What problem does it solve? Reviewing a whole project — not just a diff — usually produces shallow, unverified findings that fall apart when someone tries to act on them. This Skill runs a full review campaign: measured ground truth, parallel expert lanes, adversarial falsification of the findings, severity calibration against production reality, a reviewed implementation plan, waved execution in isolated worktrees, and a join review on every merge. ## Core Features & Use Cases - Ground-truth baseline: Runs the project's build, test, and lint commands and verifies every claim at path:line before dispatching any reviewer. - Parallel expert lanes with evidence grades: Derives a lane roster from the repository's own stacks and directories, with a strict finding contract (MEASURED / READ / INFERRED / UNVERIFIED) and explicit permission to refute the brief. - Adversarial verification and severity calibration: An independent pass tries to falsify the record, and findings are checked against the live system so dormant defects are not treated as hotfixes. - Planned, gated implementation: Findings become work packages with gates that have been watched fail, executed in waves with a join review on every merge. - Use Case: Before a major release, ask for a full audit of the repository; the Skill returns a graded findings record, a reviewed plan, and merged fixes with decision records. ## Quick Start Ask the agent to run a complete project-scope code review of this repository and produce a graded findings record with an implementation plan.
