rlm-orchestratorlisted

# RLM-Style Recursive Orchestrator Implement the orchestrator pattern from RLM research to handle arbitrarily large contexts and complex multi-part tasks. The main conversation acts as the recursive coordinator, spawning depth-1 subagents and aggregating results. ## Core Principle > "No single language model call should require handling a huge context." > — RLM Research (arXiv:2512.24601) Since Claude Code subagents cannot spawn children (architectural limit), the main conversation becomes the "recursion stack," enabling functional depth >1. ## When to Use This Skill **Ideal for:** - Tasks requiring >100K tokens of context - Multi-file analysis or refactoring - Research tasks with many sources - Batch processing with independent partitions - Any task showing signs of context rot (degraded recall, repeated mistakes) **Not ideal for:** - Simple single-file changes - Tasks requiring tight sequential dependencies - Quick exploratory questions ## The RLM Orchestration Pattern ``` Main Session (orchestrator/recursion stack) │ ├─[DECOMPOSE]─ Analyze task, identify independent partitions │ ├─[SPAWN BATCH 1]──┬── Subagent A (fresh 200K context) → summary │ ├── Subagent B (fresh 200K context) → summary │ └── Subagent C (fresh 200K context) → summary │ ├─[AGGREGATE]─ Combine results, identify gaps │ ├─[SPAWN BATCH 2]──┬── Subagent D (uses batch 1 results) → summary │ └── Subagent E (u