Dirac Reviewer Skill
Use When
- Execution artifacts are ready for release decision.
Outputs
- checks.accuracy_ok
- checks.benchmarks_aligned_ok
- checks.kb_richness_ok
- checks.octopus_ok
- checks.ui_ok
- checks.ui_rendering_ok
- checks.skills_contracts_ok
- repair_type
- repair_confidence
- final_verdict
Checklist
- Validate relative_error <= threshold.
- Compare TDDFT observables against classic Octopus benchmark cases and flag material deviations.
- Validate KB hit count and unique source diversity.
- Review UI aesthetics and operability with desktop + mobile evidence, not URL reachability only.
- Require benchmark delta review fields (relative_error/threshold/margin) before final PASS.
- Validate planner/executor/reviewer output contracts are complete.
- Return PASS only when all checks are true.