// Skill profile
microsoft/skill-recorder is a desktop application that turns a single on-screen demonstration into a reusable agent skill or automation. You record a task once — the app captures screen activity locally — and when you choose Analyze, GitHub Copilot reconstructs the session as an intent with ordered steps. You can then generate a SKILL.md procedure (for agent reuse) or a scheduled Automation (for recurring execution). The tool generalizes from one example, so recording one form submission can produce a skill that handles all similar forms. Supports optional voice narration transcribed on-device via Whisper in up to 99 languages. Includes an app-owned terminal with secret/PII scanning. Prefers generating steps that use native tools (like the gh CLI) over replaying UI clicks. Built with Electron and TypeScript, targeting macOS and Windows 11. 5.2K GitHub stars, MIT licensed.
Skill Recorder is Microsoft’s approach to the “show once, reuse forever” problem in agent workflows. Instead of manually writing SKILL.md files that describe how to perform a task, you demonstrate the task once on screen, and the tool reconstructs it as a structured procedure using GitHub Copilot. The key insight is generalization from a single example — recording yourself filling out one expense report can produce a skill that handles any expense report with the same structure.
One-shot recording: Start recording with Ctrl+Shift+R (or Cmd+Shift+R on Mac), perform the task, stop recording. The app captures screen activity locally — nothing leaves your machine until you explicitly choose to analyze.
Copilot-powered reconstruction: When you choose Analyze, GitHub Copilot processes the recording and reconstructs the session as an intent (what the task accomplishes) with ordered steps (how to do it). The reconstruction prefers native tool commands over UI replay — if you used a web form to create a GitHub issue, the generated skill might use the gh CLI instead.
Two output modes: Generate a SKILL.md procedure that agents can follow on demand, or create a scheduled Automation for recurring tasks. The SKILL.md is a read-only preview you can review before saving.
Voice narration: Optional microphone capture, transcribed on-device via Whisper in up to 99 languages. Narration adds context that pure screen capture misses — explaining why you clicked something, not just that you clicked it.
Security-aware terminal: The app provides its own terminal for tasks that involve command-line steps. It includes automatic secret and PII scanning to prevent sensitive values from ending up in generated skills.
Skill Recorder currently depends on GitHub Copilot CLI (pinned to version 1.0.78) for the analysis step, so you need an active Copilot subscription. The tool targets macOS as the primary platform with Windows 11 support. Recording quality depends on performing the task cleanly — false starts and corrections during recording can confuse the reconstruction. The generalization step works best for structured, repeatable tasks and may struggle with tasks that require significant judgment or context.
Developers and teams who want to turn manual procedures into agent-executable skills without writing SKILL.md files by hand. Particularly useful for teams onboarding new members, where experienced developers can record their workflows once and share the generated skills. Also valuable for individuals with recurring tasks that follow a consistent pattern but aren’t worth hand-coding an automation for.
AI agents that work well with Skill Recorder.
Output-style skill that reshapes every coding agent reply to be action-first, numbered, and free of preamble — designed for ADHD-friendly workflows but useful for anyone who wants concise responses.
Meta-skill that designs domain-specific agent teams — generates specialist agents, orchestrator skills, and team architectures from six battle-tested patterns.
Marketplace of 100+ product management agentic skills — covers roadmapping, PRDs, user research synthesis, prioritization frameworks, and stakeholder communication.