Build Self-Improving Agent Skills with cognee and n8n

Vasilije Marković's guest post discusses a workflow for maintaining and updating agent skill files using n8n and cognee, focusing on minimizing the neglect of these files over time. The method involves a maintenance loop where underperforming skill reviews trigger a sequence that records feedback, proposes a rewrite through cognee, and routes the proposal through an approval process in n8n. The proposed changes are then applied only after approval, with a detailed before-and-after diff available for inspection. The process can be run using a visual editor in n8n, which handles orchestration, scoring, and notifications, or through a self-hosted setup using cognee's Python SDK for more customization. This ensures that skills remain up-to-date with the project's current requirements, thus preventing oversight issues, such as missing authorization checks, from recurring.