OpenAI Codex Operator

Run OpenAI Codex CLI from OpenClaw for coding tasks in a target project directory. Use when the user asks OpenClaw to use Codex for implementation, debugging...

Security Scan

This skill is internally consistent: it simply documents and automates invoking the OpenAI Codex CLI against a target repository and does not request unrelated credentials, installs, or elevated persistence.

Purpose & Capability

Name/description match the contents: SKILL.md, references, and example script all describe running the Codex CLI (codex / codex exec) in a specified workdir. No unrelated binaries, env vars, or config paths are requested.

Instruction Scope

Instructions stay within the stated purpose (running Codex in a target repo, foreground/background modes, polling). They do not instruct reading arbitrary system files or unrelated credentials. Note: the skill does not provide guidance about avoiding sending secrets to Codex — the CLI may transmit repository contents to OpenAI during its operation, so users should be warned to scrub sensitive data before running.

Install Mechanism

No install spec is provided (instruction-only), and the included script only instructs installing the official npm package (@openai/codex) if codex is missing. There are no downloads from untrusted URLs or extract actions.

Credentials

The skill declares no required environment variables or credentials. It does mention Codex/CLI authentication in references (ChatGPT login or API key), which is expected for this integration and is not requested by the skill itself.

Persistence & Privilege

The skill does not request always:true, does not modify other skills or global agent settings, and has no install-time persistence. Autonomous invocation is allowed (platform default) but is not combined with other concerning privileges.

Assessment

This skill appears to do only what it claims: run the local Codex CLI against a repository. Before installing or invoking it, ensure you understand that the Codex CLI will contact OpenAI and may send repository contents (including secrets or PII) to OpenAI; remove or redact sensitive files or use a sanitized checkout. Also confirm you trust the locally installed `@openai/codex` package and its authentication mechanism (ChatGPT login or API key) before granting access.

Use this skill to reliably call Codex CLI from OpenClaw.

Core rules

1. Verify Codex CLI exists (codex --version) before first task.
2. Always run Codex through OpenClaw exec with pty:true.
3. Always set explicit workdir to the target repository.
4. For long tasks, use background:true and track via process.
5. Report clear milestones: started, waiting-input, finished/failed.

Execution patterns

One-shot coding task

Use:

• exec.command: codex exec "<task>"
• exec.pty: true
• exec.workdir: <repo path>

Interactive session

Use:

• exec.command: codex
• exec.pty: true
• exec.workdir: <repo path>

Long-running background task

1. Start with exec(background:true, pty:true, workdir, command:"codex exec ...")
2. Record returned sessionId.
3. Poll with process action:poll.
4. Read output with process action:log.
5. If Codex asks for input, use process action:submit.

Recommended prompts

• "Implement <feature> with tests, run tests, and summarize changed files."
• "Find root cause for failing CI in this repo and propose minimal fix."
• "Review current branch diff and list high-risk issues first."

Guardrails

• Do not claim files were changed unless logs show completion.
• If codex is missing or auth fails, return exact remediation steps.
• Keep OpenClaw tool config (pty/workdir/background) separate from CLI args.

References

• references/codex-doc-summary.md
• references/codex-usage-recipes.md
• scripts/run-codex-example.sh