claude-gpt-workflow (Claude Skill) — 설치 & 라이브 데모

왜 쓰나요

핵심 기능

Role separation: Claude strategizes, Codex executes
Adversarial plan review between two models
Three feedback loops (A: plan, B: review/fix, C: batch)
Shell-based integration via ask_codex.sh
Quality gate on each iteration

라이브 데모

실제 사용 모습

claude-gpt-workflow-skill.replay ▶ 준비됨

0/0

설치

작성자 저장소 · Claude Skills 공식 문서

사용 사례

실전 활용법: claude-gpt-workflow

Stress-test a plan by having another model attack it

👤 Engineers doing high-stakes refactors ⏱ ~45 min intermediate

언제 쓸까: Before touching code on a migration or risky refactor, you want a second model to poke holes.

흐름

Draft a plan with Claude

Write a plan.md for migrating our auth module from JWT to session cookies.✓ 복사됨

→ Step-by-step plan written to plan.md
Trigger adversarial review

/plan-review plan.md✓ 복사됨

→ Codex returns critique with risks Claude missed
Iterate until both agree

Incorporate Codex's critique, re-submit, repeat until clean.✓ 복사됨

→ Plan reaches stable state

결과: A plan that has survived real adversarial review, not just self-grading.

함정

Endless review loops — Cap at 3 rounds — diminishing returns after

함께 쓰기: filesystem · git

Delegate a batch refactor to Codex with Claude reviewing

👤 Teams doing mechanical multi-file changes ⏱ ~60 min intermediate

언제 쓸까: A rename / signature change across 50+ files.

흐름

Plan and approve

Plan + review a rename of <old> to <new> across the codebase.✓ 복사됨

→ Approved plan.md
Execute

/plan-execute plan.md — Codex does the edits, Claude reviews each batch.✓ 복사됨

→ Files change in batches, each passes Claude's review

결과: A clean multi-file change with two sets of eyes on every batch.

함정

Codex does something Claude would have rejected — The /plan-execute loop catches it — but only if quality criteria in the plan are explicit

함께 쓰기: git

Implement a feature with iterative quality refinement

👤 Devs who've seen one-shot AI code fall over on edge cases ⏱ ~90 min advanced

언제 쓸까: A feature complex enough that a single generation won't get it right.

흐름

Write plan with acceptance criteria

Plan a feature with explicit acceptance tests, then /plan-execute.✓ 복사됨

→ Plan with testable criteria
Watch the loop

Let the review/fix loop run — summarize each iteration.✓ 복사됨

→ Iterations converge to passing criteria

결과: A feature that passes explicit acceptance, not just a commit that compiles.

조합

다른 MCP와 조합해 10배 효율

claude-gpt-workflow-skill + git

Commit after each loop iteration so you can revert if the loop diverges

Before each /plan-execute iteration, create a git tag.✓ 복사됨

claude-gpt-workflow-skill + filesystem

Plan and critiques live in the repo

Write plan.md and review-notes.md to docs/ and keep them in git.✓ 복사됨

도구

이 MCP가 노출하는 것

도구	입력	언제 호출	비용
ask_codex.sh	request string	Delegate a coding task	Codex API tokens
/plan-review	plan file path	Before execution	Codex API tokens
/plan-execute	approved plan path	After /plan-review passes	Codex tokens per iteration

비용 및 제한

운영 비용

API 쿼터: Codex API tokens — potentially 2–5x a single-model workflow since both models run
호출당 토큰: High — plan + review + execute
금액: Two API subscriptions (Claude + OpenAI)
팁: Reserve this for risky work; don't use it for small one-file edits

보안

권한, 시크릿, 파급범위

자격 증명 저장: OPENAI_API_KEY in env for Codex CLI

데이터 외부 송신: Code leaves your machine to both providers — check corporate policy

Two-model workflow doubles the surface that sees your source.

문제 해결

자주 발생하는 오류와 해결

ask_codex.sh: command not found

Codex CLI not on PATH or not installed

확인: which codex

Plan-execute loops forever

Acceptance criteria too vague — tighten the plan or cap iterations

Models disagree and plan never stabilizes

Force a tiebreaker round with a shorter scope

대안

claude-gpt-workflow 다른 것과 비교

대안	언제 쓰나	단점/장점
Single-model Claude Code	The task isn't risky enough to justify two-model overhead	Fewer adversarial signals
agent-skill-tdd	Discipline comes from tests, not from another model's critique	No cross-model review

claude-gpt-workflow