codex-cli (claude-codex-skill)

원본 보기

Claude Code 안에서 OpenAI Codex CLI를 호출하는 스킬. `/codex-cli review|impl|image` 세 모드. **이미지는 ChatGPT 로그인만으로 생성 — `OPENAI_API_KEY` 없이도 작동** (Codex CLI 내장 `image_gen.imagegen` 툴 활용). Windows 0.128.0의 sandbox copy bug까지 우회.

작성자: cskwork
라이선스: MIT
트리거: /codex-cli / use codex / ask codex / have codex / second-opinion review / generate image

#skill#claude-code#codex#image-gen#code-review#sandbox#openai

한 줄

Codex CLI에 숨겨진 ChatGPT-auth 이미지 생성을 발굴 + Codex의 본업(review / impl)을 Claude에 깔끔하게 위임하는 스킬. Codex codex login(ChatGPT) 하나면 끝, API key 별도 X.

세 가지 모드

Invocation	What	Example
`/codex-cli review`	diff를 Codex에 second-opinion review로	`/codex-cli review --base main`
`/codex-cli impl <prompt>`	Codex에 non-interactive 코딩 작업 위임	`/codex-cli impl "add JWT refresh to api/auth.ts"`
`/codex-cli image <prompt>`	Codex 내장 image gen — API key 불필요	`/codex-cli image "isometric CPU diagram, neon"`

자연어로도 트리거 가능 — “ask codex to review my diff”, “have codex generate an isometric CPU diagram” (스킬의 when_to_use 필드가 감지).

핵심 함정 (스킬이 알아서 처리)

이미지 생성의 디스커버리 문제 — 모델이 OPENAI_API_KEY나 curl/python을 언급한 prompt를 받으면 shell-out 경로로 빠지고 실패. 스킬은 “generate and save”라고만 시켜 내장 툴을 고르게 한다.
Windows sandbox copy bug — codex-cli 0.128.0 Windows에서 이미지 생성은 성공하나 workspace 복사가 CreateProcessAsUserW failed: 5로 실패. 스킬이 ~/.codex/generated_images/<session>/ig_*.png에서 직접 찾아 복사
Sandbox safety — -s danger-full-access / --dangerously-bypass-approvals-and-sandbox는 명시 per-run approval 없이 절대 사용 X

설치

# macOS / Linux
curl -fsSL https://raw.githubusercontent.com/cskwork/claude-codex-skill/main/install.sh | bash

# Windows PowerShell
iwr -useb https://raw.githubusercontent.com/cskwork/claude-codex-skill/main/install.ps1 | iex

수동: SKILL.md를 ~/.claude/skills/codex-cli/SKILL.md (또는 Windows %USERPROFILE%\.claude\skills\codex-cli\SKILL.md)에 복사 → Claude Code 재시작 → /codex-cli 확인.

전제

Claude Code 설치
Codex CLI ≥ 0.128.0 (codex --version)
codex login status = “Logged in” (ChatGPT 로그인이면 충분, API key 로그인도 OK)

SKILL.md 본문 (그대로 복사)

---
name: codex-cli
description: Delegate code review, code implementation, or image generation to the OpenAI Codex CLI for a second opinion. Image generation works through ChatGPT login alone — no OPENAI_API_KEY required.
when_to_use: User types /codex-cli, or asks to "use codex" / "ask codex" / "have codex" do X, or wants a second-opinion code review, or requests an image that Codex can produce.
allowed-tools: Bash(codex *) Bash(cp *) Bash(ls *) Bash(Get-ChildItem *) Bash(Copy-Item *) Read
---

Wrap the local codex CLI (codex --version >= 0.128.0) so Claude Code can hand
off three kinds of work without leaving the conversation:

- review — second-opinion code review on a diff
- impl — non-interactive coding task in a sandbox
- image — image generation via Codex built-in image_gen.imagegen tool
  (no API key, ChatGPT login is enough)

## Verify once per session

codex --version          # codex-cli >= 0.128.0
codex login status       # must say "Logged in"

If not logged in, tell the user to run `! codex login` from the prompt. Do not
log in on their behalf.

## Pick the mode

Parse $ARGUMENTS. First token selects review | impl | image; rest is prompt or
flags. If /codex-cli invoked with no subcommand, infer:

- mentions "review", "diff", "PR", "second opinion" -> review
- mentions "image", "picture", "render", "draw", "그려" -> image
- otherwise -> impl

State the mode you picked so the user can correct you.

## Mode: review

codex review --uncommitted              # staged + unstaged + untracked
codex review --base main                # PR-style against base branch
codex review --commit <sha>             # one specific commit
"Focus on concurrency. Skip nits." | codex review --uncommitted -

After it returns, paraphrase findings; group by CRITICAL / HIGH / MEDIUM / LOW
per ~/.claude/rules/common/code-review.md. Cite file:line per finding.

