CoderLM — Recursive LLM File Explorer

What it does

CoderLM turns a coding assistant into a recursive file-discovery LLM (RLM). When a task involves many files or the total content exceeds a comfortable context size, CoderLM discovers relevant files with shell tools, peeks strategically, and spawns sub-agents to divide-and-conquer the task. Outputs from high-volume commands are automatically truncated to avoid flooding the agent's context window.

When to use it

Use CoderLM when you need the agent to work across large repositories (>10 files), run broad codebase audits (TODOs, type errors, refactors), or perform tasks that benefit from controlled recursion and isolated sub-tasks (linting, large-scale search-and-replace, multi-file fixes).

What's included

• Scripts: none detected in the repo snapshot (has_scripts=false)
• References: none detected (has_references=false)
• Instructions: Usage and examples show a CLI style: coderlm <agent> --prompt <file> [--max-depth N] [--allowedTools TOOLS] with concrete examples for codex, claude, and gemini CLI. The skill describes how it parses tasks, discovers files, and delegates to sub-agents while applying context guards.

Compatible agents

Designed for coding-capable agents and CLIs (Codex/Copilot-style agents, Claude Code, Gemini CLI). It integrates with agents that can run shell tools and spawn sub-agents.