Using Deliberation with the Codex plugin
This page is for Claude Code users who use, or are considering, OpenAI's Codex plugin. openai/codex-plugin-cc (external) runs Codex reviews from Claude Code and hands tasks to Codex. Deliberation asks several model families for a second opinion and runs a review loop across them. The two plugins do different jobs and install side by side.
Facts about the Codex plugin come from its README as of 2026-11-17: 35,182 GitHub stars, Apache-2.0, latest release v1.0.6 (published 2026-07-08).
What each one does
| Codex plugin | Deliberation | |
|---|---|---|
| Models | Codex (OpenAI) | GPT through the Codex CLI, Gemini through an AI Studio API key (Antigravity CLI opt-in), Grok, OpenRouter models, local Ollama and LM Studio models |
| Review | /codex:review (read-only, current changes or --base <ref>), /codex:adversarial-review (takes focus text) |
/deliberation:ask-all (parallel answers from the panel), /deliberation:consensus (multi-round review until the panel agrees), eight expert personas |
| Hand off work | /codex:rescue lets Codex investigate and fix, /codex:transfer moves the session into Codex |
GPT is advisory only. Only Gemini can edit files, through Deliberation's Gemini bridge (Claude Code plugin) |
| Background jobs | --background, /codex:status, /codex:result, /codex:cancel |
Calls run in the foreground. panel plus parallel ask-one calls show per-provider progress |
| Review gate | Optional Stop hook (/codex:setup --enable-review-gate) that blocks Claude from stopping until a Codex review passes |
None. Deliberation is advisory and never blocks |
| Hosts | Claude Code | Claude Code plugin, plus an MCP server for any MCP host and native plugins for Codex, Cursor, Kiro, Antigravity, Gemini CLI and OpenCode |
Where the Codex plugin fits better
- You want Codex to make the change, not only comment on it.
/codex:rescuewrites code; GPT calls in Deliberation are read-only. - You want to keep working in Codex afterwards.
/codex:transferandcodex resumecontinue the same thread in the Codex app or TUI. - You want to run long reviews in the background and check on them later.
- You want a hard stop: the review gate keeps Claude from finishing while Codex still finds issues. Watch your usage limits if you turn it on; the plugin README warns that the loop can run long.
- You only use OpenAI models and do not need other model families.
What Deliberation adds
- Reviewers from other model families: Gemini, Grok, and any OpenRouter or local model. Different families tend to flag different problems.
- A review loop with rules against self-approval. In
/deliberation:consensus, Claude records its verdict before it sees the panel. A round converges only when every responding model approves (at least two by default), no accepted critical issue is left, and Claude approves too. Claude must give a written reason for every issue it dismisses. - Expert personas (Security Analyst, Plan Reviewer, Writing Editor and five more) that apply to any provider.
- Hosts other than Claude Code.
They share one Codex login
Both plugins run your local codex CLI with its current login. If codex login works for one,
it works for the other. The Codex plugin talks to the Codex app server; Deliberation runs
codex exec --sandbox read-only for each GPT call.
Model settings are separate. The Codex plugin reads ~/.codex/config.toml and a trusted
project's .codex/config.toml. Deliberation reads the same ~/.codex/config.toml unless you pin
providers.codex.model and providers.codex.reasoningEffort in
~/.config/deliberation/config.json. The pinned values then apply to Deliberation's calls only.
Pin them when, for example, you want a cheaper model for Deliberation's quick second opinions and a stronger
one for Codex tasks.
Calls from both plugins count against the same Codex usage limits.
Install both
/plugin marketplace add openai/codex-plugin-cc
/plugin install codex@openai-codex
/plugin marketplace add antonbabenko/agent-plugins
/plugin install deliberation@antonbabenko
/reload-plugins
/codex:setup
/deliberation:setupCommand names do not collide: one set starts with /codex:, the other with /deliberation:.
If you let /deliberation:setup install its short aliases (/ask-gpt, /consensus and others),
it asks before it overwrites a command with the same name.
Workflows
Plan with the panel, build with Codex
- Write the plan in Claude Code.
- Run
/deliberation:consensuson the plan with theplan-reviewerorarchitectexpert. Fix the panel issues you accept. /codex:rescueto implement a step, or implement it in Claude Code./codex:reviewon the result.
Two kinds of review before a merge
-
/codex:review --base mainfor a Codex review of the branch. -
Run the command below for answers from the other model families:
text/deliberation:ask-all Review the current diff for real bugs, security issues, and missing tests. -
Fix the issues from both reviews. Where they disagree, ask a narrower follow-up instead of picking a side.
Break a deadlock
When Codex and Claude disagree on a design, run /deliberation:consensus on the question.
Gemini, Grok and any configured OpenRouter models review the same plan, and the final report
lists every dismissed issue and why.