A comprehensive framework for designing, configuring, and scaling custom sub-agents within the Claude Code ecosystem.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install agent-development
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 agent-development 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 Development for Claude Code?
This skill provides a technical foundation for engineering high-performance custom agents that extend the capabilities of Claude Code. It focuses on moving beyond simple command execution toward sophisticated multi-agent orchestration. By leveraging Openclaw Skills for agent development, developers can create specialized sub-agents with precise tool access, clear delegation triggers, and robust self-documenting prompts.
The framework addresses critical engineering challenges such as memory management for long-running sessions, the trade-offs between remote API calls and local sub-agents, and the transition from imperative instructions to declarative goal-setting. It ensures that agents remain autonomous, context-aware, and capable of performing repetitive tasks that require significant technical judgment.
Agent Development for Claude Code Use Cases
- Creating custom agents that automatically trigger based on specific keywords or file types.
- Scaling code audits across large repositories using parallel sub-agent workflows.
- Resolving memory crashes in Claude Code by optimizing the Node.js execution environment.
- Designing automated agent pipelines where tasks are sequentially handed off between specialized roles.
How Agent Development for Claude Code Works
- Define the agent role using a strong trigger pattern in the YAML frontmatter to enable reliable auto-delegation by the primary model.
- Configure specific tool access, such as restricting Bash usage in favor of specialized Read/Write tools to minimize manual approval prompts.
- Optimize the local environment by increasing the Node.js heap size to support larger context windows and longer agent life-cycles.
- Write declarative prompts that define the desired outcome and success criteria rather than micromanaging tool invocations.
- Implement a continuous improvement loop where every discovered bug or requirement is encoded directly into the agent prompt to ensure future sessions maintain high quality.
Agent Development for Claude Code Setup
To prevent memory issues during heavy agent usage, update your shell configuration:
export NODE_OPTIONS="--max-old-space-size=16384"
source ~/.bashrc
Configure your .claude/settings.json to allowlist common commands and streamline the Openclaw Skills workflow:
{
"permissions": {
"allow": [
"Write", "Edit", "Bash(ls *)", "Bash(grep *)"
]
}
}
Agent Development for Claude Code Data Schema & Taxonomy
Agents are defined in the .claude/agents/ directory using Markdown files with YAML frontmatter. The structure follows this schema:
| Attribute | Description |
|---|---|
| name | The unique identifier used for explicit delegation. |
| description | The criteria used by Claude to decide when to invoke the agent. |
| tools | A list of capabilities granted to the agent (e.g., Read, Write, Bash). |
| model | The specific Claude model assigned to the agent (Sonnet is recommended for most tasks). |
| prompt | The Markdown-formatted instructions defining the agent's process and output formats. |
Agent Development for Claude Code Advanced Features
- Strong Trigger Patterns: Utilize specific phrasing like "MUST BE USED when" to force proactive agent delegation.
- Command Permission Allowlists: Define safe Bash commands in the settings to reduce interactive approval fatigue.
- Pipeline Sequencing: Establish numbered workflows where predecessor agents explicitly point to the next sub-agent in a chain.
- Parallel Execution: Launch multiple identical agents with segmented item lists to process large datasets across independent context windows.
- Self-Documentation: Use structured sections in prompts like "Common Issues" and "Quality Checklists" to capture and persist technical learnings.
SKILL.md
Loading
Related Openclaw Skills
Web Accessibility (WCAG 2.1 AA)
A comprehensive developer framework for building WCAG 2.1 AA compliant websites using semantic HTML, ARIA, and keyboard-first navigation.
Job Auto-Apply Skill
Automate job searches and application submissions across multiple platforms like LinkedIn and Indeed using Clawdbot.
TokenWatch
TokenWatch is a comprehensive spending monitor that helps developers track and optimize AI token costs across multiple providers in real-time.
Prompt Performance Tester
A model-agnostic benchmarking tool that compares performance, cost, and quality across nine major AI providers simultaneously.
Vercel AI SDK Core
A production-grade toolkit for building backend AI applications with the latest Vercel AI SDK v6 stable features, including the new Output API and agent abstractions.
AI SDK UI - Frontend React Hooks
A professional toolkit for building high-performance AI chat interfaces and managing complex UI states using Vercel AI SDK v6 React hooks.
Featured*
