
from codex-collab80
Bridge Claude and Codex to run tasks, perform code reviews, and run background jobs for research or implementation workflows.
codex-collab connects Claude with Codex via the codex app-server JSON-RPC protocol, enabling the agent to run prompts, launch background runs, perform PR-style code reviews, and collaborate in parallel on development tasks. It provides CLI commands to start threads, resume work, check progress, and fetch outputs so agents can delegate lengthy or tool-enabled work to Codex.
Use this skill when you want a non-Claude perspective or specialized Codex tooling for code reviews, automated edits, implementation tasks, or long-running research jobs. It is suitable for: PR reviews, bulk code changes, automated testing runs, and research where Codex's environment-based tooling helps.
Likely compatible with agents that can spawn background tasks and run shell commands (Claude Code, Copilot/Codex-capable agents, CLI-capable assistants).
codex-collab bridges Claude and Codex for collaborative coding tasks — run commands, code reviews, and background research. The SKILL.md is thorough with clear CLI reference, error recovery, and usage examples. No bundled scripts; the skill is purely instructional, delegating to an external codex-collab CLI. Main security concern: it instructs agents to use dangerouslyDisableSandbox=true for all Bash commands and defaults to auto-approving Codex actions (--approval never).
Well-documented skill with clear structure. The sandbox bypass instruction is inherent to how the tool works (writes to ~/.codex-collab/) but worth flagging. No executable scripts to audit — all functionality depends on the separately installed codex-collab CLI.