What problem does it solve? AI agents and engineers frequently claim something is fixed, merged, running, or root-caused based on stale memory, proxy signals, or pattern-matching instead of direct observation. This Skill defines a single shared evidence rule that blocks unverified claims in PR bodies, status reports, and debugging explanations. ## Core Features & Use Cases - Evidence gate for claims: Requires same-turn, first-hand observation before stating anything is fixed, working, merged, or running, with an UNVERIFIED: prefix as the mandatory fallback. - Structured investigation sequence: Restate the reported behavior, build an executable repro against the current target, isolate the cause with a controlled comparison, then explain and re-verify. - Anti-pattern catalog: Documents concrete failure modes such as proxy proof, stale state, wrong-target repros, wrong-ref file:line citations, and log-line misattribution. - Use Case: While diagnosing a UI bug, instead of citing a passing test, you open the actual screenshot, run a repro that visibly fails against the current build, isolate the cause with a MutationObserver, and only then explain and fix it. ## Quick Start Apply the prove-it evidence rule to investigate why the dashboard panel fails to render and show me the repro output before explaining the cause.
