Skillget

ask-codex

agent
Guvenlik Denetimi
Uyari
Health Uyari
  • License — License: Apache-2.0
  • No description — Repository has no description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 6 GitHub stars
Code Gecti
  • Code scan — Scanned 3 files during light audit, no dangerous patterns found
Permissions Gecti
  • Permissions — No dangerous permissions requested
Purpose
This tool is an unofficial Claude Code skill that acts as a bridge, allowing Claude Code to delegate implementation and exploration tasks directly to the OpenAI Codex CLI.

Security Assessment
Overall risk: Low. The tool is designed specifically to execute shell commands (routing tasks to the Codex CLI), which is its intended function. A code scan found no dangerous patterns, no hardcoded secrets, and no dangerous permissions requested. While it does facilitate command execution and relies on network requests to the Codex API, it does not access your local sensitive data beyond standard repository files and does not obfuscate its actions.

Quality Assessment
The project is very new and has low community visibility, currently sitting at only 6 GitHub stars, which means it has not been broadly tested or vetted by a large audience. Additionally, the repository lacks a description in its metadata, though the README thoroughly explains its functionality. On the positive side, it uses a standard Apache-2.0 license and is actively maintained, with the most recent code push occurring today.

Verdict
Use with caution — the code is safe and well-intentioned, but its low community adoption means it may contain undiscovered bugs or edge cases.
README.md

Ask Codex

Ask Codex is an unofficial Claude Code skill that helps Claude Code delegate repository work to the Codex CLI. It provides routing guidance, single-turn delegation, brokered multi-turn sessions, and review-loop conventions for collaboration between Claude Code and Codex.

This project is not affiliated with or endorsed by Anthropic or OpenAI.

中文文档: README.zh-CN.md

What It Does

  • Routes implementation, exploration, verification, and review-heavy work from Claude Code to Codex.
  • Provides ask_codex.sh for one-shot Codex delegation.
  • Provides codex_broker.sh for long-running or interactive Codex sessions.
  • Includes a Windows PowerShell helper, ask_codex.ps1.
  • Keeps Claude Code responsible for user interaction, scope control, and final verification.

Requirements

  • Claude Code with skill support.
  • Codex CLI installed, authenticated, and available as codex.
  • jq.
  • Bash for scripts/ask_codex.sh and scripts/codex_broker.sh.
  • PowerShell 5.1+ for scripts/ask_codex.ps1 on Windows.

Installation

Recommended installation with the skills CLI:

npx skills add bookSiYi1412/ask-codex -g -a claude-code

Manual installation:

mkdir -p ~/.claude/skills
git clone https://github.com/bookSiYi1412/ask-codex ~/.claude/skills/ask-codex
chmod +x ~/.claude/skills/ask-codex/scripts/*.sh

Restart Claude Code if needed so the skill metadata is reloaded.

Skill Name

The skill is named ask-codex.

Claude Code should trigger it when a task involves implementation work, repository exploration, data or log inspection, broad codebase search, impact analysis, verification-heavy shell work, or explicit requests to use Codex.

Claude Code Usage Examples

Use the skill explicitly when you want Claude Code to delegate work to Codex.

Simple implementation task:

/ask-codex Implement pagination for the user list. Keep the existing filters and add tests.

Read-only repository exploration:

/ask-codex Find where tenant IDs are derived and propagated. Do not modify files. Summarize findings with file paths and command evidence.

Parallel implementation tasks:

/ask-codex Run two independent Codex tasks in parallel:
1. Write docs/code_architecture.md explaining the module structure, data flow, and evaluation design.
2. Write docs/quickstart_guide.md with step-by-step setup and experiment-running instructions.

Review both outputs after Codex finishes and report quality issues before finalizing.

Long-running guided work:

/ask-codex Use a brokered Codex session to implement the migration. Monitor progress, send follow-up instructions if needed, then review the final diff.

Tip: when a Codex task is running in the foreground inside Claude Code, press Ctrl+B to move it to the background so you can keep working while it runs.

Direct Script Usage

One-shot delegation:

~/.claude/skills/ask-codex/scripts/ask_codex.sh \
  "Add error handling to api.ts" \
  --file src/api.ts

Read-only exploration:

~/.claude/skills/ask-codex/scripts/ask_codex.sh \
  --read-only \
  "Find all call sites for the tenant ID derivation logic"

Brokered session:

bash ~/.claude/skills/ask-codex/scripts/codex_broker.sh start \
  "Implement the requested change" \
  --file src/main.ts

Check broker status:

bash ~/.claude/skills/ask-codex/scripts/codex_broker.sh status <broker_id>

Send follow-up instructions:

bash ~/.claude/skills/ask-codex/scripts/codex_broker.sh send <broker_id> \
  "Also add validation for the new input."

Runtime Files

The scripts may create local runtime artifacts:

  • .runtime/
  • .sessions/

These files can contain prompts, model responses, repository paths, and task details. They are ignored by Git and should not be published.

Verification

Run lightweight syntax checks before publishing changes:

scripts/check.sh

The check script runs Bash syntax checks and uses shellcheck when it is installed.

Origins and Attribution

This project is based on and substantially extends oil-oil/codex, a Claude Code skill for delegating coding tasks to Codex CLI.

Major additions in this repository include expanded routing guidance, brokered session support, additional failure handling, open-source project documentation, and release-oriented maintenance files. See NOTICE for attribution details.

License

Apache License 2.0. See LICENSE.

Yorumlar (0)

Sonuc bulunamadi