Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
What This Skill Does
Mulch is a CLI tool that records agent learnings (failures, conventions, decisions, patterns) to a git-tracked append-only store. Agents run `mulch prime` at session start to load prior expertise and `mulch record` during work to capture new knowledge. Learnings persist across sessions and are shared across teammates working in the same project.
Unlike in-session memory or ad-hoc notes, Mulch stores learnings in git-tracked JSONL files that are queryable, linkable, and visible to all agents and teammates in the same project.
When to use it
- Recording why an API call failed so future sessions skip the same mistake
- Capturing a corrected coding convention after user feedback mid-session
- Loading accumulated project expertise before starting a complex debugging task
- Documenting an architectural decision and its rationale for future reference
- Promoting a repeatedly useful pattern to shared CLAUDE.md or AGENTS.md
Example Workflow
Here's how your AI assistant might use this skill in practice.
User corrects agent: 'No, use pnpm not npm in this project'
- 1Runs `mulch search "npm"` to check for existing related records
- 2Determines no existing record covers this convention
- 3Runs `mulch record backend --type convention --content "Use pnpm not npm"` to save the learning
- 4Flags the convention as a candidate for promotion to CLAUDE.md
Convention saved to `.mulch/backend.jsonl`; future sessions load it automatically via `mulch prime`
Requirements
Accounts, API keys, or tools you or your AI assistant may need to set up while using this skill.