[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$f2i38lggy1eet6":3},{"code":4,"message":5,"data":6,"timestamp":36},10000,"success",{"full_name":7,"position":8,"description":9,"language":10,"stars":11,"forks":12,"stars_gained":13,"age_days":14,"hn_threads":15,"one_liner":16,"summary_md":17,"topic":18,"github_topics":21,"license":32,"homepage":33,"updated_at":34,"history":35},"microsoft\u002Fskill-recorder",13,"Desktop app that records your on-screen work session and uses the GitHub Copilot SDK to reconstruct it as an intent + ordered steps, then builds a reusable Skill or Automation for Microsoft Autopilots like Scout, Microsoft Copilot Cowork, or Copilot Studio.","TypeScript",5020,553,650,88,[],"Record a desktop task once, then turn that session into a reusable SKILL.md or scheduled automation for Microsoft Copilot agents.","Skill Recorder is aimed at people who already have a capable agent but no practical way to describe a repetitive procedure precisely enough for it to follow. It trades a GUI automation editor for a single recorded demonstration.\n\n## The problem\n\nYou want your agent to handle a recurring task — filing a ticket, submitting a form, pulling numbers from a dashboard — but writing the procedure by hand means guessing the exact steps, tools and inputs the model should use. Watching a person do it is the cheapest source of ground truth. Before, you would either script the clicks or write a long prompt from memory. Skill Recorder changes the capture step: you perform the task normally while the app logs windows, pages, commands and optional narration, then Copilot summarizes the session into an intent plus ordered steps. The result is a procedure you can review and hand to an agent instead of reconstructing it yourself.\n\n## What's different\n\nThe output is deliberately not a click replay. The README says generated skills and automations prefer the agent's native tools, such as `gh` or `web_fetch`, and generalize from a single example, so recording one form submission can teach the agent to handle the others. That reframes the recording as evidence about intent and sequence, while the builder decides how to express each step. Analysis and generation go through the GitHub Copilot SDK\u002FCLI, and both the describer and builders have a fixture-based eval suite you can run locally — a useful sign that the generalization step is treated as something to measure, though the fixtures are synthetic recordings rather than your workflows.\n\nThere are two practical trade-offs. First, generalization cuts both ways: an agent encouraged to reinterpret steps may choose a tool or path you did not expect, which is why the editable analysis and optional read-only SKILL.md preview matter. The preview only gives visibility into the generated instructions; it is not described as a security guarantee. Second, the privacy boundary is explicit and worth reading: capture, storage, frame extraction and narration transcription happen locally, but Analyze uploads the event timeline, extracted screen images, narration text and requested terminal-output excerpts to GitHub's cloud for Copilot processing. Recordings therefore contain real credentials, customer data or internal URLs unless you curate the session.\n\n## Who should try it\n\nGood fit: an agent-platform user who repeats a browser- or desktop-heavy procedure and can point to one clean run as the template. The evals and editable step list make it reasonable to iterate toward a usable skill rather than accept the first draft. Poor fit: someone who needs an auditable, exact UI macro, or whose task depends on screens Copilot cannot see or describe. Also a poor fit if policy forbids sending screen imagery and activity metadata to a cloud service, since analysis is the core step and there is no documented offline analysis path.\n\n## Requirements and limitations\n\nA GitHub account with Copilot access is required, and installation is a source release: one pinned command downloads a Node.js runtime, builds the release commit locally and registers a relaunchable app. The first install takes several minutes and needs a machine willing to compile an Electron app; Git and a separate Node.js install are not needed for this path, though manual developer setup requires Node.js 24. macOS is the primary target, Windows 11 x64\u002FARM64 is supported with a separate validation note, and Ubuntu appears in the install command list.\n\nmacOS requires Screen Recording permission on first launch, and the recorded terminal is optional but enables terminal-command context. Ongoing upkeep means relaunching \"Skill Recorder (Source)\", re-recording when a workflow drifts, and updating through the documented install\u002Fupdate path rather than expecting an auto-updater.\n\n## HotRepo's take\n\nThe only stored signal is one weekly TypeScript board observation on 2026-10-25: +650 stars that week, rank #13 — a young repository (created 2026-07-29) with no daily-board history in HotRepo's coverage, so persistence cannot yet be judged. Before investing time, verify the install path for your OS in the [latest release](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fskill-recorder\u002Freleases\u002Flatest) and read the data-handling section of the [project README](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fskill-recorder\u002Fblob\u002Fmain\u002FREADME.md), particularly what Analyze uploads.",{"slug":19,"name":20},"ai-engineering","AI Engineering",[22,23,24,25,26,27,28,29,30,31],"agent-skills","ai-agents","automation","copilot","copilot-cli","copilot-cowork","copilot-studio","electron","microsoft-scout","screen-recording","MIT",null,"2026-10-25T12:17:00Z",[],1792968031]