metareview

Orchestrates parallel multi-domain code reviews and consolidates findings into a deduplicated report.

Updated Nov 21, 2025
One-click install
npx skills add https://github.com/alatyshau/duet --skill metareview-alatyshau
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: metareview
Source: https://github.com/alatyshau/duet/tree/main/packages/skills/modes/metareview
Command: npx skills add https://github.com/alatyshau/duet --skill metareview-alatyshau

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Changes that span multiple domains (code, specs, tests, security, AI instructions) are hard to review thoroughly with a single perspective, leading to missed issues, duplicated findings, and false positives. ## Core Features & Use Cases - Domain-based reviewer selection: Maps changed files to expert reviewer roles (Python Engineer, Test Engineer, Spec Architect, Security Analyst, and more) and launches them as parallel sub-agents. - Review of Reviews: The orchestrator verifies each finding against primary sources, dismisses false positives, and deduplicates issues found by multiple reviewers. - Structured output: Produces a metareview.md with a summary table, action clusters with dependency chains, and a dismissed-issues table that serves as a strict spec for the executor. - Use Case: After implementing a feature that touches Python code, API contracts, and spec documents, run a metareview to get verified, deduplicated issues grouped into actionable clusters instead of three overlapping raw review reports. ## Quick Start Enter METAREVIEW mode to review the current change across all affected domains and produce a consolidated metareview.md report.

Frequently Asked Questions about metareview

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I review a change that spans multiple domains like code, tests, and specs?

Use a metareview workflow: analyze the intent and the git diff, map each changed file to a domain, assign an expert reviewer per domain, and launch all reviewers as parallel sub-agents. The orchestrator then verifies and merges their findings.

What is the difference between a regular code review and a metareview?

A regular review covers a single domain, while a metareview handles changes touching two or more domains requiring different expertise. It adds an orchestration layer that selects reviewers, challenges their findings, and deduplicates issues across reports.

How do I handle false positives from multiple AI code reviewers?

The orchestrator performs a Review of Reviews: it reads primary sources to verify each claim, dismisses findings that are by-design or explicitly deferred, and records dismissed issues with reasons in a dedicated table so executors do not fix non-issues.

When should a security reviewer be included in a code review?

Include a Security Analyst whenever the change touches file I/O, path handling, or external input. A single file can trigger multiple reviewers, for example an endpoint change may need Python, API, and Security reviewers simultaneously.

What are the limitations of an orchestrated multi-agent review?

The orchestrator does not write a final verdict or fix issues; it only produces the consolidated metareview.md. Sub-agents cannot read instruction files themselves, so all review rules must be injected verbatim into each prompt.