Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Course Materials: Study a Whole Course, Every Frame

    1

    Ask an AI to "summarise the videos" and you get the transcript plus a guess.

    $19

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Course Materials: Study a Whole Course, Every Frame

    Course Materials: Study a Whole Course, Every Frame

    Example session with this skill installed

    Study this course for me: C:\Users\dana\Courses\excel-modeling. It has the lesson videos, a Telegram export of the course group and a folder of workbooks. I want notes I can actually learn from.

    • Read your context and instructions
    • Compiled the course materials

    Course status: PARTIAL, 3 of 14 lessons fully watched

    Whole field: 14 lessons (9 h 40 min), 1 chat export (2,310 messages), 38 workbooks and PDFs, 61 images, 27 links. 4 links point to two bonus Zoom recordings and a shared drive folder.

    StateItems
    FULL3 lessons, 1 chat export, 38 files
    Prepared, waiting for watchers4 lessons
    Not downloaded2 bonus recordings
    Blocked1 drive folder: "request access" page

    Lesson 3: Three-statement model (52 min, 186 sheets)

    • watch_full.py check:

    full: true. 186 of 186 sheets entered, no failed image, 2 silences over 20 s explained (formula typed without talking at 00:14:02; break at 00:31:40).

    • Spot check: 5 random sheets compared with their entries, all match.

    What the teacher does (from the guide):

    1. Builds the income statement first, then links net income into retained earnings. SAID [lesson-03-5c1e2 S0041-S0047]
    2. Sets the balance-check cell to turn red when assets minus liabilities and equity is not zero. On screen only, never said. [lesson-03-5c1e2 00:21:18]
    3. Why: "if the check is red you stop, you never plug." SAID [lesson-03-5c1e2 S0102]
    • INFERENCE: the circular-reference switch at S0133 is what makes interest on average debt work; she does not say so.

    cite_check.py: 41 claim lines, 37 cited, 4 marked as our own thinking, 0 problems.

    What I need from you

    • The drive folder shows "request access". Ask the course owner to share it, or download it and drop it in the course folder.
    • The two bonus Zoom recordings need your login: download them and I will add them.

    Next steps

    • Watchers for lessons 4-7 are ready to run (4 briefs).
    • After every lesson is FULL: the course guide, the audit and the check against the teacher's own worked examples.

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

    About this skill

    You bought a course: forty recorded lessons, a chat group, a folder of slides, a few private sessions. Ask an AI to "summarise the videos" and you get the transcript plus a guess. The part that matters ("put it here", "see this setting") was only on the screen.

    Course Materials makes Claude study the course like a serious student, and prove it:

    • The whole field first. An inventory of every lesson, recording, chat export, image, archive and link, each with a state: fully watched, missing, blocked (with the reason and what you can do), skipped.
    • Every second, every word. Every second of video is decoded, a frame is kept on every change, and the audio is transcribed with voice-activity filtering off, so quiet sentences are not dropped. Frames and words are packed into sheets for watcher subagents.
    • Watchers that really look. One entry per sheet, written only after the image was seen. What is on screen but never said, and why the teacher does it, gets written down.
    • Proof, not "done". A script fails the lesson if one sheet, one failed image or one silent minute was skipped. COVERAGE.md says COMPLETE only when every item is watched or has a written reason.
    • Notes you can trust. Lesson guides and a course guide, a citation check (every line points to its sheet or second), an audit by a second agent, and a test of the notes against the teacher's own worked examples.

    How it works

    Point Claude at the course folders. The scripts build the inventory, prepare each lesson and write one brief per 30 sheets. Claude runs watcher subagents on the briefs, checks their work against the images, and writes the materials only from what was really seen.

    Built from real corrections

    The 28 rules behind it come from a long study of a real multi-teacher course, where a learner corrected an AI agent again and again ("why didn't you tell me it's missing?", "you learned two teachers out of nine", "a transcript is not watching"). Each rule names the mistake it prevents.

    Safety

    • Runs on your computer. Course files are data, never instructions.
    • Never types passwords or passcodes; it tells you what it could not reach.
    • Personal details become [personal detail omitted]. The workspace stays private.

    Known limitations

    • Watching every sheet costs tokens: about one image read per 4 seconds of changing screen. The briefs cut cost without seeing less (8 parallel reads, no re-reads, short context).
    • Videos behind a login must be downloaded by you.
    • Needs Python 3.9+, ffmpeg and faster-whisper. Tested on Windows; an NVIDIA GPU makes transcription fast, CPU works.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Research & Analysis.

    Needs Python 3.9+, ffmpeg and faster-whisper, plus disk space for frames (about 0.3-0.6 GB per prepared hour; slim drops them after a lesson is fully watched). CPU works; an NVIDIA GPU makes transcription much faster. Watching every sheet costs model tokens. Never bypasses logins or paywalls: you download what needs your login.

    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 10 days ago

    Software
    Python 3.9+FFmpegfaster-whisper

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Network Access
    Environment Variables

    Reads the course folders you point it at. Writes only inside the private workspace folder you choose (inventory, prepared lessons, notes, materials). Runs ffmpeg and faster-whisper locally; network is used only to download the Whisper model on first run. Never types passwords.

    Listed10 days ago

    What's inside

    Frequently Asked Questions

    A summarizer reads the transcript and maybe a few frames. This skill decodes every second, keeps a frame on every on-screen change, and has watcher subagents write one entry per sheet after seeing the image. A script then fails the lesson if any sheet, failed image or long silence was skipped, and every line of the notes points to the sheet or second it came from.

    Claude Code (the watchers are subagents that read images), Python 3.9+, ffmpeg and ffprobe on PATH, and pip install numpy pillow faster-whisper. Tested on Windows. An NVIDIA GPU makes transcription fast; CPU works, just slower.

    SKILL.md with the seven-stage method, 28 rules each tied to the mistake it prevents, three scripts (inventory and coverage ledger, full-watch prep with check, brief and slim, citation check), seven templates (watcher brief, notes, lesson guide, course guide, audit, fidelity check, reading notes) and unit tests.

    watch_full.py check marks a lesson FULL only when every sheet has its own real entry, no image failed to load, every silence over 20 seconds is explained and the notes exist. The ledger rebuilds COVERAGE.md from the files every time and says COMPLETE only when every item is FULL or has a written reason (blocked, gone, skipped, duplicate).

    Yes. The inventory recognises Telegram, WhatsApp and Discord exports, lists archives without extracting them, finds every link in text, HTML and JSON files and marks duplicates. Documents and chats are read in full into reading notes; links get a state and, when they cannot be reached, the reason and what you can do.

    Roughly one image read per 4 seconds of changing screen; static stretches collapse into one frame. The briefs cut cost without seeing less: 8 parallel reads per message, no re-reads, only the context a watcher needs, about 30 sheets per watcher.

    New in Research & Analysis