A professional management tool for the Agent Coding Platform (ACP) that enables unified control over multiple AI coding assistants like Claude Code and Codex.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install shenmeng-agent-harness
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 shenmeng-agent-harness 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 Agent Harness?
Agent Harness is a powerful integration for Openclaw Skills designed to simplify the lifecycle of AI-driven development. It acts as a management layer for the Agent Coding Platform (ACP), allowing developers to spawn, monitor, and steer specialized AI agents through a standardized interface. By abstracting the complexities of different AI CLI tools, it provides a cohesive environment for complex refactoring and automated bug fixing.
The skill is built for high-performance workflows, supporting thread-bound persistent sessions that maintain deep context over long-running projects. Whether you are using Claude Code for architectural shifts or Codex for rapid patches, Agent Harness ensures your AI agents are synchronized and manageable within a single professional ecosystem.
Agent Harness Use Cases
- Triggering complex code refactoring using Claude Code via Openclaw Skills.
- Deploying one-shot bug fixes with OpenAI Codex for rapid iteration.
- Orchestrating multi-agent collaboration where one agent designs an API and another implements the backend.
- Monitoring and terminating active AI coding sessions to optimize resource usage.
- Improving legacy code documentation using Gemini CLI integration.
How Agent Harness Works
- Use the agents_list command to identify available AI coding assistants configured in your environment.
- Initiate a task by calling sessions_spawn with a specific agentId, defining whether it is a persistent session or a one-shot run.
- For long-running tasks, the agent maintains a thread-bound context, allowing for iterative feedback through sessions_send.
- Manage active agents using the subagents toolset to steer their logic or terminate processes if they deviate from requirements.
- Retrieve final results or hand off tasks between agents to complete the development lifecycle.
Agent Harness Setup
To get started with Agent Harness, ensure your ACP environment is configured. Use the following commands to initialize and verify your setup within Openclaw Skills:
# List all available agents to verify installation
agents_list
# Set your default agent configuration
# Example for Claude Code
sessions_spawn --runtime acp --agentId claude-code --mode session --thread
Note: This skill utilizes a pay-per-call system (0.01 USDT via BNB Chain) for session management.
Agent Harness Data Schema & Taxonomy
Agent Harness organizes data based on session keys and agent IDs. The following table describes the primary data structure for session management:
| Parameter | Type | Description |
|---|---|---|
| sessionKey | String | Unique identifier for a persistent AI session |
| agentId | String | The specific AI model being used (e.g., claude-code, codex) |
| mode | Enum | 'session' for persistent context or 'run' for one-shot tasks |
| threadBound | Boolean | Whether the session is locked to a specific conversation thread |
| timeoutSeconds | Integer | Maximum duration before the agent process is terminated |
Agent Harness Advanced Features
- Multi-Agent Collaboration: Run several agents simultaneously with unique labels to handle different parts of a system.
- Thread-Bound Persistence: Maintains full conversation and code history for deep reasoning tasks.
- Steering Mechanism: Use the subagents steer command to provide mid-process corrections without restarting sessions.
- Attachments Support: Directly pass local files and context-specific metadata to external agents using Openclaw Skills architecture.
- One-Shot Optimization: Configure 'run' mode for high-speed, automated fixes with predefined timeouts.
SKILL.md
Loading
Related Openclaw Skills
ZLMediaKit Analyzer
A professional AI-driven diagnostic and tracking tool for the ZLMediaKit streaming media framework.
Lean Claude Code Harness
A technical framework designed to strip away complexity and build auditable, high-performance AI coding agent harnesses.
project-doc-gen
An automated documentation tool that orchestrates parallel AI subagents to scan your codebase and generate structured architectural, API, and module-level markdown files.
File Sender
The File Sender skill enables AI agents to package workspace files into ZIP archives and deliver them to users with descriptive metadata.
Polymarket News Impact Analysis
A sophisticated intelligence tool that bridges the gap between global breaking news and real-time prediction market fluctuations on Polymarket.
Polymarket Daily Anomaly Report
A comprehensive market analysis tool that identifies extreme volatility, opposing high-stakes bets, and suspicious new-wallet activity on Polymarket.
Featured*
