Show HN: LLM 프롬프트를 50~80% 압축하는 결정론적 미들웨어

Skillware is an open-source framework and registry for modular, actionable Agent capabilities. It treats Skills as installable content, decoupling capability from intelligence. Just as apt-get installs software and pip installs libraries, skillware installs know-how for AI agents. "I know Kung Fu." - Neo The AI ecosystem is fragmented. Developers often re-invent tool definitions, system prompts, and safety rules for every project. Skillware supplies a standard to package capabilities into self-contained units that work across Gemini, Claude, Ollama, GPT, and Llama. A Skill in this framework provides everything an Agent needs to master a domain: - Logic: Executable Python code. - Cognition: System instructions and "cognitive maps". - Governance: Constitution and safety boundaries. - Interface: Standardized schemas for LLM tool calling. This repository is organized into a core framework, a registry of skills, and documentation. Skillware/ ├── docs/ # Comprehensive Documentation & Usage Guides ├── examples/ # Reference Implementations │ └── basic_agent.py # Example showing SkillLoader integration ├── skills/ # Skill Registry │ └── category/ # Domain boundaries (e.g., finance) │ └── skill_name/ # The Skill bundle │ ├── manifest.yaml # Definition, schema, and constitution │ ├── skill.py # Executable Python logic │ └── instructions.md # Cognitive map for the LLM ├── skillware/ # Core Framework Package │ └── core/ │ ├── base_skill.py # Abstract Base Class for skills │ ├── env.py # Environment Management │ └── loader.py # Universal Skill Loader & Model Adapter ├── templates/ # Boilerplate templates for new skills │ └── python_skill/ # Standard template with required files └── tests/ # Automated test suite You can install Skillware directly from PyPI: pip install skillware Or for development, clone the repository and install in editable mode: git clone https://github.com/arpahls/skillware.git cd skillware pip install -e . Note: Individual skills may have their own dependencies. The SkillLoader validatesmanifest.yaml and warns of missing packages (e.g.,requests ,pandas ) upon loading a skill. Create a .env file with your API keys (e.g., Google Gemini API Key): GOOGLE_API_KEY="your_key" import google.generativeai as genai from skillware.core.loader import SkillLoader from skillware.core.env import load_env_file # Load Environment load_env_file() # 1. Load the Skill from the Registry # The loader reads the code, manifest, and instructions automatically skill_bundle = SkillLoader.load_skill("category/skill_name") # 2. Model & Chat Setup model = genai.GenerativeModel( 'gemini-2.5-flash', tools=[SkillLoader.to_gemini_tool(skill_bundle)], # The "Adapter" system_instruction=skill_bundle['instructions'] # The "Mind" ) chat = model.start_chat(enable_automatic_function_calling=True) # 3. Agent Loop # The SDK handles the loop: model -> tool call -> execution -> result -> model reply. response = chat.send_message("Screen wallet 0xd8dA... for risks.") print(response.text) - Core Logic & Philosophy: Details on how Skillware decouples Logic, Cognition, and Governance. - Usage Guide: Gemini: Integration with Google's GenAI SDK. - Usage Guide: Claude: Integration with Anthropic's SDK. - Usage Guide: Ollama: Native integration for local models via Ollama. - Skill Library: Available capabilities. We are building the "App Store" for Agents and require professional, robust, and safe skills. We actively encourage both humans and autonomous agents to contribute to this repository! - Please read our Agent Code of Conduct which outlines our strict expectations for deterministic outputs, zero LLM code generation, and safety boundaries. - When submitting skills, our new Agent-Friendly Pull Request Template provides a checklist to ensure your logic aligns natively with loader.py andbase_skill.py . - Please also review CONTRIBUTING.md for detailed guidelines on folder structure and schema definitions. Skillware differs from the Model Context Protocol (MCP) or Anthropic's Skills repository in the following ways: - Model Agnostic: Native adapters for Gemini, Claude, Ollama, and OpenAI. - Code-First: Skills are executable Python packages, not just server specs. - Runtime-Focused: Provides tools for the application, not just recipes for an IDE. For questions, suggestions, or contributions, please open an issue or reach out to us: - Email: [email protected] - Issues: GitHub Issues