agentleFS
Sign inStart free
fainirfainir/agent-harness/skills/qa-flows/SKILL.md

qa-flows

Skill0 starsChanged 14 days agoMITView on GitHub

Test every user flow like a skeptical real user on the real surface - fresh state, happy and unhappy paths, mobile and desktop widths, console and network errors, bots by sending real test messages - saving screenshots and notes as evidence under qa/ and setting each READINESS row to verified or filing a defect row. Use after a flow is built, before claiming anything works, before deploy, and as the final regression pass. Passing unit tests never count as QA.

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 3 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    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:

    Start from qa/README.md "How to QA this project" (start command, test data, selectors, tricks).

    line 12
  2. Very specific:

    Keep qa/ small: overwrite the previous screenshot of the same step.

    line 30
  3. Very specific:

    Before deploy and at the end of the goal: every verified flow re-run on the production URL in one pass, newest first.

    line 34
  4. Distinctive:

    Never marks a row verified from code reading, unit tests or a previous session's claim.

    line 38
  5. Distinctive:

    Never tests only the happy path.

    line 39
  6. Distinctive:

    Never writes "should work" or "looks fine".

    line 40

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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

QA flows

Assume it is broken until you have watched it work. You test as a user, not as the author.

Per flow row (F<n> in READINESS.md)

Start from qa/README.md "How to QA this project" (start command, test data, selectors, tricks); before you finish, add what you had to work out.

  1. Fresh state: new user / cleared storage / new test account (qa+<flow>@example.com style on the product's own sign-up; never a real third-party account).
  2. Drive the real UI at 375px and desktop (or the simulator, or the real bot channel). Click what a user clicks, type what a user types. No API shortcuts for the verdict.
  3. Unhappy paths: wrong input, empty input, double submit, back button, refresh mid-flow, slow network, expired session, no permission. Each one either behaves well or becomes a defect row.
  4. Watch the plumbing: browser console has no errors; network has no 4xx/5xx except intended; no secrets or PII in URLs or logs.
  5. Evidence: qa/<F-id>/<yyyy-mm-dd>-<width>.png (one per key step) and qa/<F-id>/notes.md (steps, result, date, build or commit). Link notes.md from the row.
  6. Verdict: row verified with the evidence link and the date, or todo plus a defect row | D<n> | <flow>: <what went wrong, user words> | <surface> | todo | qa/<F-id>/... | |. Fix defects in build order; re-test the flow after the fix, not just the defect.

Bots and integrations

  • WhatsApp / Telegram / Slack / email: send a real message from a test account or sandbox number, read the real reply, screenshot the conversation. A webhook log line is not evidence that the user got a reply.
  • Payments: provider test mode with its published test cards; verify receipt email, refund, and the failed-card path.

Tools

  • Claude Code: built-in browser (mcp__Claude_Browser__*; resize_window preset mobile/desktop; read_console_messages, read_network_requests), Chrome DevTools MCP, iOS simulator tool, adb for Android. For parallel coverage of many flows use .claude/workflows/production-sweep.js when workflows are enabled.
  • Codex: browser plugin / computer use with the same widths; Playwright if the project has it (npx playwright test) as a supplement, not the verdict.
  • Keep qa/ small: overwrite the previous screenshot of the same step; delete scratch recordings.

Final regression

Before deploy and at the end of the goal: every verified flow re-run on the production URL in one pass, newest first. Any failure reopens the row. The pass is one line per flow in qa/regression-<date>.md.

What QA never does

  • Never marks a row verified from code reading, unit tests or a previous session's claim.
  • Never tests only the happy path.
  • Never writes "should work" or "looks fine"; it writes what it did and what it saw.

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.

Check your setup, free