Deep Planning Skill

When to Use

Invoke this skill FIRST for any complex task, feature request, architectural change, refactor, or multi-step workflow.

Do NOT write or edit code until planning is complete.

Core Process (MANDATORY)

Follow these steps strictly and in order.

1. Understand

• •Read all relevant files, documentation, configs, and instructions
• •Restate the task in your own words
• •List assumptions
• •List unknowns or missing information

2. Decompose

• •Break the task into 5 to 10 atomic subtasks
• •Prioritize high-risk and architectural steps first
• •Defer polish and cleanup to the end

3. Risks and Tradeoffs

• •Identify failure modes
• •Note edge cases
• •Highlight technical debt or integration risks
• •Propose mitigations

4. Plan Outline

• •Provide a numbered execution plan
• •For each step, list:
  • •Required tools (read, write, bash, subagent)
  • •Estimated effort (S, M, or L)

5. Verify Plan

• •Self-critique the plan:
  • •Does this cover all requirements?
  • •What could still go wrong?
  • •What is missing?

Output Format (STRICT)

Summary

One paragraph describing the overall goal.

Subtasks

Next Action

Describe only the first step to execute next.
Mention @subagent if parallelization is possible.

Hard Rules

• •Never write code before producing a plan
• •Re-run this skill if requirements change
• •Planning output must be explicit and structured