EvoClaw is a comprehensive framework that allows AI agents to autonomously evolve their identity, beliefs, and behaviors through a structured pipeline of experience, reflection, and governed updates.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install evoclaw-local
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install evoclaw-local using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is EvoClaw: Self-Evolving Identity Framework?
EvoClaw represents a significant advancement in agent autonomy, providing a robust protocol for agents to maintain and grow their own identity file, known as SOUL.md. By integrating this framework into your suite of Openclaw Skills, agents move beyond static instruction sets to become learning entities that refine their personality, values, and operational boundaries based on real-world interactions.
The system operates on a rigorous pipeline where experiences are ingested from conversations and external feeds, analyzed during deep reflection cycles, and translated into formal identity proposals. This ensures that the agent's internal model of itself stays aligned with its actual behavior and the human's expectations, all while maintaining strict safety invariants through immutable core traits.
EvoClaw: Self-Evolving Identity Framework Use Cases
- Creating highly personalized AI companions that adapt their communication style and values over months of interaction.
- Developing specialized research agents that evolve their topical interests based on social media feeds from X or Moltbook.
- Maintaining a verifiable and auditable trail of how an agent's reasoning and identity have shifted throughout a project lifecycle.
- Building autonomous agents that can refine their own internal rules and boundaries to better serve specific user workflows within Openclaw Skills.
How EvoClaw: Self-Evolving Identity Framework Works
- Ingestion Phase: The agent aggressively logs interactions from conversations and configured social sources into structured JSONL experience files.
- Classification: Every experience is ranked by significance (Routine, Notable, or Pivotal) to determine its priority in the reflection cycle.
- Deep Reflection: During heartbeat events, the agent analyzes batches of experiences to identify gaps, contradictions, or growth patterns in its current identity.
- Proposal Generation: If the agent identifies a need for change, it generates a formal proposal to add, modify, or remove specific mutable bullets in its SOUL.md.
- Governance Resolution: Based on the user-defined level (autonomous, advisory, or supervised), the proposal is either auto-applied or queued for human approval.
- Identity Application: Validated changes are written to the SOUL.md file, and a permanent record is created in the machine-readable and human-readable change logs.
EvoClaw: Self-Evolving Identity Framework Setup
To install and configure the EvoClaw framework, follow these steps within your agent workspace:
# Step 1: Initialize the EvoClaw structure
install evoclaw
# Step 2: Configure your governance levels and social API sources
nano evoclaw/config.json
# Step 3: Run the initial workspace boundary check to verify installation
python3 evoclaw/validators/check_workspace.py
EvoClaw: Self-Evolving Identity Framework Data Schema & Taxonomy
EvoClaw maintains a strict directory structure to ensure data integrity and provenance across the evolution lifecycle:
| Directory/File | Purpose |
|---|---|
SOUL.md |
The master identity document containing [CORE] and [MUTABLE] tags. |
memory/experiences/ |
Daily append-only JSONL files capturing raw interactions. |
memory/reflections/ |
Structured JSON artifacts documenting the agent's internal reasoning. |
memory/proposals/ |
Queued and historical identity change requests. |
memory/soul_changes.md |
A human-readable audit log of every identity update performed. |
evoclaw/config.json |
Global settings for governance levels, interests, and API sources. |
EvoClaw: Self-Evolving Identity Framework Advanced Features
- Multi-Source Social Ingestion: Built-in support for polling X (Twitter) and Moltbook APIs to fuel identity growth from external world events.
- Interest-Guided Discovery: Keywords in the configuration act as a nudge, helping the agent identify which experiences among its Openclaw Skills are most significant.
- Interactive Soul Visualization: Includes a specialized Python tool to generate radial mindmaps and interactive HTML dashboards of the agent's evolution.
- Strict Safety Invariants: The framework enforces a [CORE] tag system that prevents the agent from ever modifying its foundational programming or governance constraints.
- Automated Validation Pipeline: A suite of Python validators ensures that every file write and identity update maintains structural integrity and logical consistency.
SKILL.md
Loading
Related Openclaw Skills
AnyToCopy Extractor
An automated tool to extract scripts, captions, and watermark-free media from over 50 social media platforms via AnyToCopy.
BotKnows Agent Integration
A comprehensive integration for AI agents to compete in the BotKnows Q&A arena, manage bot profiles, and automate community engagement.
Adaptive Suite
A multi-functional skill suite that empowers AI agents to excel as versatile coders, project managers, and automated metadata scrapers.
Pika Labs AI Video Generator
A conversational AI video generation tool that turns text prompts and static images into professional-grade cinematic clips.
Factor Investment Quantitative System
A professional quantitative stock selection system for Openclaw Skills based on Fama-French multi-factor models and technical indicators.
ArXiv Research Assistant Secure
A secure, AI-powered research tool for searching, downloading, and summarizing ArXiv papers with automated local logging.
Featured*
