What problem does it solve? AI assistants often report tasks as complete based on their own summary rather than actual verification, leading to false claims about file edits, calculations, data pulls, and integrations. This Skill enforces a verify-before-claiming discipline so every "done" is backed by a real check, not a claim. ## Core Features & Use Cases - Evidence-Based Verification: Runs concrete checks (grep for old/new values, recompute totals, re-count rows, re-open generated files) against the authoritative source before any completion claim. - False-Green Detection: Catches six categories of misleading passes, including wrong replacements, self-signed integration tests, proxy-context runs, and source-reads mistaken for behavior proof. - Self-Healing Loop: Auto-fixes caught failures, re-runs the same check, and escalates to a human with full context after two failed attempts instead of looping silently. - Use Case: After batch-updating a version string across five files, the Skill greps all files for the old value (expecting zero hits) and the new value (expecting one canonical form), then reports the proof before saying done. ## Quick Start Before telling me the task is finished, run the prove-it verification and show me the actual check results.
