A meta-analytical automation tool that observes AI agent performance and suggests optimized instruction updates based on semantic log patterns and real-time monitoring.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install self-evolving-agent
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 self-evolving-agent 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 Self-Evolving Agent?
The Self-Evolving Agent is a sophisticated meta-layer designed to improve the performance of Openclaw Skills by analyzing interaction logs across multiple sessions. By utilizing semantic embeddings via local Ollama instances, it identifies structural patterns, recurring errors, and user frustrations that traditional keyword matching often misses. The system operates with a high degree of transparency and safety, ensuring that while it suggests optimizations for the core AGENTS.md configuration, no changes are ever applied to your environment without explicit human approval.
In its version 5.0 release, this skill introduces real-time streaming monitoring and fleet analysis capabilities, allowing for cross-instance pattern detection in multi-agent environments. It transforms raw log data into actionable intelligence, providing a feedback loop that measures the actual effectiveness of previous improvements to ensure the agent's behavior is constantly trending toward higher quality and reliability.
Self-Evolving Agent Use Cases
- Identifying recurring infinite loops or execution retry failures in complex coding tasks.
- Detecting semantic patterns of user frustration across different languages using local embeddings instead of hardcoded keywords.
- Analyzing performance trends across a fleet of multiple agent instances to identify systemic instruction weaknesses.
- Automatically generating GitHub Issues for proposed system improvements to maintain a structured development roadmap.
How Self-Evolving Agent Works
- The system collects logs from various sessions, structured execution events, and cron job results into a structured JSON format.
- Semantic analysis is performed using Ollama nomic-embed-text to identify patterns and complaints with a low false-positive rate.
- A trend analyzer compares current data against a four-week history to categorize issues as emerging, resolved, or stable.
- Cross-instance fleet analysis identifies common failure points and successes across different agent models and environments.
- The benchmark stage measures the frequency of specific failure patterns before and after previous proposals were applied to verify effectiveness.
- A synthesis agent generates a detailed proposal with Markdown diffs, delivering it via Discord, Slack, or Telegram for user approval.
Self-Evolving Agent Setup
Install the skill by cloning the repository and running the interactive setup wizard:
git clone https://github.com/Ramsbaby/self-evolving-agent ~/openclaw/skills/self-evolving-agent
bash ~/openclaw/skills/self-evolving-agent/scripts/setup-wizard.sh
To enable the command-line interface, add the bin directory to your shell path:
export PATH="$HOME/openclaw/skills/self-evolving-agent/bin:$PATH"
source ~/openclaw/skills/self-evolving-agent/bin/sea-completion.bash
Self-Evolving Agent Data Schema & Taxonomy
The skill manages its analytical lifecycle using the following data structure within the data/ directory:
| Path | Description |
|---|---|
proposals/ |
Stores generated improvement suggestions in JSON and Markdown formats. |
benchmarks/ |
Contains the results of effectiveness measurements for applied changes. |
trends/ |
Stores weekly trend reports (Emerging/Resolved/Stable patterns). |
stream-alerts/ |
A queue for real-time monitoring notifications and threshold alerts. |
fleet/ |
Multi-instance comparative analysis results. |
Self-Evolving Agent Advanced Features
- Real-time streaming log monitor that provides threshold-based alerts in under 30 seconds.
- Zero-cost operation mode using local Ollama models (like llama3.1 or mistral) for both embeddings and proposal synthesis.
- Multi-platform delivery support for Discord, Slack, Telegram, and generic Webhooks.
- Integrated GitHub Issues management to track proposed optimizations from creation to resolution.
- Local web dashboard for visualizing quality trends, pattern frequencies, and rule effectiveness.
SKILL.md
Loading
Related Openclaw Skills
OpenClaw Self-Healing System
An autonomous 4-tier recovery system that monitors gateway health and uses AI to diagnose and repair persistent failures.
MemoryBox
MemoryBox is a zero-dependency CLI tool designed to maintain a lean 3-tier hierarchical memory structure for Openclaw Skills.
Skill Optimizer
An automated diagnostic tool that analyzes AI conversation logs to evaluate performance and generate data-driven optimizations for agent skills.
WeChat Moments Copywriting Guide
A conversational AI guide designed to help users rediscover their authentic voice and write personalized WeChat Moments posts through guided reflection.
Playwright MCP
A high-performance browser automation skill that enables AI agents to interact with websites, fill forms, and extract data via the Playwright framework.
Step ASR
A high-performance audio transcription tool that converts Chinese and English speech into text using the StepFun streaming API.
Featured*
