Create Plans

What it does

This skill generates detailed, step-by-step plans to achieve user goals by breaking them down into smaller, independently executable phases. Each plan focuses on a maximum of 2-3 tasks and is designed for high quality output. The system prioritizes "atomic" planning – many small, focused plans are preferred over fewer large ones – to avoid performance degradation in the AI model (Claude).

When to use it

• When you need a structured approach to completing complex goals.
• For tasks that can be broken down into smaller, manageable steps.
• To ensure high-quality output by avoiding overly long or complex prompts.
• When needing automated execution with human checkpoints for verification and decision making.

Key capabilities

• Atomic Planning: Breaks down large goals into small, focused plans (2-3 tasks each).
• Autonomous Execution: Plans are executed independently with fresh context to prevent quality degradation.
• Checkpoint Integration: Includes checkpoint:human-verify and checkpoint:decision types for human confirmation and architectural choices.
• Automated Deviation Handling: Automatically addresses bugs, missing critical elements, blockers, and logs enhancements.
• CLI/API Prioritization: Prefers automated execution via CLI/API tools over requiring human action whenever possible.

Example prompts

• "Create a plan for setting up a database schema."
• "Generate a plan to configure a database client."
• “Outline the steps needed to create API routes.”

Tips & gotchas

• Context is Key: The skill performs best with focused, concise instructions. Quality degrades significantly when context exceeds 50% (especially for Claude).
• Embrace Atomicity: Expect multiple small plans rather than one large plan.
• Human Checkpoints: Be prepared to review and confirm work at designated checkpoints (checkpoint:human-verify and checkpoint:decision).