What problem does it solve? Teaching an AI agent a repetitive workflow normally requires writing detailed instructions by hand. This Skill lets you simply demonstrate the task once — in a browser or in native desktop apps — while it records your actions, then turns that recording into a reusable agent skill with semantic locators and verification steps. ## Core Features & Use Cases - Browser recording via Playwright: Captures every user action into events.jsonl with multiple selector candidates (testId, role, id, text, css) plus trace chunks with DOM snapshots, on macOS, Windows, and Linux. - Desktop recording via pynput: Captures global mouse/keyboard events, active-window changes, click-moment screenshots, and macOS accessibility descriptions of clicked elements at roughly 0-3% CPU. - Skill generation from evidence: Summarizers convert recordings into compact markdown timelines that the agent uses to generate a new SKILL.md with semantic Playwright locators or accessibility-tree targets. - Use Case: Demonstrate filling out a weekly report in a web app once; the agent records the session and produces a reusable skill that replays the workflow with parameterized dates and values. ## Quick Start Tell your agent to watch you perform a task and turn the demonstration into a reusable skill, then start the recorder with a named session.
