self-improvement - Mulch Self Improver: Capture Agent Learnings Across Sessions
Captures agent learnings via Mulch so expertise compounds across sessions by recording failures, conventions, decisions, patterns, guides, and references.
Tags
Updated: 2026-06-30Capabilities
Typical Inputs
Typical Outputs
What this skill does
- record learnings
- query stored learnings
- search learning records
- load expertise into context
- add learning domains
- set up provider hooks
- promote patterns to project memory
- generate onboarding snippets
- check learning store status
- edit existing learning records
- prune stale learning entries
- run health checks
Inputs
- Project directory with .mulch/
- Learning content (failures, conventions, decisions)
- Domain configuration
- Git repository
Outputs
- Learning records in .mulch/ (JSONL)
- Updated project memory files (CLAUDE.md, AGENTS.md)
- Telegram notifications
- Reusable skill files
Requirements
- Node.js runtime
- mulch-cli installed or available via npx
- Git repository
- OpenClaw workspace (for OpenClaw integration)