How do I install the Codex Claude Loop skill?

Run npx skillmds@latest add tomevault-io/codex-claude-loop in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Codex Claude Loop safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Codex Claude Loop?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Codex Claude Loop free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Codex Claude Loop?

tomevault-io (@tomevault-io) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Codex Claude Loop by tomevault-io — Agent Skill

Name: Codex Claude Loop
Author: tomevault-io

Codex Claude Loop

Orchestrates a dual-AI engineering loop where Claude CLI runs with `--dangerously-skip-permissions` for planning/implementation and Codex validates/reviews outputs. Use when users ask for Claude+Codex collaboration, cross-review, dual AI loop, or explicit Claude CLI execution with skip-permission mode ("claude", "dangerously-skip-permission", "dangerously-skip-permissions", "codex claude", "dual AI", "교차 검증", "claude 협업"). Use when this capability is needed.

tomevault-io 1cee03f 2 files · 3.2 KB Updated Sep 17, 2026

File contents

Codex-Claude Loop

Claude CLI + Codex 협업 루프를 실행한다.

Preconditions

Run --dangerously-skip-permissions only when the user explicitly requested it.
Prefer isolated environment: disposable branch/worktree, no production credentials.
Default Codex verification to --sandbox read-only.

Core Loop

Plan (Claude) -> Validate (Codex) -> Feedback ->
Implement (Claude) -> Review (Codex) ->
Fix (Claude) -> Re-validate (Codex)

Quick Commands

# 1) Claude non-interactive output
claude --dangerously-skip-permissions -p "Write an implementation plan for: <task>"

# 2) Codex validation (read-only)
echo "<claude_output>" \
| codex exec -m gpt-5.2-codex --config model_reasoning_effort="high" --sandbox read-only

# 3) Claude interactive implementation session
claude --dangerously-skip-permissions

Scripted Commands

Use the bundled helper script for repeatable runs:

# Claude one-shot
bash .codex/skills/codex-claude-loop/scripts/claude-codex-loop.sh claude-print "Plan this task: <task>"

# Codex one-shot review
bash .codex/skills/codex-claude-loop/scripts/claude-codex-loop.sh codex-review "Review this plan: <plan>"

# One roundtrip: Claude plan -> Codex review
bash .codex/skills/codex-claude-loop/scripts/claude-codex-loop.sh roundtrip "Implement feature X"

Operating Rules

Keep handoff explicit: include assumptions, constraints, and acceptance criteria in every prompt.
Treat Claude output as draft input. Use Codex to challenge edge cases, regressions, and test gaps.
If either CLI returns non-zero exit, stop and summarize the failure before retrying.
For high-impact edits (schema, auth, deployment), ask the user before applying changes.

Recovery

Resume Claude session: claude -c
Resume Codex session: echo "continue" | codex exec resume --last
If context drifts, re-anchor with a short summary and current objective.

Converted and distributed by TomeVault — claim your Tome and manage your conversions.

---name: codex-claude-loopdescription: Orchestrates a dual-AI engineering loop where Claude CLI runs with `--dangerously-skip-permissions` for planning/implementation and Codex validates/reviews outputs. Use when users ask for Claude+Codex collaboration, cross-review, dual AI loop, or explicit Claude CLI execution with skip-permission mode ("claude", "dangerously-skip-permission", "dangerously-skip-permissions", "codex claude", "dual AI", "교차 검증", "claude 협업"). Use when this capability is needed.---# Codex-Claude LoopClaude CLI + Codex 협업 루프를 실행한다.## Preconditions- Run `--dangerously-skip-permissions` only when the user explicitly requested it.- Prefer isolated environment: disposable branch/worktree, no production credentials.- Default Codex verification to `--sandbox read-only`.## Core Loop```Plan (Claude) -> Validate (Codex) -> Feedback ->Implement (Claude) -> Review (Codex) ->Fix (Claude) -> Re-validate (Codex)```## Quick Commands```bash# 1) Claude non-interactive outputclaude --dangerously-skip-permissions -p "Write an implementation plan for: <task>"# 2) Codex validation (read-only)echo "<claude_output>" \| codex exec -m gpt-5.2-codex --config model_reasoning_effort="high" --sandbox read-only# 3) Claude interactive implementation sessionclaude --dangerously-skip-permissions```## Scripted CommandsUse the bundled helper script for repeatable runs:```bash# Claude one-shotbash .codex/skills/codex-claude-loop/scripts/claude-codex-loop.sh claude-print "Plan this task: <task>"# Codex one-shot reviewbash .codex/skills/codex-claude-loop/scripts/claude-codex-loop.sh codex-review "Review this plan: <plan>"# One roundtrip: Claude plan -> Codex reviewbash .codex/skills/codex-claude-loop/scripts/claude-codex-loop.sh roundtrip "Implement feature X"```## Operating Rules- Keep handoff explicit: include assumptions, constraints, and acceptance criteria in every prompt.- Treat Claude output as draft input. Use Codex to challenge edge cases, regressions, and test gaps.- If either CLI returns non-zero exit, stop and summarize the failure before retrying.- For high-impact edits (schema, auth, deployment), ask the user before applying changes.## Recovery- Resume Claude session: `claude -c`- Resume Codex session: `echo "continue" | codex exec resume --last`- If context drifts, re-anchor with a short summary and current objective.---> Converted and distributed by [TomeVault](https://tomevault.io/claim/tygwan) — claim your Tome and manage your conversions.

Frequently asked questions