subagent-driven-development

What This Skill Does

Subagent-Driven Development is an advanced architectural pattern for the OpenClaw AI agent that transforms complex project execution into a reliable, automated assembly line. By dispatching a fresh, specialized subagent for every discrete task in an implementation plan, the system effectively mitigates the risk of 'context pollution'—the common AI failure mode where previous task state or accumulated hallucinations interfere with current objective precision.

This skill enforces a rigorous two-stage review process: first, a Spec Reviewer verifies that the implementation strictly adheres to the requested functional requirements, and second, a Code Quality Reviewer assesses the codebase against industry best practices. This layered approach ensures that code is not just functional, but also maintainable and clean before the agent proceeds to the next item in the backlog.

Installation

You can install the skill directly via your terminal using the OpenClaw/Clawbot package manager. Ensure your environment has the necessary permissions to spawn sub-processes:

npx clawhub@latest install subagent-driven-development

Use Cases

This skill is optimized for structured, predictable development environments. It is best utilized when you have already defined a clear roadmap or technical specification.

1. Refactoring Legacy Modules: When you have a list of decoupled functions that need migration to a new framework.
2. Feature Expansion: When implementing multiple independent API endpoints or CRUD interfaces defined by an existing OpenAPI spec.
3. Test Suite Generation: When generating unit tests for an existing codebase where each module is isolated.

Example Prompts

1. "Execute the 'user-auth-implementation' plan using the subagent-driven-development skill. Start by parsing the plan in docs/roadmap.md and ensure every unit test passes before marking each task as complete."
2. "I have an implementation plan for adding five new validator functions to the engine. Use subagent-driven-development to implement them, ensuring that the spec reviewer checks for edge-case coverage on every single one."
3. "Follow the subagent-driven-development process for the migration plan. Don't worry about existing debt, just focus on the task list provided and ensure the code quality reviewer enforces the project's Airbnb style guide."

Tips & Limitations

• Context management: While the subagent is 'fresh,' the skill still consumes shared environment context. Ensure your README and architecture.md files are up to date.
• Task Granularity: This skill works best when tasks are atomic. If a task requires more than 50 lines of complex logic, consider decomposing it further before starting the subagent process.
• Bottlenecks: Because this skill enforces a two-stage review, initial progress may seem slower than manual prompting. However, the reduction in debug-cycles usually results in a faster total completion time.
• Interaction: Avoid manual intervention while the agent is in the 'dispatch' loop; let the review cycle conclude before attempting to steer the agent, as the internal feedback loop is designed to resolve compliance issues automatically.