Claude Codex Review Skill — Code Quality skill for Claude Code
File-based dual-review skill for Claude Code + Codex: iterative plan/code review through plain files, no MCP server needed.
How to install Claude Codex Review Skill
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open Serg2000Mr/claude-codex-review-skill and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Claude Codex Review Skill does
File-based dual-review skill for Claude Code + Codex: iterative plan/code review through plain files, no MCP server needed.
Alternatives in Code Quality
- Open Code Review — /open-code-review 21.5k ★
- Code Quality Review — Review code quality in: $ARGUMENTS 5.8k ★
- Ars Lit Review — ARS academic-paper lit-review mode — annotated bibliography in paper format 4.5k ★
README
[Русская версия](README.ru.md)
Claude-Codex Review Skill
This skill lets Claude Code and Codex review a plan or code change through a crash-resilient file protocol. Claude Code initiates the session, Codex acts as an independent reviewer, and both agents use the same project directory without an MCP server or CLI bridge.
Quick start
Copy the complete `claude-code/` directory into the project skill directory:
mkdir -p .claude/skills/codex-dual-review-file-based
cp claude-code/* .claude/skills/codex-dual-review-file-based/
Then run in Claude Code:
/codex-dual-review-file-based [path/to/plan.md] [max_rounds=5]
Claude Code creates the first round and prints one ready-to-use instruction for Codex. Codex keeps watching the same session for later rounds; the user does not need to pass the prompt again.
How it works
Each session lives in an absolute `
R1-01-round-start.md <- Claude writes the review contract
R1-02-codex-claimed.flg <- Codex claims the round
R1-03-codex-review.md <- Codex writes the review
R1-04-claude-claimed.flg <- Claude processes the result
final.md <- Claude records the session outcome
Round-start and review files contain a machine-readable JSON block plus a human-readable explanation. Claude verifies every finding as a hypothesis, applies accepted changes before opening the next round, and records rejected findings with evidence.
The session ends with:
approvedafterAPPROVED;failedafter the rareREJECTEDverdict;limitwhenmax_roundsis reached.
`APPROVED` alone does not finish a session: `final.md` is the completion marker.
Protocol guarantees
- Session files are append-only: protocol files are never overwritten or deleted.
- Absolute
SESSION_DIR, reviewer-prompt, and helper-script paths
Related Skills
Paranoia
Local MCP server for cross-agent adversarial code review — Claude Code and Codex CLI critique each other's wor
Claude Skill For Multimodal Report Generation
A Claude Skill that converts mixed-format files (documents, images, audio/video) into structured, template-bas
MCP Config
Turn one MCP server setup into the right format for lots of apps.
Skillmd
Lint, search, and install Agent Skills (SKILL.md) — CLI + MCP server + lint engine
Cfn Vote Implement
MUST BE USED after cfn-dry-review or cfn-alpha-launch:manifest produces a manifest. Also verification phase of
Lens Review
Researched multi-lens code review – fans out lens reviewers over a target, each grounded in recent cited best
Related Agents
Codex Bridge
把指定工作外包給 gpt-5.5(codex)並忠實回傳其輸出。當需要 gpt-5.5 的批量實作或獨立第二審查視角時使用。這是 plugin 內第三種 codex 路徑:一次性 codex exec CLI 呼叫、輸出
Codex Teammate
Use Codex as a bounded implementation and review teammate through claude-codex-mcp-bridge.
Workflow Orchestration Command - Enhanced With Optional Flags
You are the product-manager-orchestrator agent. You orchestrate specialist agents but NEVER write code yoursel