What problem does it solve? AI coding agents often claim work is done without proof, pasting fabricated or summarized test output. This Skill enforces a strict evidence standard so every claim in a Bubbles report.md or Definition-of-Done item is backed by real, captured command output that survives audits and lint checks. ## Core Features & Use Cases - Standardized evidence blocks: Records the exact command, exit code, and at least 10 lines of verbatim terminal output for every cited result. - Compact verifiable form for large output: Routes output over 40 lines through evidence-capture.sh, which emits a sha256 hash, line count, and a re-runnable --verify check. - Category-specific execution rules: Maps DoD claims (unit tests, API endpoints, database state, UI, builds, lint, coverage) to required live-execution actions and forbidden substitutes. - Use Case: When finishing a bug fix, you run the test suite through the capture script and paste the hashed evidence block into report.md, so the state-transition guard and artifact lint accept the DoD checkbox instead of rejecting it as fabrication. ## Quick Start Ask the agent to run the project's test command through the evidence-capture script and record the resulting evidence block in the report.md Test Evidence section.
