Does Deep Review work with local uncommitted changes?

Yes, running the /deep_review command without arguments will automatically identify and review your local uncommitted changes.

How does the skill handle disagreements between agents?

The orchestrator uses an evidence-based synthesis phase. For example, if a Skeptic finds a bug but the Advocate defends it, the orchestrator checks for specific file-and-line evidence to determine the winner.

Can I use Deep Review for large Pull Requests?

Yes, though if changes exceed 50KB, the skill will intelligently summarize less-critical files or suggest splitting the review to maintain high analysis quality.

What is an adversarial code review?

It is a review process where three distinct AI agents—an Advocate, a Skeptic, and an Architect—analyze your code with opposing goals to ensure every angle, from logic to architecture, is thoroughly tested.

Is my source code kept after the review?

No. Deep Review writes a temporary context file to the system /tmp directory for the subagents to read and explicitly deletes it immediately after the synthesis is complete.

Deep Review

Name: Deep Review
Author: shouenlee

byshouenlee

0•

コラボレーションツール

Performs rigorous code reviews by synthesizing three adversarial perspectives to identify bugs, architectural flaws, and design rationale.

Deep Review is a specialized Claude Code skill designed to eliminate reviewer blind spots through a multi-agent adversarial framework. Instead of a single analysis, it spawns three parallel subagents—the Advocate, the Skeptic, and the Architect—who independently evaluate code changes from distinct viewpoints. This tension surfaces hidden bugs, edge cases, and structural smells that a single pass might miss. The process concludes with a synthesized report that reconciles conflicting views and provides a machine-readable findings block, making it an essential tool for high-stakes PR audits and complex refactoring tasks.

主な機能

01Parallel processing for fast, cost-effective code audits

02Automated conflict resolution based on evidence-driven reasoning

030 GitHub stars

04Adversarial Three-Agent Analysis (Advocate, Skeptic, Architect)

05Machine-readable structured findings for CI/CD integration

06Support for local diffs, Pull Requests, specific commits, and branch comparisons

ユースケース

01Rigorous security and edge-case testing for mission-critical modules

02Architectural impact assessment for large-scale code refactors

03Pre-submission audit of complex Pull Requests to catch regressions

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add shouenlee/ghcp-dev-plugin deep_review

For use in Claude.ai and ChatGPT

主な機能

01Parallel processing for fast, cost-effective code audits

02Automated conflict resolution based on evidence-driven reasoning

030 GitHub stars

04Adversarial Three-Agent Analysis (Advocate, Skeptic, Architect)

05Machine-readable structured findings for CI/CD integration

06Support for local diffs, Pull Requests, specific commits, and branch comparisons

ユースケース

01Rigorous security and edge-case testing for mission-critical modules

02Architectural impact assessment for large-scale code refactors

03Pre-submission audit of complex Pull Requests to catch regressions

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add shouenlee/ghcp-dev-plugin deep_review

For use in Claude.ai and ChatGPT