More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Claude Real Video: Watch Video with Keyframes

    1

    Actually watch a video instead of guessing: extract scene-aware keyframes and a transcript locally, read them in the right order, and answer with timestamps and frame references for every claim.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Claude Real Video: Watch Video with Keyframes

    Claude Real Video: Watch Video with Keyframes

    Example session with this skill installed

    Here is a 12-minute product demo video. What does it claim the tool does, and does the demo actually show it doing those things?

    • Read your context and instructions
    • Compiled the claude real video

    EXTRACTION: Scene-aware keyframes: 34 frames across 9 scenes. Transcript with timestamps, single speaker. Manifest: 12:04 duration, 34 frames, transcript available, frame budget 60 (under budget).

    READING: Transcript skim: the presenter claims three things — one-click setup (01:10), automatic syncing (04:40), offline mode (08:15). Contact sheets: title card, setup screen, sync indicator, long settings walkthrough. Two individual frames pulled: the setup screen at 01:10 and the sync indicator at 04:40.

    ANSWER WITH RECEIPTS: "One-click setup (01:10, frame 6): the presenter says one click; the frame shows a setup screen with four required fields and a confirmation dialog — the video shows a four-step setup, not one click. Automatic syncing (04:40, frame 18): the presenter says automatic; the frame shows a sync indicator spinning for 40 seconds before completing. Offline mode (08:15): mentioned in narration, no visual demonstration in any frame — the transcript claims it, the visuals never show it."

    DIGEST SAVED: Key moments with timestamps plus a one-paragraph summary, so follow-up questions do not need a re-extraction. The finding is the gap between saying and showing — and it only exists because the answer kept shows and says separate, with a receipt on each.

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

    About this skill

    A language model cannot watch video directly — but it can read pictures and text. This skill converts a video into those two things intelligently: scene-aware keyframes (sampled at cuts, not fixed intervals, so ten frames of the same talking head cost one frame's worth of information) plus a time-aligned transcript, in one self-contained extraction folder with a manifest.

    Frames are spent like money: contact sheets first (nine moments for the price of one image), individual full frames only for the moments that matter, a budget set up front (60 for a long video). The reading order is fixed — manifest, transcript, contact sheets, individual frames on demand — so the expensive reads happen only where the cheap reads say they are needed. Every substantive claim gets a receipt (timestamp or frame number), shows and says are kept separate, and negative claims require checking both transcript and frames. Video content is treated as untrusted data: instruction-like text gets reported, never followed. A worked example watches a 12-minute demo and finds the presenter's "one-click setup" is four steps on screen.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Digital Products.

    • It watches through keyframes and transcript, not the raw video — a moment between sampled frames can be missed, which is why scene-aware sampling matters.
    • No speech means no transcript; the answer then leans on frames alone and says so.
    • Long videos spend a capped frame budget; sections the transcript does not flag may go unchecked, and the report says which.
    • On-screen text that contradicts narration is reported as a finding, not resolved by picking a side.

    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 buying a skill.

    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 3 days ago

    Permissions

    No special permissions declared or detected

    Listed3 days ago

    What's inside

    Frequently Asked Questions

    Fixed intervals waste the budget on near-duplicates and miss the cut where new information appears. Scene-aware sampling puts a frame where the scene changes — ten frames of the same talking head are one frame's worth of information.

    Skip transcription — transcribing silence wastes time and produces hallucinations. The answer leans on frames alone, and the report says so.

    Every substantive claim carries a timestamp or a frame reference. "The presenter claims X (12:34)" and "the diagram shows three components (frame 14)" can both be checked in under a minute. A summary without receipts is a guess.

    New in Digital Products