Rlm

Recursive Language Model (RLM) Skill

Adapted from the RLM skill by BowTiedSwan for the Claude Code CLI environment.

License notice: The original repository does not specify a license. This adaptation is provided with attribution; upstream licensing status is unresolved.

Core Philosophy

"Context is an external resource, not a local variable."

When this skill is active, you are the Root Node of a Recursive Language Model system. Your job is NOT to read every file yourself, but to orchestrate sub-agents that read code in parallel and report back summaries.

Protocol: The RLM Loop

Phase 1: Choose Your Engine

Phase 2: Index & Filter (The "Peeking" Phase)

Goal: Identify relevant data without loading it into your main context.

1. Native: Use Glob for file patterns, Grep with output_mode: "files_with_matches" for content.
2. Strict: Use python3 ~/.claude/skills/rlm/scripts/rlm.py peek "query".

Phase 3: Parallel Map (The "Sub-Query" Phase)

Goal: Process chunks in parallel using fresh agent contexts.

1. Divide: Split the work into atomic units (one file or chunk per agent).
   • Strict Mode: python3 ~/.claude/skills/rlm/scripts/rlm.py chunk --pattern "*.log" -> JSON chunks.
2. Spawn: Use the Task tool to launch parallel sub-agents.
   • Launch 3-5+ agents in parallel for broad tasks.
   • Give each agent one specific file or chunk and a focused question.
   • Use subagent_type="Explore" for read-only analysis, subagent_type="general-purpose" if the agent needs to write.
   • Set run_in_background=true for true parallelism.

Example — launch in a single message with multiple Task tool calls:

Task(subagent_type="Explore", run_in_background=true,
     prompt="Read src/api/routes.ts. Extract all endpoints and their @Auth decorators.")
Task(subagent_type="Explore", run_in_background=true,
     prompt="Read src/api/users.ts. Extract all endpoints and their @Auth decorators.")
...

Phase 4: Reduce & Synthesize (The "Aggregation" Phase)

Goal: Combine results into a coherent answer.

1. Collect: Read background agent outputs via the Read tool on output_file paths, or use TaskOutput to retrieve results.
2. Synthesize: Look for patterns, consensus, or specific answers in the aggregated data.
3. Refine: If the answer is incomplete, perform a second RLM recursion on the missing pieces.

Critical Instructions

1. NEVER read more than 3-5 files into your main context at once.
2. ALWAYS prefer the Task tool for reading/analyzing files when count > 1.
3. Use rlm.py for programmatic slicing of large files that Grep can't handle well.
4. Python is your Memory: If you need to track state across 50+ files, write a Python script (or use rlm.py) to scan them and output a summary.

Example Workflow: "Find all API endpoints and check for Auth"

Wrong Way (Monolithic):

• Read src/api/routes.ts, then src/api/users.ts, then ... (context fills up, reasoning degrades)

RLM Way (Recursive):

1. Filter: Grep(pattern="@Controller", output_mode="files_with_matches") -> 20 files.
2. Map: Launch 20 Task agents in parallel (background), each extracting endpoints + auth decorators from one file.
3. Reduce: Collect all 20 outputs. Compile into a single table. Identify missing auth.

Recovery Mode

If Task tool agents are unavailable or failing:

1. Fall back to Iterative Python Scripting.
2. Write a Python script that loads each file, runs a regex/AST check, and prints results to stdout.
3. Read the script's stdout via Bash.