Overview

Provides a local, guardrail-backed evolution loop for OpenClaw. It snapshots current status, generates and tests candidate skills, promotes successful ones, and supports rollback.

How This Skill Works

A local controller script evolvectl.sh orchestrates the workflow. Using commands like evolve plan, evolve generate <slug>, evolve test <slug>, evolve promote <slug>, and evolve rollback <slug>, it handles status snapshots, candidate creation, testing, promotion, and rollback. The EVOLVECTL environment variable can override the script location.

When to Use It

• You want to safely explore new skill ideas without destabilizing OpenClaw.
• You need to capture a snapshot of the current OpenClaw state before making changes.
• You want to generate and empirically test skill candidates before activation.
• A candidate fails during testing and you need a clean rollback.
• You want a centralized, local evolution controller that you can override with EVOLVECTL.

Quick Start

1. Step 1: evolve plan
2. Step 2: evolve generate <slug>
3. Step 3: evolve test <slug> (then evolve promote <slug> or evolve rollback <slug>)

Best Practices

• Plan with evolve plan before generating candidates to set guardrails.
• Use meaningful slugs and keep versioned candidates to track changes.
• Always snapshot status and test results before promotion.
• Verify EVOLVECTL or specify the script path to avoid unintended runs.
• Record logs and keep a clear rollback path for each candidate.

Example Use Cases

• Plan a new candidate named 'adaptive-logging', generate, test, and promote it if tests pass.
• If a candidate causes issues, run evolve rollback <slug> to revert.
• Override evolvectl.sh location by exporting EVOLVECTL=/custom/path/evolvectl.sh.
• Before making changes, run evolve plan to snapshot status.
• Promote only after automated tests verify improvements.

Frequently Asked Questions