cc-plugin-codex

Use Claude Code from inside Codex for code reviews and delegated tasks.

This repository is maintained by Sendbird and follows the overall shape of openai/codex-plugin-cc, but in the opposite direction: Codex hosts the plugin and delegates work to Claude Code.

What You Get

• $cc:review for a normal read-only Claude Code review
• $cc:adversarial-review for a steerable challenge review
• $cc:rescue, $cc:status, $cc:result, and $cc:cancel to delegate work and manage tracked jobs
• $cc:setup to verify Claude Code readiness, hook installation, rescue-agent wiring, and review-gate state

How This Differs From Upstream

The goal is to stay close to the upstream OpenAI plugin's UX, but Claude Code and Codex expose different runtime surfaces.

| Topic | openai/codex-plugin-cc | This repository | | --- | --- | --- | | Host app | Claude Code hosts the plugin | Codex hosts the plugin | | User command surface | Claude slash commands such as /codex:review | Codex skills such as $cc:review | | Delegated runtime | Codex app-server + broker | Fresh claude -p subprocess per invocation | | Review gate subject | Reviews the previous Claude response before Claude stops | Reviews the previous Codex response before Codex stops | | Rescue agent | Plugin-local Codex rescue agent inside Claude | Global cc-rescue agent in ~/.codex/agents | | Model / effort flags | Codex model names and Codex effort controls | Claude model names and Claude effort values: low, medium, high, max |

Where This Goes Further

In addition to mirroring the upstream command surface, this repository adds a few implementation-focused optimizations:

• The stop-time review gate computes a turn baseline and skips Claude review entirely when the most recent Codex turn made no net edits, which helps avoid unnecessary token spend.
• Nested helper sessions suppress stop-time review and unread-result prompts, so the review gate stays attached to the user-facing Codex thread instead of recursive child runs.
• Background Claude jobs track unread/viewed state and session ownership, which makes $cc:status, $cc:result, and follow-up rescue flows safer for concurrent work.
• The installer is idempotent and manages the personal marketplace entry, hooks, and global cc-rescue registration together, so install and reinstall are a single step.

Requirements

• Codex with hook support
• Node.js 18 or later
• Claude Code CLI installed and authenticated
  • claude auth login, or
  • ANTHROPIC_API_KEY set in the environment

Install

Choose either install path below. Both install the plugin into ~/.codex/plugins/cc, create or update ~/.agents/plugins/marketplace.json, enable cc@local-plugins in ~/.codex/config.toml, enable codex_hooks = true, and install Codex hooks plus the global cc-rescue agent.

npx

npx cc-plugin-codex install

Shell Script

curl -fsSL "https://raw.githubusercontent.com/sendbird/cc-plugin-codex/main/scripts/install.sh" | bash

Update

Rerun either install command. The installer is safe to run again and will refresh the installed copy in place.

npx cc-plugin-codex update
curl -fsSL "https://raw.githubusercontent.com/sendbird/cc-plugin-codex/main/scripts/install.sh" | bash

Uninstall

npx cc-plugin-codex uninstall
curl -fsSL "https://raw.githubusercontent.com/sendbird/cc-plugin-codex/main/scripts/uninstall.sh" | bash

Manual Install

If you want to work from a local checkout instead of the one-shot installer:

mkdir -p ~/.codex/plugins
git clone https://github.com/sendbird/cc-plugin-codex.git ~/.codex/plugins/cc
cd ~/.codex/plugins/cc
node scripts/local-plugin-install.mjs install --plugin-root ~/.codex/plugins/cc

If Claude Code is not installed yet:

npm install -g @anthropic-ai/claude-code
claude auth login

Then run:

$cc:setup

After install, you should see:

• the $cc:* skills listed in Codex
• the global cc-rescue agent installed under ~/.codex/agents/cc-rescue.toml

One simple first run is:

$cc:review --background
$cc:status
$cc:result

Usage

$cc:review

Runs a normal Claude Code review on your current work.

Use it when you want:

• a review of your current uncommitted changes
• a review of your branch compared to a base branch like main

It supports --base <ref>, --scope <auto|working-tree|branch>, --wait, --background, and --model <model>.

Examples:

$cc:review
$cc:review --base main
$cc:review --background

This command is read-only. When run in the background, use $cc:status to check progress and $cc:cancel to stop it.

$cc:adversarial-review

Runs a steerable review that questions the chosen implementation and design.

Use it when you want:

• a review before shipping that challenges the direction, not just the code details
• review focused on design choices, tradeoffs, hidden assumptions, and alternative approaches
• pressure-testing around specific risk areas like auth, data loss, rollback, race conditions, or reliability

It uses the same target selection as $cc:review, including --base <ref>, and also accepts extra focus text after the flags.

Examples:

$cc:adversarial-review
$cc:adversarial-review --base main challenge whether this was the right caching and retry design
$cc:adversarial-review --background look for race conditions and question the chosen approach

This command is read-only. It does not fix code.

$cc:rescue

Hands a task to Claude Code through the global cc-rescue agent.

Use it when you want Claude Code to:

• investigate a bug
• try a fix
• continue a previous Claude task
• take a cheaper or faster pass with a smaller Claude model

It supports --background, --wait, --resume, --resume-last, --fresh, --write, --model <model>, --effort <low|medium|high|max>, and --prompt-file <path>.

Examples:

$cc:rescue investigate why the tests started failing
$cc:rescue fix the failing test with the smallest safe patch
$cc:rescue --resume apply the top fix from the last run
$cc:rescue --model sonnet --effort medium investigate the flaky integration test
$cc:rescue --background investigate the regression

$cc:status

Shows running and recent Claude Code jobs for the current repository.

Examples:

$cc:status
$cc:status task-abc123

$cc:result

Shows the final stored Claude Code output for a finished job.

When available, it also includes the Claude session ID so you can reopen that run directly with:

claude --resume <session-id>

Examples:

$cc:result
$cc:result task-abc123

$cc:cancel

Cancels an active background Claude Code job.

Examples:

$cc:cancel
$cc:cancel task-abc123

$cc:setup

Checks whether Claude Code is installed and authenticated.

It also verifies:

• hook installation
• global cc-rescue registration
• current review-gate state for this workspace

Enabling Review Gate

$cc:setup --enable-review-gate
$cc:setup --disable-review-gate

When the review gate is enabled, the plugin uses a Stop hook to run a targeted Claude Code review based on Codex's previous response. If that review finds issues, the stop is blocked so Codex can address them first.

[!WARNING] The review gate can create a long-running Codex/Claude loop and may drain usage limits quickly. Only enable it when you plan to actively monitor the session.

Typical Flows

Review Before Shipping

$cc:review --background
$cc:status
$cc:result

Challenge A Design Choice

$cc:adversarial-review --background question the retry, rollback, and caching strategy

Hand A Problem To Claude Code

$cc:rescue fix the failing test with the smallest safe patch

Continue A Previous Claude Task

$cc:rescue --resume apply the top fix from the last run

License

Apache-2.0