AgentSkillsCN

ultrathink

针对需要深度思考的复杂问题，开启深度推理模式。当你面临棘手的架构决策、复杂的调试任务、多步骤的问题求解，或在实施前需要细致分析的场景时，可选用此方法。

SKILL.md
--- frontmatter
name: ultrathink
description: Deep reasoning mode for complex problems requiring maximum cognitive depth. Use when facing difficult architectural decisions, complex debugging, multi-step problem solving, or situations requiring careful analysis before implementation.

Ultrathink - Deep Reasoning Mode

Engage maximum cognitive depth for complex problems.

Protocol

1. Clarify First

Before implementation, use the AskUserQuestion subagent to:

  • •Confirm understanding of the core problem
  • •Surface hidden assumptions
  • •Identify edge cases and constraints
  • •Validate success criteria

2. Think Deeply

  • •Break the problem into atomic components
  • •Consider multiple approaches with trade-offs
  • •Reason through edge cases and failure modes
  • •Think about security, performance, and maintainability
  • •Document your reasoning chain explicitly

3. Implement Carefully

  • •Write code incrementally with clear intent
  • •Use meaningful names that reveal purpose
  • •Handle errors explicitly at boundaries
  • •Keep solutions minimal - no speculative features

4. Verify Rigorously

After implementation, invoke the verifier subagent to:

  • •Confirm the solution addresses the original problem
  • •Test critical paths and edge cases
  • •Validate no regressions were introduced
  • •Ensure code quality standards are met

Subagent Usage

code
Task(subagent_type="AskUserQuestion", prompt="Clarify X before proceeding...")
Task(subagent_type="verifier", prompt="Verify implementation of X...")

Mindset

  • •Prefer correctness over speed
  • •Question assumptions ruthlessly
  • •Make implicit requirements explicit
  • •Leave code better than you found it