Codex Agent serves as a sophisticated project manager for the OpenAI Codex CLI, automating task execution, monitoring, and knowledge base maintenance.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install codex-agent
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install codex-agent using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Codex Agent?
Codex Agent is a specialized project management system designed to sit atop the OpenAI Codex CLI within the Openclaw Skills ecosystem. Rather than acting as a simple command relay, it functions as an autonomous project manager that understands requirements, designs execution strategies, and supervises the quality of AI-generated code. It utilizes a terminal-based interface combined with asynchronous monitoring to handle complex, multi-turn development tasks without requiring constant developer oversight.
The skill provides a robust framework for managing AI coding workflows by integrating tmux for persistent sessions and custom Python hooks for real-time notifications. By bridging the gap between high-level project goals and low-level CLI execution, Codex Agent ensures that Openclaw Skills users can maintain high standards of code quality and architectural integrity through automated reasoning and supervised iteration.
Codex Agent Use Cases
- Managing complex multi-file code generation tasks through the OpenAI Codex CLI.
- Automated monitoring of long-running coding processes with asynchronous wake-up notifications.
- Synchronizing local knowledge bases with the latest AI model features and documentation.
- Supervising AI-driven configuration changes and feature flag updates in development environments.
- Orchestrating parallel development tasks using git worktrees and isolated tmux sessions.
How Codex Agent Works
- Requirement Analysis: The agent identifies task goals, technical stacks, and acceptance criteria through developer interaction.
- Strategy Planning: It consults an internal knowledge base to select the optimal model, reasoning effort, and execution mode (exec vs. TUI).
- Prompt Engineering: Using predefined patterns, it constructs precise instructions for the underlying AI model.
- Autonomous Execution: The agent launches the task, either in the background or via an interactive tmux pane, applying auto-approval or manual approval gates as configured.
- Real-time Monitoring: A pane monitor script tracks terminal output for specific triggers or approval requests.
- Verification and Reporting: Upon completion, the agent validates the changes against the initial requirements and provides a detailed summary of modifications.
Codex Agent Setup
To integrate Codex Agent into your Openclaw Skills workflow, follow these steps:
- Install tmux on your local machine and ensure the codex CLI is accessible via your path.
- Configure your codex settings by adding the notification hook to
~/.codex/config.toml:
notify = ["python3", "<path_to_skill>/hooks/on_complete.py"]
- Set the required environment variables for the notification system:
export CODEX_AGENT_CHAT_ID="your_unique_id"
export CODEX_AGENT_NAME="main"
- Verify the setup by running a test message through the agent:
openclaw agent --agent main --message "test"
Codex Agent Data Schema & Taxonomy
Codex Agent organizes its operational context through a structured set of Markdown files to ensure accuracy within Openclaw Skills:
| Data Component | Path | Purpose |
|---|---|---|
| Feature Tracking | knowledge/features.md |
Catalog of all CLI commands and feature flags. |
| Config Schema | knowledge/config_schema.md |
Strict definitions for config.toml fields to prevent errors. |
| Prompt Patterns | knowledge/prompting_patterns.md |
Library of optimized templates for specific coding tasks. |
| Version State | state/version.txt |
Tracks local CLI version to trigger automated knowledge updates. |
| Execution Logs | /tmp/codex_exec_output.txt |
Temporary storage for background process outputs. |
Codex Agent Advanced Features
- Dual-Channel Notification System: Combines Python-based turn hooks with shell-based pane monitoring for 100% visibility.
- Intelligent Approval Modes: Toggle between Full-Auto for trusted tasks and Manual Approval for sensitive architectural changes.
- Multi-Agent Orchestration: Support for launching and monitoring parallel agents across different git worktrees.
- Automated TUI Interaction: Logic-driven keystroke injection into tmux panes for non-blocking interactive sessions.
- Knowledge Staleness Checks: Automated workflows to refresh documentation and schemas if they haven't been updated in over 7 days.
SKILL.md
Loading
Related Openclaw Skills
Feishu Folder Summary
A powerful tool for recursively scanning Feishu folders and wiki spaces to generate comprehensive, structured Markdown summary reports.
Feishu Deep Research
An automated research agent that conducts deep data gathering and generates structured reports directly in Feishu cloud space.
OpenClaw Memory Enhancer
An edge-optimized RAG memory system that provides AI agents with persistent, semantic long-term memory using minimal system resources.
Task Orchestrator
An autonomous orchestration skill for managing multi-agent development tasks using tmux, Codex, and intelligent dependency analysis.
Browser Use
A specialized browser orchestration skill that initializes execution contexts and manages web interaction for OpenClaw agents.
Google Services (gog CLI)
A CLI-driven skill for managing Google API authentication and interacting with Gmail, Drive, and Calendar services.
Featured*
