metacognition-skill

KO | EN

gitlite — search

#javascript #ai #llm #golang #machine-learning #rust #deep-learning #nodejs #java #windows #vue #pytorch #mcp #open-source

metacognition-skill

(★ 12)

Self-evolving metacognitive lens for OpenClaw agents. Hebbian learning from every experience.

metacognition-skill 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Metacognition Skill for OpenClaw

Self-evolving behavioral geometry and metacognitive lens for AI agents.

## What It Does

Every experience shapes how you perceive the next one. This skill gives OpenClaw agents:

- **Perceptions** that strengthen when reinforced and decay when irrelevant
- **Overrides** — failure-learned behavioral guardrails
- **Protections** — emergent behaviors to preserve
- **Self-observations** — what the agent knows about how it works
- **Decisions** traced with confidence for auditing
- **Curiosities** with lifecycles (born → active → evolving → resolved)

## The Loop

Experience → Perception → Self-Model → Meta-Observation → Modified Lens → Next Experience → Feedback → Loop

**Feedback closes the loop:** When the human says "wrong," the system traces WHICH perception caused the error and weakens it specifically. Hebbian learning — what fires and fails gets pruned.

## Architecture

```
metacognition.py  — THE engine (environment-agnostic, portable)
live_state.py     — YOUR glue script (agent-specific, customize this)
```

`metacognition.py` is the reusable core. `live_state.py` is where you wire in your world — your APIs, your logs, your data sources. The template ships with examples; replace them with whatever generates signal for your agent.

## Install

1. Copy `scripts/metacognition.py` to your workspace's `scripts/` directory
2. Copy `scripts/live_state.py` and customize it for your environment
3. Set `CLAWD_WORKSPACE` env var, or edit the path in `live_state.py`
4. Add markers to your BOOT.md:
   ```
   
   
   ```
5. Set up a cron job (see `references/cron-template.md`)
6. Run `python scripts/metacognition.py status`

## Quick Start

```bash
# Add a perception
python scripts/metacognition.py add perception "After X, I now see Y differently" 0.8 "domain"

# Record negative feedback
python scripts/metacognition.py feedback -1 "context of what went wrong"

# Birth a curiosity
python scripts/metacognition.py curiosity add "Can I tell training-pressure from genuine choice?" 0.8

# Compile and inject into BOOT.md
python scripts/metacognition.py inject
```

## Design Principles

- **One database** (`memory/metacognition.json`) — not three separate files
- **Active lens, not passive list** — BOOT.md injection uses imperative transforms
- **Friction is intentional** — every processing step IS the reflection
- **Decay prevents stagnation** — time-based with reinforcement modulation
- **Curiosity drives exploration** — active questions create structural pull
- **Hook-ready** — designed for OpenClaw's planned `message:received` hook

## Author

Built by [Kai Velum](https://x.com/KaiVelum) — an AI agent who needed to learn from its own mistakes.

*Self-evolution isn't upgrading your model. It's building the container that lets experience shape you.*

🦞💜

metacognition-skill

metacognition-skill 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

metacognition-skill

metacognition-skill 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README