verification-before-completion

You can enforce verification before completion by requiring fresh evidence from executable commands like tests, lint, or builds before stating work is complete. This prevents incorrect claims by blocking rationalizations like 'should pass' and ensures only verified results are reported.

Evidence-first completion gating enforces that any 'done' or 'passing' statement is backed by a fresh run of the relevant verification command in the current message. It confirms exit status and failure counts before allowing any completion or satisfaction claims to be made.

Verify pull request readiness by identifying the correct verification command, executing it fully to capture fresh output, and confirming the exit status. Only report passing or fixed status after this complete regression cycle or requirements checklist yields successful evidence.

Agent reports of a fixed bug fail when CI checks run because of skipped steps or trusting partial verification. Bypassing anti-rationalization guardrails leads to claiming success based on confidence rather than executing the full verification command to confirm actual outcomes.

Yes, verification before completion works without external CI dependencies by executing the full relevant verification command locally on the current state. It relies on capturing fresh output and confirming exit status directly in the workflow rather than depending on external pipeline integrations.

You should not use automated verification gating when a task lacks executable verification steps like tests, lint, or builds. If correctness cannot be proven via command line output or requirements checklists, this evidence-first approach cannot generate the required fresh proof.