yuki-yamamura/instruct.md

Execute detailed implementation instructions and adjustments based on the context file. Process multiple instructions sequentially with thorough analysis and implementation.

You must verify the following file exists before proceeding:

• ./tmp/context.md - Implementation instructions and requirements

If the file is missing:

• Stop the process immediately
• Notify the user that context.md file is not found
• Do not proceed with implementation

• Read ./tmp/context.md and parse content structure only
• Identify individual instructions separated by "=====" or similar delimiters
• Extract file references and requirements (e.g., src/components/button.tsx:L42)
• Create instruction list with dependency priority order
• Do not research or analyze instructions yet - only parse and organize

• Use TodoWrite to create todos for each instruction from Step 1
• Mark first instruction as in_progress
• Begin processing loop

IMPORTANT: Execute the following cycle for the current in_progress instruction:

1. Research Phase

   • Use sub-agents to conduct thorough investigation for this instruction
   • Use Task tool with general-purpose agent for complex research
   • Search codebase for patterns, implementations, and context related to this instruction
   • Use Web search tool for external documentation and examples
   • Use Context7 MCP for library/framework usage
   • Use Serena MCP for semantic code search

2. Problem Identification Phase

   • Read referenced files and identify current state vs this instruction's requirements
   • Compare existing implementation with expected outcome from this instruction
   • Identify gaps between current codebase state and desired end state
   • Analyze what needs to be added, modified, or removed for this instruction
   • Think harder about the root cause of each gap between expectation and reality
   • Define acceptance criteria to be fulfilled after the successful implementation

3. Implementation Phase

   • Apply research findings to solve identified problems
   • Focus on clean, maintainable implementation
   • Follow patterns discovered in research phase

4. Validation Phase

   • Run format, lint, and test tools to ensure code quality
   • Fix any failures
   • Acceptance criteria is fulfilled

5. Clean up AI comments

   • Review all AI-generated comments in the implemented code
   • Remove redundant comments that simply repeat what the code already expresses
   • Remove commented-out code blocks unless they serve a specific purpose
   • Keep only comments that provide valuable context that cannot be expressed through code structure
   • Ensure code is self-explanatory through clear naming and structure rather than excessive commenting
   • Apply the principle: "Comment what the code cannot say, not simply what it does not say"

   ref: https://github.com/97-things/97-things-every-programmer-should-know/tree/master/en/thing_16

6. Commit Changes

   Ask the subagent commit for changes.

7. Loop Control

   • If more instructions remain: mark next instruction as in_progress, return to phase 1
   • If all instructions completed: proceed to Phase 4

For multiple instructions, format as:

## Instruction 1: [instruction summary]
ref: <commit_hash>
[Implementation details, decisions, and technical explanations]

## Instruction 2: [instruction summary]
ref: <commit_hash>
[Implementation details, decisions, and technical explanations]