Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    video repro

    1

    Extract structured bug reproduction steps and clickable timestamps from screen recordings.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned

    video repro

    Example session with this skill installed

    Analyze the bug recording at ./recordings/login_error.mp4 and get the repro steps. It happened on the staging environment.

    • Read your context and instructions
    • Compiled the video repro

    I have analyzed the screen recording. The system identified 8 distinct steps leading to the authentication failure.

    Bug Analysis: Login Loop on Staging

    StepTimeActionConfidence
    10:02Navigate to https://staging.acme.corp/loginHigh
    20:05Enter "admin@test.com" into the Email fieldHigh
    30:08Click the "Next" buttonHigh
    40:12Enter password into the hidden fieldLow (Value not visible)
    50:15Click "Sign In"High
    60:22Observe the spinner for 7 seconds (Idle span)Medium
    70:30Redirected back to the empty login pageHigh
    80:31"Unexpected Token" toast appears at top rightHigh

    Summary: The user is redirected back to the login screen after a long timeout, followed by a JSON parsing error toast.

    Next steps

    1. View the interactive report with frame thumbnails at: http://localhost:4320/job/login-repro-9912
    2. Check the server logs for the "Unexpected Token" error at 0:31.
    3. Verify if the staging environment’s auth-service was responding during the 7-second idle period.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Requires: npm i -g video-repro (Node 20+ and ffmpeg)

    You give Claude a bug video. It writes the repro steps and opens them at localhost:4320, with the video player, clickable timestamps and frame thumbnails. No API key needed.

    Writing up bugs from screen recordings by hand is slow, and it's easy to miss small UI changes. This skill pulls the repro steps out of the video for you and saves them in a set format, with a local page to replay the video against each step.

    What it does

    • Step extraction: picks out user actions, typed values and UI changes from the video frames.
    • Change detection: ffmpeg keeps only the frames where the screen changed, so idle time is skipped.
    • Confidence scoring: marks steps as high, medium or low. Low means the action was unclear or happened off-screen.
    • Narration (optional): whisper.cpp transcribes spoken narration and folds it into the steps.
    • Playback page: a local page with timestamps you can click to jump to the frame for each step.

    How it works

    1. Prepare: prepare.mjs pulls out the frames where the screen changed, plus timing and idle spans.
    2. Analyze: Claude reads each frame in order and spots what changed: pages, menus, fields, toasts, modals.
    3. Write steps: noise like mouse wandering gets merged into clear, single-action steps. Anything the video doesn't show goes into gaps.
    4. Finish: finish.mjs checks result.json and saves it. Then Claude gives you the localhost link.

    Requirements

    • Node.js 20+
    • ffmpeg on PATH
    • The video-repro npm package (global install)
    • Optional: whisper.cpp for narration

    The analysis is done by Claude's own vision. No API key needed.

    Why this beats prompting it yourself

    A plain "watch this video" prompt tends to miss small UI labels and get the timestamps wrong. This skill only sends Claude the frames where the screen changed, makes it follow a fixed JSON format, and makes it flag anything it can't actually see.

    Use cases

    • QA triage: turn bug recordings into repro steps a dev can follow.
    • Documenting current behavior: record a flow and get a starting set of steps.
    • Regressions: find the exact second something breaks or disappears.

    Known limitations

    • Needs ffmpeg on PATH.
    • macOS may block videos in Downloads. Move the video to the Desktop first.
    • Runs locally only. The link works on your machine and can't be shared.

    Requires ffmpeg on PATH. macOS users may need to move files to Desktop to avoid 'Operation not permitted' errors. Best results require Node 20+.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 6 days ago

    Requires

    macOS

    Permissions

    Write Files
    Network Access
    Listed6 days ago

    What's inside

    Frequently Asked Questions

    The skill converts raw screen recordings into a structured JSON list of user actions, input values, and UI state changes, effectively turning a video into a technical bug report. It solves the problem of manual transcription and the common issue of missing subtle UI interactions that occur during a recording.

    You will need Node.js 20 or higher, the `video-repro` npm package, and ffmpeg installed on your system's PATH. For developers who want to include verbal narration in their reports, optional integration with whisper.cpp is supported.

    Unlike basic prompts, this skill uses a specialized frame-by-frame diffing logic and visual change detection to ensure accurate timestamp mapping. This prevents the AI from "hallucinating" actions during idle time and ensures every minor UI label change is captured in the structured output.

    Yes, the skill generates a local web report with interactive, clickable timestamps. This allows developers to click any step in the generated reproduction list and jump immediately to the corresponding frame in the video for visual verification.

    If you are using macOS, you must ensure your video files are moved out of the Downloads folder before processing to avoid OS-level permission errors. Additionally, while the skill is highly accurate, it includes a confidence scoring system to flag any steps where an action might be ambiguous or occur off-screen.

    New in Agents & Orchestration