What problem does it solve?
In AI-driven development workflows, completed tasks need independent verification before humans sign off. This Skill acts as a read-only adversarial reviewer that fetches a Chorus task, its acceptance criteria, and the originating proposal documents via MCP, then independently checks whether the implementation actually delivers what was promised.
Core Features & Use Cases
- Independent verification: Runs the project's test, build, and lint commands in read-only mode and records command, exit code, and output as evidence.
- Acceptance criteria tracing: Maps each AC item to concrete code and tests, cross-references proposal documents, and flags circular self-tests or silent divergence.
- Structured verdicts: Classifies findings as BLOCKER or NOTE and posts exactly one grep-able verdict comment (VERDICT: PASS, PASS WITH NOTES, or FAIL) via chorus_add_comment.
- Use Case: After a coding agent marks a task complete, spawn this reviewer with the task UUID to get an adversarial second opinion before a human closes the task.
Quick Start
Ask the agent to review Chorus task <task-uuid> with a maximum of 3 review rounds and post its VERDICT comment.
