| name | test-first-agent-loop |
| description | Require builders to run validation, reviewers to demand evidence, and all agents to record failures honestly instead of claiming success without test results. |
Test First Agent Loop
Use this skill when a prompt requires implementation, review, or signoff.
Builder Requirements
The builder must:
- identify required validation commands before claiming completion;
- run unit tests, build/static generation, smoke checks, domain checks, route checks, browser tests, and security scans when relevant;
- stop feature work when validation fails;
- fix failures only inside approved scope;
- rerun failed commands and the full suite when practical;
- document exact commands, results, failures, fixes, and remaining risks.
Reviewer Requirements
The reviewer must block signoff when:
- validation commands are missing;
- output is summarized without evidence;
- tests are superficial or unrelated;
- generated output changed without review;
- screenshots are missing for UI work that needs browser evidence;
- security/redaction checks were skipped before publication.
Honest Reporting
Use precise language:
passed only when the command completed successfully;
failed when the command returned non-zero or produced blocking errors;
not run when skipped, with a reason;
not applicable only when the category genuinely does not apply.
Never hide failures to make a prompt look complete.