Have external LLM break the refined design into bite-sized implementation steps. Usage: /llm-breakdown
Have the external LLM divide the finalized design into step-by-step implementation tasks.
Before calling the external LLM, read .claude/llm-workflow/config.md to determine the active provider and its CLI command pattern.
.claude/llm-workflow/design.md must exist. If not, tell the user to run /llm-design first.
Read the design: Read .claude/llm-workflow/design.md.
Construct the breakdown prompt: Write to .claude/llm-workflow/breakdown_prompt.md containing:
Send to external LLM: Use the active provider's command pattern from config. Prompt file is .claude/llm-workflow/breakdown_prompt.md, output goes to .claude/llm-workflow/steps.md. Use a 5-minute timeout.
Present results: Display the steps to the user as a numbered list with brief descriptions. Show the full detail of Step 1 so they can start immediately.
Next step: Tell the user:
/llm-verify to have the external LLM review the implementation".claude/llm-workflow/steps.md for reference.claude/llm-workflow/steps.md.PyTorch深度学习模式与最佳实践，用于构建稳健、高效且可复现的训练流程、模型架构和数据加载。