plan-task

When the user asks you to implement something complex:

1. First check if it's complex enough:

   • Affects 5+ files? → Use planning
   • Multiple sub-tasks? → Use planning
   • Architectural decisions? → Use planning

2. Use mcp__snipara__rlm_plan with:

   • query: The task description
   • max_tokens: 16000 for comprehensive plans
   • strategy: "relevance_first" (default) or "depth_first"

3. Or use mcp__snipara__rlm_decompose to break into chunks:

   • query: The task
   • max_depth: 2 (recommended)

4. Then query context for each sub-task as you implement

Example workflow:

1. User: "Implement OAuth integration"
2. You: rlm_plan("Implement OAuth integration", max_tokens=16000)
3. Result: Sub-queries with execution order
4. You: Implement each sub-query with rlm_context_query