Codex review is second opinion, not ground truth. If it contradicts something
already verified in this session, flag the conflict instead of silently siding
with Codex.

## Mode: impl

codex exec "<prompt>" -s workspace-write -C "<absolute path>" \
  --output-last-message codex-out.txt

codex exec "<prompt>" -s read-only -C "<path>"
cat prompt.md | codex exec -s workspace-write -C "<path>" -
codex exec resume --last "<follow-up>"

Sandbox: default workspace-write. read-only for analysis/plan. Always pass -C
absolute path. Do NOT pass -s danger-full-access or
--dangerously-bypass-approvals-and-sandbox without explicit per-run approval.

After Codex finishes, read codex-out.txt (or the diff) and summarize. Do not
rerun `codex apply` without asking.

## Mode: image

Codex has built-in image tool (image_gen.imagegen). Activates when
`codex features list` shows image_generation = stable, true. Auth uses
codex login — ChatGPT login is enough.

codex exec --skip-git-repo-check -s workspace-write -C "<absolute dir>" \
  "Generate an image of <prompt>, 1024x1024. Save the result as <name>.png in the current working directory."

CRITICAL prompt rule: do NOT tell Codex to "use the OpenAI API", "use curl",
"use python", or "use openai CLI". Those force shell-out needing
OPENAI_API_KEY. Just say generate and save — Codex picks built-in tool.

Where the file lands: ~/.codex/generated_images/<session_id>/ig_<hash>.png
Then Codex tries to copy to workspace via shell. On Windows codex-cli 0.128.0
the copy fails (CreateProcessAsUserW failed: 5) even though generation
succeeded. Do NOT escalate to danger-full-access. Copy yourself:

PowerShell:
$src = Get-ChildItem "$env:USERPROFILE\.codex\generated_images" -Recurse -Filter "ig_*.png" |
  Sort-Object LastWriteTime -Descending | Select-Object -First 1
Copy-Item $src.FullName "<output path>\<name>.png"

bash:
src=$(ls -t ~/.codex/generated_images/*/ig_*.png 2>/dev/null | head -1)
cp "$src" "<output path>/<name>.png"

Show the user the absolute path. Do not embed PNG bytes in chat.
Supported sizes: 1024x1024, 1536x1024, 1024x1536.

## Safety

- Do not send confidential code/prompts to Codex without confirming.
- Do not generate images of real identifiable people without consent.
- Do not pass --dangerously-bypass-approvals-and-sandbox (CI-only).
- Do not pass -s danger-full-access without explicit per-run user approval.

## Output discipline

- Don't stream Codex stdout unless asked. Summarize.
- Save long sessions to codex-out.txt and quote only what's relevant.
- Always state the mode and exact command so the user can rerun.

## 한 줄

Codex CLI에 숨겨진 ChatGPT-auth 이미지 생성을 발굴 + Codex의 본업(review / impl)을 Claude에 깔끔하게 위임하는 스킬. Codex `codex login`(ChatGPT) 하나면 끝, API key 별도 X.

## 세 가지 모드

| Invocation | What | Example |
|---|---|---|
| `/codex-cli review` | diff를 Codex에 second-opinion review로 | `/codex-cli review --base main` |
| `/codex-cli impl <prompt>` | Codex에 non-interactive 코딩 작업 위임 | `/codex-cli impl "add JWT refresh to api/auth.ts"` |
| `/codex-cli image <prompt>` | Codex 내장 image gen — API key 불필요 | `/codex-cli image "isometric CPU diagram, neon"` |

자연어로도 트리거 가능 — "ask codex to review my diff", "have codex generate an isometric CPU diagram" (스킬의 `when_to_use` 필드가 감지).

## 핵심 함정 (스킬이 알아서 처리)

1. **이미지 생성의 디스커버리 문제** — 모델이 `OPENAI_API_KEY`나 `curl`/`python`을 언급한 prompt를 받으면 shell-out 경로로 빠지고 실패. 스킬은 "generate and save"라고만 시켜 내장 툴을 고르게 한다.
2. **Windows sandbox copy bug** — codex-cli 0.128.0 Windows에서 이미지 생성은 성공하나 workspace 복사가 `CreateProcessAsUserW failed: 5`로 실패. 스킬이 `~/.codex/generated_images/<session>/ig_*.png`에서 직접 찾아 복사
3. **Sandbox safety** — `-s danger-full-access` / `--dangerously-bypass-approvals-and-sandbox`는 명시 per-run approval 없이 절대 사용 X

## 설치

```bash
# macOS / Linux
curl -fsSL https://raw.githubusercontent.com/cskwork/claude-codex-skill/main/install.sh | bash

# Windows PowerShell
iwr -useb https://raw.githubusercontent.com/cskwork/claude-codex-skill/main/install.ps1 | iex
```

수동: `SKILL.md`를 `~/.claude/skills/codex-cli/SKILL.md` (또는 Windows `%USERPROFILE%\.claude\skills\codex-cli\SKILL.md`)에 복사 → Claude Code 재시작 → `/codex-cli` 확인.

