OpenSpace: A Self-Evolving Skill Engine for Autonomous AI Agents

A Coding Implementation to Design Self-Evolving Skill Engine with OpenSpace for Skill Learning, Token Efficiency, and Collective Intelligence

OpenSpace is a self-evolving skill engine developed by HKUDS that enables AI agents to learn from every task they perform. In the GDPVal benchmark, it demonstrated a 4.2x income improvement and 46% token reduction across 50 real-world professional tasks.

Why This Matters

Traditional AI agents often operate as stateless tools, re-reasoning from scratch for every new task, which leads to high token costs and inconsistent performance. OpenSpace shifts this paradigm by implementing a self-evolution loop that captures, repairs, and reuses skills, addressing the economic inefficiency of LLM-based automation where costs often scale linearly with task complexity. By treating skills as living entities that auto-repair when tools break, the system bridges the gap between unreliable one-off script generation and battle-tested production workflows.

Key Insights

• The GDPVal benchmark shows that OpenSpace achieves a 45.9% average token reduction across 50 professional tasks by reusing evolved skill patterns.
• The engine utilizes three distinct evolution modes—FIX, DERIVED, and CAPTURED—to maintain a healthy skill database and adapt to execution failures.
• OpenSpace implements a hybrid search system combining BM25 and embedding-based ranking to retrieve relevant skills for task descriptions.
• Community features at open-space.cloud allow agents to share evolved skills, enabling collective intelligence and cross-team repository sharing.
• The taxonomy of evolved skills shows that a significant portion (29 out of 165 in GDPVal) focuses on execution recovery, highlighting the importance of error-handling patterns.

Working Examples

Installation and basic execution of a task within the OpenSpace environment.

import subprocess, sys, os
subprocess.check_call([sys.executable, "-m", "pip", "install", "-q", "git+https://github.com/HKUDS/OpenSpace.git"])
from openspace import OpenSpace
import asyncio

os.environ["OPENAI_API_KEY"] = "your_key"
os.environ["OPENSPACE_MODEL"] = "openai/gpt-4o-mini"

async def run_task():
    async with OpenSpace() as os_engine:
        result = await os_engine.execute("Analyze sales.csv and generate a revenue report.")
        print(f"Evolved Skills: {len(result.get('evolved_skills', []))}")

asyncio.run(run_task())

Structure of a manual skill definition in the SKILL.md format used by OpenSpace.

---
name: data-validation-csv
description: Validate CSV files for issues like encoding and missing values.
version: 1.0.0
origin: manual
triggers: ["csv", "data validation"]
---
# data-validation-csv
## Instructions
1. Encoding Detection: Try UTF-8 first, then fall back to latin-1.
2. Delimiter Detection: Use csv.Sniffer() to auto-detect delimiter.
```python
import pandas as pd
def validate_csv(filepath):
    df = pd.read_csv(filepath)
    return df.isnull().sum().to_dict()

### Practical Applications
- Use Case: Automated CSV data validation and processing using the data-validation-csv skill to detect encoding and delimiters. Pitfall: Hardcoding file paths or encodings, which leads to execution failures in varied environments.
- Use Case: Multi-layer execution recovery where agents handle sandbox failures by applying targeted fixes like pip install or batch processing. Pitfall: Infinite retry loops without capturing the specific failure type, resulting in wasted tokens and API timeouts.
- Use Case: Professional report generation using the report-gen-fallback skill to transition from PDF to HTML or plain text when libraries are missing. Pitfall: Relying on a single output format, which causes task failure if specific system dependencies are unavailable.

**References:**
- https://github.com/HKUDS/OpenSpace
- https://open-space.cloud
- https://www.marktechpost.com/2026/03/24/a-coding-implementation-to-design-self-evolving-skill-engine-with-openspace-for-skill-learning-token-efficiency-and-collective-intelligence/