Codex Bridge

Dispatch coding tasks to the local OpenAI Codex CLI with background execution, status polling, and answerable clarifying questions. Use when OpenClaw should...

Security Scan

The skill's files, scripts, and runtime instructions are coherent with its stated purpose of driving a local 'codex' CLI from OpenClaw and do not request unrelated credentials or hidden network endpoints.

Purpose & Capability

Name/description match the code and scripts: the bundle dispatches tasks to a local 'codex' CLI, requires python3 and codex binaries, stores state under ~/.codex-bridge, and manipulates files only within the declared task/workdir scope.

Instruction Scope

SKILL.md instructs the agent to dispatch, poll, ask/answer clarifying questions, and read results. The included bridge.py and helper scripts implement exactly those behaviors. The bridge reads/writes only under ~/.codex-bridge/tasks/<task-id>/ and launches the codex CLI in the specified workdir; it does not access unrelated system paths or external endpoints.

Install Mechanism

No install spec is provided (instruction-only with included scripts). Nothing is downloaded or written outside the skill's own task directory; no unexpected install sources are used.

Credentials

No environment variables or external credentials are required by the skill. Declared binary requirements (python3, codex) are appropriate for the functionality. Note: the codex CLI itself may require user credentials/config which are outside this skill's scope.

Persistence & Privilege

The skill does not request 'always: true', does not modify other skills or global agent settings, and keeps its state under ~/.codex-bridge. It runs background processes for dispatched tasks (normal for this purpose).

Assessment

This skill appears to do what it claims: run the local 'codex' CLI on a project directory and mediate clarifying Q&A via files under ~/.codex-bridge. Before installing, consider: 1) Ensure you trust the local 'codex' CLI and its configuration/credentials (the bridge will invoke it in the chosen workdir and the CLI could read or modify files there). 2) Run the bridge only against project directories you are willing to let the Codex CLI access; avoid pointing it at sensitive system paths or your home directory. 3) Inspect or back up any files in the chosen workdir if you are worried about automated modifications. 4) Logs and task state are stored in ~/.codex-bridge/tasks/; review or remove that directory if you uninstall. If you want additional assurance, run the scripts manually in a safe test repo and observe behavior before enabling them in production workflows.

Delegate coding tasks from OpenClaw to the local codex CLI and manage them asynchronously. This skill provides a file-based bridge so OpenClaw can:

• dispatch tasks
• poll status and recent output
• relay clarifying questions
• send answers
• collect final results

When to Use

• Build scripts (Python, Bash, etc.)
• Implement or refactor code in an existing project
• Run larger multi-file coding tasks in the background
• Delegate coding work while keeping OpenClaw responsive
• Handle tasks that may require clarifying questions mid-run

When NOT to Use

• Quick factual questions or explanations
• Small code snippets that OpenClaw can write directly
• Non-coding tasks
• Tasks that should not invoke a local coding agent/CLI

Dispatch a Task

~/.openclaw/skills/codex-bridge/codex-bridge-dispatch.sh \
  --task-id <descriptive-name> \
  --workdir <project-directory> \
  --prompt "<detailed coding task>"

Prompt Writing

Include:

• what to build/fix
• file paths if known
• expected behavior/output
• language/framework preferences
• constraints (tests, style, no new deps, etc.)

Example:

~/.openclaw/skills/codex-bridge/codex-bridge-dispatch.sh \
  --task-id scripts-csv-parser \
  --workdir ~/projects/data-tools \
  --prompt "Create parse_orders.py. Read orders CSV, keep shipped rows, group by customer_id, and write summary CSV with columns customer_id, order_count, total_amount. Use pandas. Add basic CLI args and error handling."

Check Status

~/.openclaw/skills/codex-bridge/codex-bridge-status.sh --task-id <id>

Common status commands:

~/.openclaw/skills/codex-bridge/codex-bridge-status.sh --list
~/.openclaw/skills/codex-bridge/codex-bridge-status.sh --task-id <id> --output
~/.openclaw/skills/codex-bridge/codex-bridge-status.sh --task-id <id> --question
~/.openclaw/skills/codex-bridge/codex-bridge-status.sh --task-id <id> --result
~/.openclaw/skills/codex-bridge/codex-bridge-status.sh --task-id <id> --log

Answer Clarifying Questions

When status is waiting_for_answer, read the pending question and send a response:

~/.openclaw/skills/codex-bridge/codex-bridge-status.sh --task-id <id> --question
~/.openclaw/skills/codex-bridge/codex-bridge-answer.sh --task-id <id> --answer "<answer text>"

The bridge resumes the same Codex session after the answer is written.

Workflow

1. Dispatch task with a clear prompt.
2. Report the task ID.
3. Poll status/output periodically.
4. If status becomes waiting_for_answer, read --question, relay to user, then write answer with --answer.
5. When status is complete, read --result and summarize outcomes.
6. If status is error, inspect --log and --output.

Notes and Limits

• Uses the local codex CLI (codex exec and codex exec resume).
• Clarifying questions are implemented via a lightweight text marker protocol in the prompt wrapper.
• Bridge state is stored in ~/.codex-bridge/tasks/<task-id>/.
• Commands run via Codex in the specified --workdir.
• For unattended runs, the bridge times out after 10 minutes waiting for an answer and resumes with a default/sensible fallback.