What problem does it solve? When triaging user feedback about agent behavior, you need to know whether a reported bug actually reproduces or whether a fix produces the intended result. This Skill automates that judgment by comparing the current state of a feedback-case directory (research.json, tree.gedcomx.json, results/) against the reporter's prose description of what went wrong or what should have happened. ## Core Features & Use Cases - Bug Reproduction Verification: Run with --against=what-went-wrong to confirm the reported bug reproduces against the agent_did field in _feedback/feedback.json. - Fix Validation: Run with --against=desired to verify a fix matches the agent_should_have description. - Structured Verdicts: Outputs a matches / partial / does-not-match verdict with a bullet list of concrete discrepancies mapped to the prose. - Fixture Validation: Aborts early with a specific message when feedback.json is missing, unparseable, or has a wrong schema_version, and handles legitimately empty fields without false aborts. - Use Case: After unzipping a feedback case with scripts/setup-feedback-case.sh and letting the agent run, invoke /compare-state --against=what-went-wrong to decide whether to iterate on a fix or escalate as not reproducible. ## Quick Start Run /compare-state --against=what-went-wrong inside an unzipped feedback-case directory to check whether the reported bug reproduces.
