autoskill

Automate workflow analysis by running a local screenpipe daemon to observe screen activity, which then suggests new scientific skills based on detected recurring patterns. This eliminates manual analysis and automatically proposes compositions tailored to your specific repetitive research tasks.

Local embeddings and screenpipe minimize data exposure during research automation by processing user activity entirely on your machine. Screenpipe observes workflows locally, while sentence-transformers generate embeddings without sending sensitive screen data to external servers.

An LLM backend like LM Studio is required to suggest new skills based on recurring workflow patterns. The skill generation process relies on a local large language model to analyze screenpipe activity data and propose tailored scientific skill compositions.

Screenpipe monitors screen activity by running a local daemon that continuously observes user workflows over a set period, such as the last 24 hours. It feeds this activity data to local embeddings and an LLM backend to propose new scientific skills.

Python dependencies required to analyze user activity and generate skills include screenpipe, sentence-transformers, httpx, and pyyaml. These packages enable local screen monitoring, embedding generation, backend API communication, and configuration management for the workflow analysis pipeline.