What problem does it solve?
LLM-generated project proposals often look well-structured but contain hidden gaps: missing requirements, vague acceptance criteria, wrong task dependencies, or hallucinated technical details. This Skill performs an adversarial, read-only audit of a Chorus proposal so human reviewers get a concrete, evidence-based verdict instead of rubber-stamping.
Core Features & Use Cases
- Full proposal audit: Fetches the proposal, its comments, the originating idea, and its elaboration via Chorus MCP tools, then checks PRD completeness, specificity, feasibility, and module contracts.
- Task DAG validation: Verifies task granularity, acceptance-criteria quality, requirement-to-task coverage, and dependency correctness (missing or circular edges).
- Structured verdict output: Classifies findings as BLOCKER or NOTE and posts exactly one grep-able verdict comment (
VERDICT: PASS, VERDICT: PASS WITH NOTES, or VERDICT: FAIL) via chorus_add_comment.
- Use Case: A PM agent drafts a proposal for a new feature. Spawn this reviewer with the proposal UUID; it cross-references every requirement against task ACs and posts a FAIL verdict with evidence when it finds an uncovered requirement.
Quick Start
Ask the agent to review proposal <proposal-uuid> with at most 3 review rounds and post its VERDICT comment.
