wtf is openskills?

OpenSkills is a toolkit that lets you give AI agents specialized abilities, like summarizing meetings, processing Word and Excel files, or turning PDFs into articles, without overloading the AI with information it doesn't need at any given moment. You define a skill as a simple markdown file describing what it does, drop in any reference documents or helper scripts, and the framework handles the rest. When a user asks a question, the agent automatically figures out which skill fits and activates it. The core idea is called "progressive disclosure," and it works in three stages. First, the agent only looks at lightweight metadata, a skill's name, description, and trigger words, to decide which skill might be relevant. Once it picks one, it loads that skill's detailed instructions. Only then, if needed, does it pull in specific reference documents or run scripts. This layered approach keeps the AI focused and efficient, rather than dumping every possible instruction and document into the conversation at once. Reference documents can be loaded always, only when a stated condition is met, or at the AI's own discretion based on the question. For anything involving scripts, say, parsing a PDF or generating charts, the project can run them inside an isolated sandbox environment for safety. Files you reference locally get uploaded to the sandbox automatically, and any output files get downloaded back to you. The framework supports several AI providers, including OpenAI, Azure OpenAI, Ollama, and others. This would appeal to developers or teams building AI assistants that need real, modular capabilities rather than just chat. The included examples, meeting summarization, office document processing, weekly report aggregation, give a concrete sense of the use cases. The markdown-based skill format keeps the barrier to entry low, and the auto-discovery of reference files means you can drop documents into a folder and let the AI decide when they're relevant.