parallel-expert-review

Reviews code changes and refactor plans using parallel architect and engineer subagents.

2|Updated Aug 2, 2026
One-click install
npx skills add https://github.com/Arasz/ai-raccoon --skill parallel-expert-review-arasz
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: parallel-expert-review
Source: https://github.com/Arasz/ai-raccoon/tree/main/.ai-badger/skills/learned/uncategorized/parallel-expert-review
Command: npx skills add https://github.com/Arasz/ai-raccoon --skill parallel-expert-review-arasz

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Single-perspective code reviews miss architectural flaws or implementation bugs, and stale plan documents lead reviewers to critique code that no longer exists. This Skill coordinates multiple expert reviewers and integrates their findings into one verified report. ## Core Features & Use Cases - Parallel expert dispatch: Sends an architect subagent (structure, composition, naming) and a domain engineer subagent (idioms, correctness, edge cases) to review the same change concurrently, then merges findings by theme. - Adversarial conflict resolution: When experts disagree, each defends its position and the orchestrator settles disputes by reading the actual merged source rather than trusting lane counts. - MoE plan review variant: Ground-truths every claim in a refactor plan against current code before dispatch, routes open questions through an owner-gate decision form, and produces before/after mermaid architecture diagrams. - Use Case: Before starting a large refactor, dispatch all expert personas against the plan document, collect a unified findings report with severity ratings, answer owner questions via decision cards, and receive before/after module diagrams — all before writing any code. ## Quick Start Review this pull request with parallel architect and engineer experts, integrate their findings, and flag any conflicts for adversarial resolution.

Frequently Asked Questions about parallel-expert-review

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

FAQPage Schema
How do I review a pull request with multiple AI expert agents?

Dispatch an architect subagent focused on structure, composition, and naming alongside a domain engineer subagent focused on idioms, correctness, and edge cases. Give each a self-contained prompt with the worktree path and file list, then merge their findings by theme.

How to resolve conflicting code review findings from different reviewers?

Present each agent's position to the other and ask them to defend or concede, then decide based on evidence from the actual source code. Never settle disputes by counting which lane reported more — read the merged file directly to verify each claim.

Can AI agents review a refactor plan before implementation starts?

Yes, but first verify every factual claim in the plan against current code, since plans go stale between writing and review. Feed experts the verified ground truth explicitly, keep the review read-only, and stop at the owner-gate before any coding begins.

What happens if a PR gets merged during an ongoing code review?

Do not assume the review is still valid on the old head. Fetch and reset to origin/main, diff the reviewed head against the merge commit, re-run all gates on the merged state, and re-verify every finding against the merged files.

When should I not use parallel multi-agent code review?

Avoid it for trivial single-file changes where one reviewer suffices, and never start implementation immediately after the review phase ends. The workflow caps concurrency at three subagents and requires waiting for explicit owner approval before coding.