-evidence-<api port>). down never touches evidence. The journey exports its directory to the spec and stores:

  • <journey>-<n>-<step>.png: screenshots the journey takes at each step (action, then result);
  • <journey>.log: the Playwright run;
  • <journey>-playwright/: the trace (pnpm exec playwright show-trace <trace.zip>) and, on failure, error-context.md with the page snapshot;
  • anything you add: save psql / curl output there too (| tee "$VERIFY_EVIDENCE/x.txt").

Proof standards:

  • Drive the real user path: the UI or the public API. No internal setters, no test-only routes, no SQL writes to fake a state the feature should have produced.
  • Capture the action and the resulting state, not only the last screen.
  • Check side effects next to what is visible: the row in Postgres, the document in search.
  • A journey that fails because the seed lacks something is a failure; write the reason in that feature's Gotchas. Never weaken the journey to make it pass.
  • In a pull request, name what you drove, link or attach the screenshots, and say what you could not drive (mail, payments, LLM) instead of implying it was covered.

Cleanup

.agents/skills/verify-freehire/stack.sh down

Stops the processes up recorded (by pid, never by name), removes its Meilisearch container, drops verify_<port>, deletes the state dir. It leaves the shared compose Postgres running and the evidence folder in place. Run it after every failed attempt too.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.