openclaw-self-evolving

🧠 Self-Evolving Agent

*Your AI agent reviews its own logs and proposes behavior improvements — weekly, automatically.*

**Stop making the same mistakes. Let your agent learn from them.**

[](https://github.com/Ramsbaby/openclaw-self-evolving/stargazers) [](https://github.com/Ramsbaby/openclaw-self-evolving/actions/workflows/ci.yml) [](LICENSE) [](#) [](#) [](#) [](#)

If this improved your agent's behavior, a ⭐ helps others find it.

[⚡ Quick Start](#-quick-start) · [🔍 What It Detects](#-what-it-detects-6-pattern-types) · [📋 Real Results](#-real-results) · [🤖 Claude Code](#-claude-code--agentsmd)

---

Quick Install

curl -fsSL https://raw.githubusercontent.com/Ramsbaby/openclaw-self-evolving/main/install.sh | bash

Then point it at your logs:

# edit config.yaml — set agents_md and logs_dir
nano ~/.local/share/openclaw-self-evolving/config.yaml

---

What Is This?

**Self-Evolving is a weekly agent improvement pipeline** — no LLM, no API calls, no cloud.

It reads your agent's session logs, finds patterns of bad behavior (retry loops, broken rules, user frustration), and surfaces exact `AGENTS.md` or `CLAUDE.md` rule changes you can approve or reject in under a minute.

**The agent doesn't change itself.** You approve every change. That's the point.

---

The Problem

AI agents make the same mistakes repeatedly. Nobody has time to manually review thousands of conversation logs. The mistakes keep accumulating, silently.

Week 1: Agent calls git directly — you correct it
Week 2: Same mistake again
Week 3: Same mistake
Week 4: Still happening — 3 weeks wasted

**Self-Evolving automates the review** — and brings you a short list of what to fix, every week.

Week 1: Agent calls git directly 4 times despite CLAUDE.md rule
Week 2: Same mistake, 3 more times
Week 3: Self-Evolving flags it → you approve the stronger rule → never happens again

---

How It Works

...