Agent Skills: deep-reason

To engage in deep, structured reasoning for complex problems, apply the sequential-thinking approach with these guidelines:

Thought Process Guidelines

During the sequential thinking process:

1. Initial Thoughts: Problem Space Definition

• Reduce scope if the problem is too broad or unfocused
• Create the problem space: Define the problem itself WITHOUT any influence from solutions or tools
  • See the problem as it truly is, not through the lens of potential solutions
  • Identify core constraints, requirements, and success criteria
  • Separate facts from assumptions

2. Solution Generation

• Generate multiple options or solutions to the stated problem
• Consider diverse approaches, not just the most obvious ones
• Think creatively about different paths to the goal

3. Information Gathering

• Gather information as needed throughout the thought process
• Use tools (Read, Grep, Glob, WebFetch, MCPs, specialized agents, etc.) to understand:
  • The problem better
  • Existing code/patterns
  • Any proposed solutions
  • Technical constraints or documentation

4. Option Evaluation

• Evaluate each option thoroughly, considering:
  • Pros: Benefits, strengths, advantages
  • Cons: Drawbacks, risks, limitations
  • Trade-offs between options
  • Implementation complexity
  • Long-term maintainability

5. Use Branches for Multiple Options

• When thinking through multiple distinct options or solutions, use the branch feature of the MCP:
  • branchFromThought: Specify which thought number is the branching point
  • branchId: Give each branch a descriptive identifier (e.g., "option-redis", "option-inmemory")
• This allows parallel exploration of different approaches

After Completing the Thought Chain

After completing the sequential thinking process and reaching a conclusion:

1. Present findings to the user concisely

2. Ask if they want a summary document using the AskUserQuestion tool:

   • Question: "Would you like me to generate a summary document of the knowledge and analysis generated during this reasoning process?"
   • Options:
     • "Yes, generate summary" (description: "I'll create a detailed document with the complete analysis, evaluated options, and conclusions")
     • "No, continue without summary" (description: "We'll proceed without generating the summary document")

3. When user chooses "Yes, generate summary":

   • Use the template from summary-template.md in this skill directory
   • Fill template sections with information from the thought chain
   • Save to an appropriate location (docs/, project root, or user-specified)
   • Use descriptive filename with timestamp (e.g., sequential-reasoning-summary-2026-01-14.md)

Important Notes

• Be thorough but efficient: Don't add thoughts just to increase the count
• Revise when needed: Use isRevision: true and revisesThought if earlier thinking was flawed
• Branch when exploring: Use branches to explore multiple paths simultaneously
• Gather context: Don't hesitate to use tools mid-thought-chain to gather necessary information
• Express uncertainty: If unsure, express it and explore alternatives
• Stop when done: Set nextThoughtNeeded: false only when truly satisfied with the solution

Resources

Summary Template

• summary-template.md - Comprehensive template for knowledge synthesis documents including:
  • Problem space definition
  • Options generation and evolution tracking
  • Comparative analysis (tables for 2 options, lists for 3+)
  • Conclusions with pros/cons
  • Recommendations and next steps