Orchestrates multi-agent panels to review plans, code, and designs with structured scoring.
npx skills add https://github.com/HACK-WU/skills --skill review-panel-hack-wu
Please help me install this Agent Skill. Skill: review-panel Source: https://github.com/HACK-WU/skills/tree/main/skills/review-panel Command: npx skills add https://github.com/HACK-WU/skills --skill review-panel-hack-wu
.codebuddy/review-panel/{task-name}/ for full transparency and later inspection. - Iterative Improvement Loop: The main agent can revise the plan up to 3 times based on judge feedback, re-running challenger evaluation each round. - Use Case: Ask the agent to run a review panel on a JWT authentication design; it produces the scheme, gathers independent objections, collects scored evaluations, and delivers a final report highlighting consensus objections and adopted suggestions. ## Quick Start Ask the agent to use the review panel to evaluate your design document or code change, for example by saying: use the review panel to review my authentication scheme.High-intent search queries and answers about installing and using this skill.
Explicitly ask the agent to use the review panel, for example by saying 'use the review panel to review this design'. The main agent writes the scheme to disk, launches two isolated challengers and three judges, then delivers a final report with scores and verdicts.
The panel uses two isolated challengers that raise objections with alternatives, three judges that score feasibility, risk coverage, completeness, and adoption, and a main agent that acts as final judge and can iterate the plan up to three times.
It is not triggered automatically. The skill only activates when the user explicitly requests a review panel, using phrases like 'review panel' or 'use the review panel'. Ordinary requests like 'review this' do not enable it.
All artifacts are persisted under .codebuddy/review-panel/{task-name}/, including scheme.md, code-context.md, challenger outputs, judge scores, and final-report.md. Files are kept after the review so users can inspect the full debate process.
A single sub-agent failure does not block the review. The main agent can retry the failed role or skip it, record that the role did not respond, and continue the remaining stages of the panel.
mr-daedaliumGuide structured design sessions to produce action-oriented design documents.
DektoraCreate, revise, and manage Architectural Decision Records for software architecture.
AutumnsGroveDesign RESTful and GraphQL APIs with OpenAPI specification generation.
bordenetScores code changes with three-critic review and flags anti-patterns before commit.