cdx

Full-surface Codex CLI integration for Claude Code.

Skill-first: say "spawn codex on the failing tests", "get a codex second opinion on this diff", or "have codex fix this in the background" and Claude picks the right flags, runs codex exec, parses structured output, and acts on it — verifies findings against the repo, applies and tests patches, resumes threads. A real two-model loop, replacing the official openai-codex plugin's locked-down verbatim forwarder.

Verified against codex-cli 0.144.0. Requires Node >= 20 and a logged-in Codex CLI (codex login).

Install

claude plugin marketplace add nuko-nova-dynamics/cdx
claude plugin install cdx@cdx

Or from a local clone: claude plugin marketplace add /path/to/cdx.

Philosophy

• Natural language is the interface. The driving-codex skill triggers on any mention of Codex; slash commands are optional aliases.
• Claude chooses the flags. Sandbox, model, effort, web search, images, schemas — selected per task by documented heuristics, never asked of you.
• Structured output over prose relay. Reviews and reports come back as JSON (--output-schema) that Claude parses, verifies, and acts on.
• Codex owns its state. Sessions live in ~/.codex; cdx keeps no job registry, no state files, no companion daemon. One small runner script tames the JSONL event stream; everything else is the CLI itself.

Commands

Command	Purpose
/cdx:task	Delegate with full flag control: [--bg|--wait] [--model m|spark] [--effort ...] [--sandbox ro|write|full] [--search] [--image f] [--schema name|file] [--resume [id]|--fresh] [--local] [-c k=v] <prompt>
/cdx:review	Structured review Claude verifies finding-by-finding, or --native for Codex's built-in reviewer (--uncommitted|--base <ref>|--commit <sha>)
/cdx:fleet	2–4 parallel Codex workers: decomposed subtasks, multi-angle opinions, A/B implementations
/cdx:session	list, resume <id|--last>, fork <id>, apply <task_id>
/cdx:cloud	Codex Cloud: exec, list, status, diff, apply
/cdx:setup	Doctor: install, auth, version drift vs the verified flag map, config pitfalls, live smoke

Skills

• driving-codex — the core: invocation contract, flag heuristics, session management, fleet fan-out, failure signatures. Reference: skills/driving-codex/references/flag-map.md (stamped with the verified CLI version).
• codex-structured-output — bundled schema library (review-findings, verdict, task-report, patch-plan) + rules for ad-hoc schemas.
• prompting-codex — GPT-5.x-Codex prompt contracts, adapted from openai/codex-plugin-cc (Apache-2.0, see NOTICE).

Safety model

• codex exec is inherently non-interactive (no approval flag exists); the runner therefore requires an explicit sandbox: ro (read-only) for review/research, write (workspace-write) for fixes, full (danger-full-access) only on explicit request with one confirmation per session.
• The --dangerously-bypass-* flags are never used, under any circumstances.

Testing

node --test tests/*.test.mjs   # unit tests (stub codex, free)
bash tests/smoke.sh            # real Codex, ~5 cheap spark calls

License

Apache-2.0. Portions adapted from openai/codex-plugin-cc — see NOTICE.