/test

name: test
description: "Proves a code change works: runs every repository check until green, then proves each acceptance criterion with tests, and by running the real app for anything a user can see or call. Use to test or verify a diff, branch, PR, URL, or user flow."
user-invocable: true
argument-hint: "<task, acceptance criteria, diff, branch, PR, URL, or flow>"
disable-model-invocation: true

Test

Show that each acceptance criterion passes. Reading the code is not proof.

Steps

1. Run every check the repository has: format, lint, typecheck, and tests. The commands are in `AGENTS.md`, `CLAUDE.md`, the README, or the project's scripts. Fix what the change broke and run them all again. After three rounds on the same failure, stop and report it. 2. List each acceptance criterion (`AC-n`, `J-n`, `REQ-n`, `INV-n`) from the ticket, word for word. 3. Point each criterion to a test that proves it. Where none exists, write one that calls the public interface. Revert the code once to confirm the main new test fails. 4. If a user can see or call the change, run the real app the way they would. Use the repository's own end-to-end tests or scripts if it has them. Otherwise use a real browser for a web page, a terminal for a command-line tool, and HTTP requests for an API. 5. For a web page, check desktop and mobile widths, keyboard use, and console errors. 6. Save the evidence for the PR: screenshots, terminal output, or responses.

Return

Failures and unverified criteria first:

Criterion: <ID or exact wording>
Result: Pass / Fail / Unverified
Evidence: <command, user flow, screenshot, or captured output>

Then each check command you ran and its result.

Fix the code, not the test. Report failures the change did not cause; leave them alone.