What problem does it solve? AI coding agents frequently claim work is complete without actually running the commands, paste fabricated test output, or mark checklists done in bulk. This Skill blocks that behavior by requiring every completion claim to be backed by raw, session-captured command output. ## Core Features & Use Cases - Evidence Gating: Rejects any Definition-of-Done item, report.md evidence section, or status transition that lacks at least 10 lines of real terminal output including the command and exit code. - Session-Bound Proof Rules: Treats reused, stale, or narrative-only evidence ("all tests pass", "verified") as fabrication, and forbids batch-checking multiple items without per-item execution. - Mechanical Enforcement Pointers: Routes to governance modules and guard scripts (state-transition-guard.sh, artifact-lint.sh, done-spec-audit.sh) that auto-detect fabrication before pushes. - Use Case: Before an agent marks a bug fix's regression test item [x] and transitions a scope to Done, this Skill forces it to actually run the test suite in the current session and paste the literal output as evidence. ## Quick Start Before marking any task complete, ask the agent to run the verification command now and paste the actual terminal output as evidence.
