Codex Collab
JSON-RPC bridge that lets Claude Code delegate tasks and PR reviews to Codex and read the results back.
Test report
- Verdict
- Works with setup
- Score
- Tested
- Jul 13, 2026
- Environment
- Claude Code 2.x (agent harness)
- Upstream re-checked
- Aug 10, 2026 · 812a43e
Since our 13.07 test, the author released 0.3.0 with self-update and skill synchronization — this is precisely the type of change that could overturn the verdict, so Stage 0 was re-examined: api.github.com is polled daily for the latest release (only User-Agent is sent, no tokens or secrets), disabled via CODEX_COLLAB_NO_UPDATE_CHECK=1, nothing is applied automatically. We checked manually, not by text: added a junk line to the installed SKILL.md, a bare `codex-collab skill sync` printed a diff, exited with an error, and DID NOT touch the file; it applied only with `--yes`; SKILL.md itself explicitly forbids the agent from passing `--yes` on its own initiative. The only caveat is that `update --yes` pulls the release tarball from github.com and runs its install.sh without checksum verification, but only after explicit consent. Unlike last time, Bun was present, and install.sh worked properly: it built a 0.31 MB bundle + broker, unpacked it into ~/.claude/skills/codex-collab, installed a shim, and without codex on PATH, it honestly fails with a clear list of prerequisites. End-to-end A/B was not done: there is no working authorized codex CLI on the machine (alias points to a non-existent binary), so we did not see real Codex runs — this is an output assessment, not a guess. Trigger was lowered 5 -> 4 based on our re-reading (upstream did not change the description): 'use PROACTIVELY... second opinions on code' catches the usual 'look at my PR' where Codex was not called.
Scored on four weighted criteria — install, triggering, output vs. baseline, docs. How scoring works
- Installs cleanly 5/5
- Triggers reliably 4/5
- Output vs. baseline 6/10
- Docs & honesty 4/5
What Codex Collab does
Drives the Codex CLI's app-server JSON-RPC protocol so Claude can start threads, run prompted tasks, request code reviews, and handle mid-turn questions/approvals from Codex without leaving the session. Triggers when the user explicitly asks to invoke, delegate to, or get a second opinion from Codex, and is also written to fire proactively whenever an independent non-Claude perspective on code, plans, or architecture would help. Requires Bun and the Codex CLI installed and on PATH.
How to install Codex Collab
git clone https://github.com/Kevin7Qi/codex-collab.git
cd codex-collab && ./install.sh # нужны Bun >=1.0 и авторизованный CLI codex на PATH; ставит ~/.claude/skills/codex-collab/ и шим в ~/.local/bin
Skills live in ~/.claude/skills/ (global) or .claude/skills/
(per-project). Restart Claude Code after installing.
Commands — how to trigger Codex Collab
-
/codex-collabJSON-RPC bridge that lets Claude Code delegate tasks and PR reviews to Codex and read the results back.
It also activates on plain-language prompts like these:
-
Ask Codex for a second opinion on this PR before I merge it -
Delegate this refactor task to Codex and bring back the results -
Have Codex review my architecture plan and report back its notes
Frequently asked questions
- Is the Codex Collab skill free?
- Yes. The skill itself is free from Kevin7Qi/codex-collab. SkillProof publishes the install command and an independent test verdict at no cost.
- Does Codex Collab work with Claude Code?
- We tested it with Claude Code 2.x (agent harness) on Jul 13, 2026. Verdict: Works with setup. Since our 13.07 test, the author released 0.3.0 with self-update and skill synchronization — this is precisely the type of change that could overturn the verdict, so Stage 0 was re-examined: api.github.com is polled daily for the latest release (only User-Agent is sent, no tokens or secrets), disabled via CODEX_COLLAB_NO_UPDATE_CHECK=1, nothing is applied automatically. We checked manually, not by text: added a junk line to the installed SKILL.md, a bare `codex-collab skill sync` printed a diff, exited with an error, and DID NOT touch the file; it applied only with `--yes`; SKILL.md itself explicitly forbids the agent from passing `--yes` on its own initiative. The only caveat is that `update --yes` pulls the release tarball from github.com and runs its install.sh without checksum verification, but only after explicit consent. Unlike last time, Bun was present, and install.sh worked properly: it built a 0.31 MB bundle + broker, unpacked it into ~/.claude/skills/codex-collab, installed a shim, and without codex on PATH, it honestly fails with a clear list of prerequisites. End-to-end A/B was not done: there is no working authorized codex CLI on the machine (alias points to a non-existent binary), so we did not see real Codex runs — this is an output assessment, not a guess. Trigger was lowered 5 -> 4 based on our re-reading (upstream did not change the description): 'use PROACTIVELY... second opinions on code' catches the usual 'look at my PR' where Codex was not called.
- What is the Codex Collab SkillProof Score?
- 7.6/10 — installs cleanly 5/5, triggers reliably 4/5, output vs. baseline 6/10, docs & honesty 4/5.
- How do I install Codex Collab?
- Copy the install command from this page, run it in your terminal, and restart Claude Code. Skills live in ~/.claude/skills/ (global) or .claude/skills/ inside a project.
- Can I use Codex Collab with Cursor, Copilot, Gemini CLI, Codex or other AI tools?
- The SKILL.md format is native to Claude (Claude Code, Desktop, claude.ai). The instructions inside adapt to other assistants: Cursor rules, GitHub Copilot instructions, Windsurf rules, Custom GPTs, AGENTS.md for OpenAI Codex, and GEMINI.md for Google Gemini CLI — our conversion guides cover each, and the free converter on the tools page does the wrapping for you.