# Self-Improving Proactive Agent A merged OpenClaw skill that combines the best parts of **self-improving** and **proactivity**. ## What it does - learns from corrections and reflection - stores durable patterns separately from active task state - keeps momentum with proactive next moves - recovers context before asking the user to repeat themselves - uses heartbeat for useful follow-through without spamming - preserves hard boundaries for external actions and privacy ## State layout - `~/self-improving/` → durable learning - `~/proactivity/` → active execution state ## Main files - `SKILL.md` - `setup.md` - `boundaries.md` - `heartbeat-rules.md` - `learning.md` - `state.md` - `recovery.md` - `operations.md` ## Philosophy One skill, two layers: - learn better - operate better That is the whole point.