Evolve

Name: Evolve
Rating: 3.5 (1 reviews)
Author: cjalden

By cjalden 👁 46 views ▲ 0 votes

Self-improving OpenClaw plugin — measures, analyzes, and proposes behavioral improvements across a bot network

GitHub

Install

openclaw plugins install evolve

README

# Evolve

**Self-improving OpenClaw plugin.** Measures, analyzes, and systematically proposes behavioral improvements across a network of OpenClaw bot instances — without retraining the underlying model.

```
Bot network → measure → analyze → propose → human approval → Forge validation → deploy
                 ↑                                                                  |
                 └──────────────── recursive improvement ───────────────────────────┘
```

## What it does

- **Observes** every turn across all bots, annotating sessions with quality signals
- **Measures** bot performance: task resolution rate, maintenance overhead, reliability
- **Analyzes** patterns weekly, generates specific improvement proposals with evidence
- **Validates** approved proposals in a sandboxed Forge instance before production
- **Dashboards** — web UI showing network health, scores, and proposal queue

## What it doesn't do

- Modify any bot without human approval
- Retrain or fine-tune the underlying model
- Operate outside its defined scope
- Touch SOUL.md or core identity files

## Repository structure

```
packages/
  plugin/       OpenClaw plugin (TypeScript) — runs in-process, annotates turns
  analyzer/     Analysis engine (Python) — measures, detects patterns, generates proposals
  dashboard/    Web UI (served by plugin HTTP route)
scripts/        Deploy, forge-setup, launchd templates
config/         Network topology configuration
docs/           Architecture, getting started, configuration reference
```

## Quick start

```bash
# Install on a bot
openclaw plugins install evolve

# Configure (add to openclaw.json)
# See docs/getting-started.md

# Deploy across a network
bash scripts/deploy.sh --network my-network
```

## Roadmap

| Version | Status | Milestone |
|---------|--------|-----------|
| v0.1 | 🔧 In progress | Plugin skeleton, turn annotation, HTTP status endpoint |
| v0.2 | Planned | Metrics collection, daily measurement, basic scoreboard |
| v0.3 | Planned | Tier classification, session scoring |
| v0.4 | Planned | Analysis engine, proposals, Telegram alerts |
| v0.5 | Planned | Dashboard — Overview + BotDetail |
| v0.6 | Planned | Forge integration, shadow mode, approval flow |
| v0.7 | Planned | Proposals page, rejection tracking |
| v0.8 | Planned | Multi-bot rollout |
| v0.9 | Planned | Drift detection, cross-bot correlation |
| v1.0 | Planned | Recursive layer, full network operational |

## Design

See `docs/architecture.md` for the full design, including the Tier 1/Tier 2 
classification model, the Goodhart's Law mitigation strategy, and the Forge 
validation pipeline.

## License

MIT

tools