Quality Gate

Verifies that a deliverable genuinely satisfies its criteria, using real evidence rather than assertions.

When to use

• Before declaring a task complete.
• At the end of a workflow phase, before proceeding to the next.
• Before publishing, sending, or releasing anything.

When not to use

• On intermediate scratch drafts not yet consumed by anyone.
• As an infinite review loop: check once, fix, then re-check only what changed.

Procedure

1. Retrieve the criteria (from acceptance-criteria or the request).
2. For each criterion gather evidence: test or command output, generated file, described manual check, primary source.
3. Run deterministic checks (tests, linting, validations, counts) instead of estimating the result.
4. Assign an outcome:
   • pass: all criteria have positive evidence;
   • fail: at least one criterion is not satisfied (specify what to fix);
   • blocked: something is missing to perform verification (access, data, user decision).
5. Explicitly list what you could not verify: never present unverified items as verified.

Expected output

Outcome: pass | fail | blocked
1. <criterion> — ok | no | unverified — evidence: <test/command/file>
…
To fix: …
Unverified: … (reason)

Attention

• Never mark "pass" without evidence. An agent claiming to be finished is not evidence.
• For security, releases, and external actions, this gate is mandatory and never skipped to save tokens.