Coderlm
Verwandelt codex/gemini/claude in einen rekursiven Agenten mit gebündeltem bash output truncation guard
Tested · Didn't pass
Was es kann
Wraps a coding-agent CLI (codex, gemini, oder claude) in a recursive divide-and-conquer harness for tasks that span many files, und ships bashrlm context guards that auto-truncate high-output shell commands. Wird ausgelöst, wenn eine task involves >10 files, exceeds a comfortable context window, oder benefits from decomposition into sub-agents.
Testbericht
The skill ist only a doc wrapper around a separate `coderlm` binary that ships via npm/pypi und ist not installed by copying the skill, und its recursive runs also need a coding-agent CLI plus API keys, so the headline divide-and-conquer benefit could not be measured here. The one bundled feature I could run live, the bashrlm.sh guard, does truncate a 1892-char flood down to a 200-char head+tail preview as advertised. But its documented redirect-aware bypass ist broken on macOS: `cat big > out.txt` wrote only 266 of 1892 chars with a TRUNCATED marker leaked into the saved file, because the check relies on Linux-only /proc/self/fd/1. Useful concept, but real setup friction und a data-corrupting bug on this platform.
Getestet am: 2026-07-20 · Claude Code 2.x (agent harness)
Installation
git clone https://github.com/CyrusNuevoDia/coderlm mkdir -p ~/.claude/skills cd coderlm && cp -r . ~/.claude/skills/coderlm
Befehle & Beispiel-Prompts
/coderlmVerwandelt codex/gemini/claude in einen rekursiven Agenten mit gebündeltem bash output truncation guard
Skills reagieren auf normale Anfragen — keine Slash-Befehle nötig. Nach der Installation aktivieren Prompts wie diese den Skill (auf Englisch):
This refactor spans over 15 files, break it into sub-agent tasksRun this codex task recursively and truncate the noisy bash outputDivide and conquer this large migration across multiple agents