Self Evolution

README

# OpenClaw Self Evolution

OpenClaw Self Evolution turns workspace self-learning sources into reviewable execution proposals and promoted rule layers:

- `workspace/.learnings/*.md`

It does not modify OpenClaw core. It only:

- parses structured learnings/errors/feature requests
- rebuilds low-noise priors/proposals files
- promotes reusable rules into an injected rules layer
- exposes status/refresh tools
- injects promoted rules and priors before agent execution

It does not own instinct formation. Instinct / habit rollout belongs to `memory-enhanced`.

## Tools

- `self_evolution_status`
- `self_evolution_refresh`
- `self_evolution_proposals`
- `self_evolution_rules`

## Outputs

- `~/.openclaw/workspace/.openclaw/self-evolution-priors.md`
- `~/.openclaw/workspace/.openclaw/self-evolution-status.json`
- `~/.openclaw/workspace/.openclaw/self-evolution-proposals.json`
- `~/.openclaw/workspace/.openclaw/rules/rule-candidates.json`
- `~/.openclaw/workspace/.openclaw/rules/promoted-rules.json`
- `~/.openclaw/workspace/.openclaw/rules/promoted-rules.md`