Dual Review File Based — AI skill for Claude Code
File-based cross-review and correction between independent AI agents / Файловое кросс-ревью и исправление между независимыми ИИ-агентами.
How to install Dual Review File Based
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open Serg2000Mr/dual-review-file-based and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Dual Review File Based does
File-based cross-review and correction between independent AI agents / Файловое кросс-ревью и исправление между независимыми ИИ-агентами.
Alternatives in AI
- Codex — openai-codex Adversarial code review, Codex CLI integration, cross-model analysis 9.6k ★
- Claudexor — Multi-harness control plane for Claude Code, Codex, Cursor, and OpenCode: quota-aware rotation across multiple 425 ★
- Skill Porter - Cross-Platform Skill Converter — This skill automates the conversion between Claude Code skills and Gemini CLI extensions, enabling true cross 160 ★
README
[Русская версия](README.ru.md)
Dual Review File Based
Two AI agents review plans and code iteratively, coordinating through plain files with no server or orchestrator needed.
Why
AI-generated plans and code benefit from a second opinion before implementation. But setting up a review pipeline between agents usually requires MCP servers, orchestrators, or custom infrastructure. This skill replaces all of that with a simple file-based protocol: agents read and write markdown files in a shared directory, each round producing a structured review with verdicts and evidence-backed issues.
Roles are not tied to applications: Codex → Codex, Codex → ZCode, Claude Code → Codex, and other pairings all work. For cross-model review, select different models inside those applications. A separate chat using the same model is also supported and still provides independent context. Both agents work in the same project directory; no MCP server or CLI bridge is required.
How it works
The initiator gives an independent agent the task and a list of files. The reviewer examines the actual project files and writes structured findings. The initiator verifies each finding, applies justified fixes, and opens another round. The updated files are reviewed again until approval or the configured round limit.
Session state lives in `.dual-review/` inside the project, so the review can continue after a restart or interruption. By default, the reviewer receives the task manually once and then follows subsequent rounds in the same session.
Installation
Choose the bundle for the application where the initiator manages the session. Do not mix files from the two bundles, and finish active sessions before updating an installed copy.
Codex
Copy the contents of `codex/` to `.agents/skills/claude-dual-review-file-based/` in your project:
New-Item -ItemType Directory -Force .agents/skills/claude-dual-review-file-based
Copy-Item codex/* .agents/skills/claude-dual-review
Related Skills
Model Crosscheck
Adversarial multi-model planning & code review for Claude Code: /dual-plan and /dual-review (Claude + OpenAI C
Santa Review
Adversarial dual-review convergence loop — two independent model reviewers must both approve before code ships
V Review Plan
Run an independent cross-model (Codex) adversarial review of a Compound V plan/manifest before dispatch, then
Multi Agent Protocol
A four-party collaboration protocol for running AI agent teams on real work — auditable authorization, durable
Astra Review
Send Claude-written code to GPT-6-Astra (ChatGPT, via Codex CLI) for an independent cross-provider review, the
Neo Agent Brain
Point a cross-model AI engineering team at your own codebases. The Agent OS keeps named AI maintainers' identi
Related Agents
Swe Gardener
Use this agent when you need dual-altitude software engineering that oscillates between architecture (system-w
Deep Repo Auditor
Cross-validated deep repository audit — dual AI models (Sonnet + Codex) in parallel, consolidated report with
Winterbotham
Consolidates review findings into canonical reports. Two modes - cross-discipline (merges Fagan + Oppenheimer