What problem does it solve? It turns a planned action checklist (actions.md) into real code changes while preserving an auditable trail of what changed in a legacy system and which business rules must keep holding true in future extractions. ## Core Features & Use Cases - Phased task execution: Runs pending [ ] actions phase by phase (Preparação, Testes, Núcleo, Integração, Polimento), respecting dependencies and parallel markers, checking off [X] boxes and appending to an append-only progress.jsonl log. - Legacy impact auditing: Generates legacy-impact.md mapping every touched file to components in architecture.md, classifying impact type and severity, and listing preserved versus modified domain rules. - Regression watch tracking: Generates regression-watch.md with stable watch IDs for modified or removed rules, so future re-extractions can verify nothing silently broke. - Dual anchoring: Works anchored on legacy extraction artifacts (architecture.md + domain.md) or greenfield artifacts (prd.md + SDD specs), aborting cleanly when neither exists. - Use Case: After running /reversa-to-do to produce an actions.md plan for a feature, invoke this skill to implement the tasks, then hand off to /reversa-sync to converge the delivery into the extraction addenda. ## Quick Start Ask the agent to run the reversa-coding skill to execute the pending actions in actions.md, optionally scoping it to a phase or ID range like "only Núcleo" or "T001-T005".
