More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Stage Edit: Footage Editing with Evidence

    1

    Edit real footage on evidence, not guesses: probe every clip, source every timecode from a transcript, probe, or inspected frame, and give each cut a written reason.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Stage Edit: Footage Editing with Evidence

    Stage Edit: Footage Editing with Evidence

    Example session with this skill installed

    Edit my raw footage into a 60-second product teaser. I have three clips: a 6-minute screen recording of my app (silent), a 3-minute talking-head clip of me explaining the problem it solves, and a 40-second customer testimonial. Open on the strongest moment, cut the dead air and filler, and caption everything spoken.

    • Read your context and instructions
    • Compiled the stage edit

    EVIDENCE GATHERED

    • Screen recording: probed 362.4 s, 1440×900, 30 fps, no audio. Frames inspected: the dashboard result view is on screen at 288–304 s; the first 40 s are login and loading — dead air, excluded.
    • Talking head: probed 184.6 s, 1920×1080, speech throughout. Transcript: the problem statement lands at 22.4–38.9 s ("I was spending every Friday night rebuilding the same report"); filler and two false starts fill 96–141 s — excluded on transcript evidence.
    • Testimonial: probed 41.2 s, 1080×1920, speech. Transcript: the quotable line sits at 12.8–21.3 s ("it gave me my Fridays back").

    EDIT PLAN (plan-backed — a teaser usually gets a second cut):

    1. Testimonial 12.8–21.3 (8.5 s) — hook: the customer's strongest line opens, face to camera. Evidence: transcript + frame.
    2. Talking head 22.4–38.9 (16.5 s) — the problem, in the founder's words. Evidence: transcript.
    3. Screen recording 288.0–304.0 (16.0 s) — the result view as proof, carried under the tail of segment 2's audio. Evidence: frames.
    4. Talking head 158.2–177.0 (18.8 s) — the closing invitation and call to action. Evidence: transcript.

    Total 59.8 s. Hard cuts throughout with ambience bridging each join; captions burned last from the transcripts via the caption tool; mix measured at −14.2 LUFS integrated, −1.0 dBTP.

    FLAGGED: the testimonial's background banner is legible in frames, but its text was not checked against a brand guide — verify before publishing. No window above was estimated; each traces to a probe, a transcript position, or an inspected frame.

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

    About this skill

    The editing discipline for footage that already exists: turning raw recordings, screen captures, and talking-head clips into a finished piece where every cut can be defended. It is built against the three enemies of footage editing: guessed timecodes, silent assumptions, and revisions that turn into rebuilds. Every input is probed before anything is cut — duration, resolution, frame rate, audio presence — and the content is read the cheapest honest way: speech is transcribed into a timecoded map, silent material is read through inspected frames, and any judgment that cannot be verified is marked unverified rather than invented. A cut point is legitimate only if it comes from a user timestamp, a probe measurement, a transcript position, or an inspected frame; a moment no evidence can locate goes back to the user as a question, never an estimate. The workflow runs in two paths — one-shot for a plain trim or join, plan-backed for anything the user may revise — with segments carrying an exact window, a one-line reason, and an evidence source, scored against a rubric before assembly. Cut craft is built in: trim to the moment rather than the clip, a restrained transition vocabulary, sound bridging hard cuts, and line-specific direction for social clips, podcast repurposing, screen demos, localization, and documentary montage. Edge cases are handled explicitly — clips shorter than the brief assumes, music where a transcript was expected, user timestamps that land outside the file — and no edit ships without a pre-delivery quality gate covering the output probe, caption provenance, and a measured loudness figure.

    • It edits existing footage only; it does not generate new footage or plan a multi-source video from scratch — generation and planning are separate stages.
    • Word-level cutting depends on a usable transcript; music-only or noisy audio falls back to frame evidence and your timestamps, and the skill says so.
    • Visual judgments require frame inspection; where that is unavailable they are delivered marked unverified, never asserted.

    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 24 days ago

    Permissions

    No special permissions declared or detected

    Listed24 days ago
    Updated21 days ago

    What's inside

    Frequently Asked Questions

    Because an estimated timecode is a cut that lands on nothing — mid-word, mid-gesture, or on a loading screen. Every window in an edit from this skill traces to a probe, a transcript position, an inspected frame, or a timestamp you supplied, and the evidence note shows which.

    No. A plain trim or join runs one-shot, with no plan file — the ceremony matches the reversibility. The plan-backed path is reserved for pieces you may revise, excerpt again, or localize, where a written plan turns a revision into a one-entry change instead of a rebuild.

    The edit stops on that point and asks you for the timestamp, showing what was checked — the probed durations and the transcript search. It will not clamp to the nearest plausible spot or quietly drop the moment; a timestamp that lands outside the file usually means a different export, and the skill surfaces that mismatch instead of absorbing it.

    New in Image & Video