A professional orchestration layer for managing background AI coding sessions with advanced lifecycle control and worktree management.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install code-agent-orchestration
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 code-agent-orchestration 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 Code Agent Orchestration?
Code Agent Orchestration is a powerful tool within the Openclaw Skills ecosystem designed to manage Claude Code or Codex sessions as background coding jobs. It provides developers with a robust framework for launching multi-turn agent sessions, handling plan approvals, and managing complex git worktree strategies without leaving the chat context.
By leveraging this skill, developers can maintain deep control over autonomous agents. It bridges the gap between raw AI execution and human-in-the-loop oversight, ensuring that code changes are planned, reviewed, and merged according to strict safety and architectural guidelines.
Code Agent Orchestration Use Cases
- Orchestrating background refactoring tasks across large repositories.
- Implementing a 'Plan-First' workflow where AI agents must submit architectural changes for review before execution.
- Managing isolated development sandboxes using automated git worktrees to prevent workspace pollution.
- Resuming interrupted coding sessions or handling follow-up tasks without losing session context.
- Automating low-risk merges while escalating complex PR decisions to human developers.
How Code Agent Orchestration Works
- Session Launch: The orchestrator initializes an agent session using
agent_launch, defining the workspace and permission modes. - Plan Generation: In 'plan' mode, the agent generates a technical proposal which is held for review based on the defined approval strategy.
- Approval Lifecycle: Depending on the configuration (Ask, Delegate, or Approve), the plan is either auto-validated, reviewed by the orchestrator, or sent to the user.
- Execution: Once approved, the agent executes the coding tasks within the target directory or a dedicated worktree.
- Monitoring & Feedback: The orchestrator monitors real-time output and deterministic state fields to track progress and handle agent inquiries.
- Worktree Resolution: After completion, the system manages the transition from a sandbox worktree to the main branch via merge or PR commands.
Code Agent Orchestration Setup
To get started with this skill within the Openclaw Skills environment, ensure you have the OpenClaw CLI installed and then add the code agent package.
# Install the OpenClaw code agent package
npm install openclaw-code-agent
# Verify the installation
openclaw --version
Configuration is primarily handled via tool calls during the chat session, such as setting the workdir and permission_mode.
Code Agent Orchestration Data Schema & Taxonomy
The skill organizes session data and worktree states using a deterministic metadata schema. Key status fields include:
| Field | Description |
|---|---|
requestedPermissionMode |
The initial permission gate requested (e.g., plan, bypass). |
approvalExecutionState |
Tracks if the session is awaiting_approval or approved_then_implemented. |
worktree_status |
Monitors sandbox lifecycle states like pending_decision, merged, or released. |
session_id |
Unique kebab-case identifier for resuming multi-turn interactions. |
Code Agent Orchestration Advanced Features
- Delegated Approval: Allows the orchestrator to programmatically approve low-risk plans while providing structured rationales.
- Worktree Isolation: Support for
delegateandaskworktree strategies to keep experimental code in temporary task sandboxes. - State-Based Monitoring: Utilizes deterministic fields instead of fragile transcript parsing to determine agent completion and success.
- Safe Cleanup: Advanced
preview_safeandclean_safemodes to prune stale worktrees without losing active PRs or dirty branches. - Multi-Turn Continuity: Native support for
resume_session_idto ensure long-running tasks maintain full context over multiple interactions.
SKILL.md
Loading
Related Openclaw Skills
Threads Automation Suite (threads-skills)
A professional automation toolkit for the Threads platform powered by Chrome CDP and Python.
MX Stocks Screener
An AI-powered financial screening tool that uses natural language to filter global stocks, funds, and bonds using the Eastmoney Miaoxiang database.
mx-macro-data
A natural language interface for retrieving structured macroeconomic and commodity price data from the Eastmoney database.
Financial Information Search (mx-finance-search)
A natural language search tool providing real-time access to global financial announcements, research reports, and market news via the Eastmoney database.
Soul Generator
A sophisticated tool for generating custom AI persona configurations through celebrity distillation, personality typing, or manual descriptions.
ClawVitals
ClawVitals is a security health check skill that scans Openclaw installations to identify vulnerabilities and provide a scored safety report.
METADATA
Featured*