## 전제

- Claude Code 설치
- Codex CLI ≥ 0.128.0 (`codex --version`)
- `codex login status` = "Logged in" (ChatGPT 로그인이면 충분, API key 로그인도 OK)

## SKILL.md 본문 (그대로 복사)

````markdown
---
name: codex-cli
description: Delegate code review, code implementation, or image generation to the OpenAI Codex CLI for a second opinion. Image generation works through ChatGPT login alone — no OPENAI_API_KEY required.
when_to_use: User types /codex-cli, or asks to "use codex" / "ask codex" / "have codex" do X, or wants a second-opinion code review, or requests an image that Codex can produce.
allowed-tools: Bash(codex *) Bash(cp *) Bash(ls *) Bash(Get-ChildItem *) Bash(Copy-Item *) Read
---

Wrap the local codex CLI (codex --version >= 0.128.0) so Claude Code can hand
off three kinds of work without leaving the conversation:

- review — second-opinion code review on a diff
- impl — non-interactive coding task in a sandbox
- image — image generation via Codex built-in image_gen.imagegen tool
  (no API key, ChatGPT login is enough)

## Verify once per session

codex --version          # codex-cli >= 0.128.0
codex login status       # must say "Logged in"

If not logged in, tell the user to run `! codex login` from the prompt. Do not
log in on their behalf.

## Pick the mode

Parse $ARGUMENTS. First token selects review | impl | image; rest is prompt or
flags. If /codex-cli invoked with no subcommand, infer:

- mentions "review", "diff", "PR", "second opinion" -> review
- mentions "image", "picture", "render", "draw", "그려" -> image
- otherwise -> impl

State the mode you picked so the user can correct you.

## Mode: review

codex review --uncommitted              # staged + unstaged + untracked
codex review --base main                # PR-style against base branch
codex review --commit <sha>             # one specific commit
"Focus on concurrency. Skip nits." | codex review --uncommitted -

After it returns, paraphrase findings; group by CRITICAL / HIGH / MEDIUM / LOW
per ~/.claude/rules/common/code-review.md. Cite file:line per finding.

Codex review is second opinion, not ground truth. If it contradicts something
already verified in this session, flag the conflict instead of silently siding
with Codex.

## Mode: impl

codex exec "<prompt>" -s workspace-write -C "<absolute path>" \
  --output-last-message codex-out.txt

codex exec "<prompt>" -s read-only -C "<path>"
cat prompt.md | codex exec -s workspace-write -C "<path>" -
codex exec resume --last "<follow-up>"

Sandbox: default workspace-write. read-only for analysis/plan. Always pass -C
absolute path. Do NOT pass -s danger-full-access or
--dangerously-bypass-approvals-and-sandbox without explicit per-run approval.

After Codex finishes, read codex-out.txt (or the diff) and summarize. Do not
rerun `codex apply` without asking.

## Mode: image

Codex has built-in image tool (image_gen.imagegen). Activates when
`codex features list` shows image_generation = stable, true. Auth uses
codex login — ChatGPT login is enough.

codex exec --skip-git-repo-check -s workspace-write -C "<absolute dir>" \
  "Generate an image of <prompt>, 1024x1024. Save the result as <name>.png in the current working directory."

CRITICAL prompt rule: do NOT tell Codex to "use the OpenAI API", "use curl",
"use python", or "use openai CLI". Those force shell-out needing
OPENAI_API_KEY. Just say generate and save — Codex picks built-in tool.

Where the file lands: ~/.codex/generated_images/<session_id>/ig_<hash>.png
Then Codex tries to copy to workspace via shell. On Windows codex-cli 0.128.0
the copy fails (CreateProcessAsUserW failed: 5) even though generation
succeeded. Do NOT escalate to danger-full-access. Copy yourself:

PowerShell:
$src = Get-ChildItem "$env:USERPROFILE\.codex\generated_images" -Recurse -Filter "ig_*.png" |
  Sort-Object LastWriteTime -Descending | Select-Object -First 1
Copy-Item $src.FullName "<output path>\<name>.png"

bash:
src=$(ls -t ~/.codex/generated_images/*/ig_*.png 2>/dev/null | head -1)
cp "$src" "<output path>/<name>.png"

Show the user the absolute path. Do not embed PNG bytes in chat.
Supported sizes: 1024x1024, 1536x1024, 1024x1536.

## Safety

- Do not send confidential code/prompts to Codex without confirming.
- Do not generate images of real identifiable people without consent.
- Do not pass --dangerously-bypass-approvals-and-sandbox (CI-only).
- Do not pass -s danger-full-access without explicit per-run user approval.

## Output discipline

- Don't stream Codex stdout unless asked. Summarize.
- Save long sessions to codex-out.txt and quote only what's relevant.
- Always state the mode and exact command so the user can rerun.
````