How do I install the Openai Codex skill?

Run npx skillmds@latest add dreamlab-ai/openai-codex 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 Openai Codex 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 Openai Codex?

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 Openai Codex free to use?

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

Who published Openai Codex?

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

Openai Codex by dreamlab-ai — Agent Skill

Name: Openai Codex
Author: dreamlab-ai

Openai Codex

Delegate a coding or reasoning task to OpenAI Codex (GPT-6 Astra) via MCP for a second opinion from a non-Claude model. Use when you want to cross-check a hard algorithm, design, or refactor against a different frontier model, or get an independent bug/security/performance review of a code snippet. Not for tasks Claude handles directly, chain-of-thought traces (use deepseek-reasoning), live web research (use perplexity-research), or GitHub-PR review swarms (use github-code-review).

dreamlab-ai fdb888f 4 files · 49.8 KB Updated Sep 17, 2026

File contents

OpenAI Codex Skill

Bridges OpenAI GPT-6 Astra into the Ruflo multi-agent environment as a first-class MCP tool.

Tools

Tool	Description
`codex_generate`	Generate code, solve algorithmic problems, or get architectural advice from GPT-6 Astra
`codex_review`	Submit code for bug/security/performance review by GPT-6 Astra

When Not To Use

For tasks Claude can handle directly -- only delegate to Codex when you specifically need GPT-6 Astra capabilities
For multi-step reasoning with chain-of-thought traces -- use the deepseek-reasoning skill instead
For web research or fetching live information -- use the perplexity-research or gemini-url-context skills instead
For code review on GitHub PRs with swarm coordination -- use the github-code-review skill instead

Status: reference-only, not the live bridge

The bundled mcp-server/ (this skill's codex_generate/codex_review tools, calling the raw openai SDK directly) is not registered as an MCP server anywhere — it is legacy code kept only because flake.nix (~~573-578), the repo's prefetch-hashes.sh, and the entrypoint (config/entrypoint-unified.sh ~841) prefetch it as an npm closure. It is not symlinked into `~~/.claude/skills(this skill is absent fromregistered-skills.txt) and no reconciler auto-discovers entry_point/mcp_server` frontmatter for it. Read it as documentation of a superseded approach, not as something to debug when Codex delegation fails.

The live GPT-6 Astra bridge is the consultant-codex MCP server (/opt/agentbox/mcp/consultants/package/codex/server.js, registered in skills/mcp.json), reached on demand by the MCP client — not a standing supervisord program. Its tools are consult, health, and cost_estimate. Use it directly, or via the codex-companion skill's higher-level commands.

Architecture (legacy bridge, as designed)

Would run as devuser, invoked directly over stdio MCP protocol (no pseudo-user; the openai-user/UID-1002 model is retired estate-wide)
API key injected from $OPENAI_API_KEY environment variable
Superseded by consultant-codex — see Status above

Configuration

Variable	Default	Description
`OPENAI_API_KEY`	(required)	OpenAI API key
`OPENAI_DEFAULT_MODEL`	`gpt-6-astra`	Model to use for completions

Addendum (2026-09-08): large-context reviews go through `codex exec`

This MCP bridge and the consultant-codex server take a pasted excerpt only. For reviews that must read a whole repository or a multi-document bundle (the model has a one-million-token context), use the direct CLI path documented in skills/codex-companion/SKILL.md under "Two consultation paths". Model for both paths: gpt-6-astra.

---name: openai-codexdescription: Delegate a coding or reasoning task to OpenAI Codex (GPT-6 Astra) via MCP for a second opinion from a non-Claude model. Use when you want to cross-check a hard algorithm, design, or refactor against a different frontier model, or get an independent bug/security/performance review of a code snippet. Not for tasks Claude handles directly, chain-of-thought traces (use deepseek-reasoning), live web research (use perplexity-research), or GitHub-PR review swarms (use github-code-review).---# OpenAI Codex SkillBridges OpenAI GPT-6 Astra into the Ruflo multi-agent environment as a first-class MCP tool.## Tools| Tool | Description ||------|-------------|| `codex_generate` | Generate code, solve algorithmic problems, or get architectural advice from GPT-6 Astra || `codex_review` | Submit code for bug/security/performance review by GPT-6 Astra |## When Not To Use- For tasks Claude can handle directly -- only delegate to Codex when you specifically need GPT-6 Astra capabilities- For multi-step reasoning with chain-of-thought traces -- use the deepseek-reasoning skill instead- For web research or fetching live information -- use the perplexity-research or gemini-url-context skills instead- For code review on GitHub PRs with swarm coordination -- use the github-code-review skill instead## Status: reference-only, not the live bridgeThe bundled `mcp-server/` (this skill's `codex_generate`/`codex_review` tools, calling theraw `openai` SDK directly) is **not registered as an MCP server anywhere** — it is legacycode kept only because `flake.nix` (~573-578), the repo's prefetch-hashes.sh, and theentrypoint (`config/entrypoint-unified.sh` ~841) prefetch it as an npm closure. It is notsymlinked into `~/.claude/skills` (this skill is absent from `registered-skills.txt`) and noreconciler auto-discovers `entry_point`/`mcp_server` frontmatter for it. Read it asdocumentation of a superseded approach, not as something to debug when Codex delegationfails.The live GPT-6 Astra bridge is the **`consultant-codex`** MCP server(`/opt/agentbox/mcp/consultants/package/codex/server.js`, registered in `skills/mcp.json`),reached on demand by the MCP client — not a standing supervisord program. Its tools are`consult`, `health`, and `cost_estimate`. Use it directly, or via the `codex-companion`skill's higher-level commands.## Architecture (legacy bridge, as designed)- Would run as `devuser`, invoked directly over stdio MCP protocol (no pseudo-user; the `openai-user`/UID-1002 model is retired estate-wide)- API key injected from `$OPENAI_API_KEY` environment variable- Superseded by `consultant-codex` — see Status above## Configuration| Variable | Default | Description ||----------|---------|-------------|| `OPENAI_API_KEY` | (required) | OpenAI API key || `OPENAI_DEFAULT_MODEL` | `gpt-6-astra` | Model to use for completions |## Addendum (2026-09-08): large-context reviews go through `codex exec`This MCP bridge and the `consultant-codex` server take a pasted excerpt only. For reviewsthat must read a whole repository or a multi-document bundle (the model has aone-million-token context), use the direct CLI path documented in`skills/codex-companion/SKILL.md` under "Two consultation paths". Model for both paths:`gpt-6-astra`.

Frequently asked questions