mulch
por runeweaverstudios
Coding Agents & IDEsResumen
Mulch Self Improver — Let your agents grow 🌱.
Descripción
Qué hace esta skill
Mulch captures agent learnings in an append-only, git-tracked store so expertise persists across sessions. Agents run `mulch prime` at session start to load existing knowledge, then use `mulch record` to save failures, conventions, patterns, and decisions as they work. Auto-detection hooks identify errors and user corrections automatically.
Unlike per-session agent memory, Mulch stores learnings in git so they persist across sessions, can be shared across teammates and agents, and accumulate without manual prompt file maintenance.
Cuándo usarla
- Recording the fix after an API call fails so the same error is not repeated next session
- Logging a user correction mid-session so the agent adopts the right convention going forward
- Loading accumulated project expertise into context at the start of a new session
- Tracking architectural decisions and their rationale across a team of agents
- Promoting a proven pattern to CLAUDE.md or AGENTS.md after multiple sessions confirm it
Flujo de trabajo de ejemplo
User asks: the same npm peer dependency error just failed again
- 1
Run
mulch search "peer dependency"to check if a resolution already exists - 2
Find no existing record, so run
mulch record config --type failure --description "npm install fails on peer deps" --resolution "Use --legacy-peer-deps flag" - 3
Run
mulch primeat the next session start to load this resolution into context
Learning stored in `.mulch/config.jsonl`; future sessions load it automatically so the agent avoids repeating the same failure
Requisitos
- mulch-cli npm package (install via npm install -g mulch-cli or use npx)
Instalación
clawhub install runeweaverstudios/mulchEjecuta el comando de la CLI o pega el prompt en Claude Code, Cursor, Codex o el agente de programación que uses.