skill-verification-gate

To prevent premature delivery claims, identify the completion claim, run a fresh verification command, read the output, and only then declare success. This five-step process requires fresh evidence to stop misreporting in development workflows.

Verification evidence in software development is fresh output from executing tests or builds that substantiates a completion claim. It guards against hallucinations and misreporting by requiring proof before declaring bugs fixed or builds succeeding.

To stop hallucinated task completions in multi-provider environments, enforce a verification gate requiring fresh evidence. The IDENTIFY, RUN, READ, VERIFY, ONLY THEN process checks claims against actual command output before any delivery is finalized.

To enforce a verification gate before a pull request, apply the five-step process: IDENTIFY the claim, RUN a fresh command, READ the output, VERIFY the results, and ONLY THEN declare success. This guards PRs and commits against premature claims.

Without fresh verification evidence, claim validation is vulnerable to hallucinations, misreporting, and premature deliveries. Relying on prior assertions instead of actively running verification commands leaves multi-provider workflows exposed to declaring tasks complete without substantiation.