SkillsSkill for Claude

Practice sheet from music lessons

A private instrument teacher gives Claude the last two to four lesson recordings of one student and the practice points they want on the sheet. Claude prices the recordings against the teacher's Vivu plan, uploads them to a private project, runs one precise search per point, reads the transcript inside each returned window to keep only the minutes where that point is taught, picks a frame of the demonstration from the local file, and writes a practice sheet with the lesson, the time, the teacher's words, the frame and this week's assignment. The value is in what the recording shows: teachers say "not like this, like this", and the transcript keeps only the word "this". Because everything in a lesson is explanation, a search for one point can land on the point taught a minute earlier, so every window stays a candidate until the transcript shows the teacher explaining that point.

Maintained by Vivu. Updated 2026-11-19.

Download

lesson-practice-sheet.zip

10 KB. Unzips to lesson-practice-sheet/SKILL.md. Upload the zip as it is in the Claude app, or unzip it into your skills folder for Claude Code.

SHA-256 648b88b4fa15f10f8fd99475587e322dab86d249fdc79dc354f648e80cd181b3

At a glance

What the Practice sheet from music lessons skill does, where it runs, what it needs, and when it asks
Looks forThe minutes where the teacher explains and shows each practice point, plus a frame of the hand, arm or posture being demonstrated, which a transcript reduces to "this". Every window is checked against the lesson transcript before it reaches the sheet, and the teacher checks every frame.
Runs onClaude Code on your computer (the terminal or the Code tab of Claude Desktop), because it reads local lesson files and runs ffmpeg. No residential IP is needed since nothing is downloaded, and nothing recurs, so no scheduler is involved.
Needs
  • The Vivu connector with write access, to create a private project, open its upload page and search.
  • The lesson recordings as local video files, because Vivu indexes the files and every frame is cut from them.
  • A transcript per lesson (a caption export or a local speech to text tool), to tell this point from the next and quote the teacher.
  • A shell with ffmpeg and ffprobe, to measure lessons and cut contact sheets and frames.
  • An upload path: the upload page opened in your own browser, or a browser tool that can attach local files.
  • The student's consent to recording, and a guardian's written consent for a minor.
Your Vivu planIndex minutes equal the total length of the lessons you upload, and each point costs one precise search. The skill measures the lessons and shows the minutes and credits against your plan from vivu_get_usage before anything is uploaded.
Asks you firstAt the start, to confirm the student's consent (a guardian's written consent for a minor); lessons without it are left out. Then before uploading (the lesson list and its index minutes), after drafting the queries, before writing the full sheet (one sample row first), and before deleting any project or video. It sends nothing; you share the sheet.

The skill does its video work through the Vivu connector. If the connector is not in your Claude yet, add it first; the skill checks that it is connected before it does anything else.

Before you run it

  • Record and index lessons only with the student's consent; for a minor, a guardian's written consent is required or the skill is not used.
  • Lesson recordings show the student, so the skill creates a private project; the videos stay in your Vivu project until you delete them.
  • Searches for one point can return the neighbouring point of the same lesson; the skill checks every window against the transcript and counts the false ones.
  • Picking the demonstration frame is judgment from the pictures and the demo can sit outside the window; check every frame before sending.
  • Claude in Chrome uploads at most 10 MB per call, so long lessons go through your own browser or the Vivu web app.
  • The sheet lists what you taught; it does not grade the student, recognize faces or voices, or search by the sound of the playing.
  • The skill sends nothing; you share the sheet with the student and their guardian yourself.

Start it

Once the skill is installed, ask for the task in your own words. Naming the skill is the most reliable way to have Claude use it. For example:

Make a practice sheet for my student from the last three lesson recordings in ~/Lessons/AB. The points are: relax the right wrist, rushing in bar 12, and let the hand land before closing.

In Claude Code you can also type /lesson-practice-sheet. Claude asks for anything the request leaves out, most important first.

What is inside

  1. When to use
  2. Working principles
  3. What you need before starting
  4. Inputs to collect
  5. Files and state
  6. Step 1: Check the Vivu connector and the setup
  7. Step 2: Turn the practice points into queries
  8. Step 3: Measure the lessons and price them
  9. Step 4: Upload and index
  10. Step 5: Search each point
  11. Step 6: Check every window against the transcript
  12. Step 7: Pick the demo frame
  13. Step 8: Write the practice sheet
  14. Compliance
  15. Known failure modes

The full skill

This is lesson-practice-sheet/SKILL.md from the download, as Claude reads it: the frontmatter first, then the instructions.

---
name: lesson-practice-sheet
description: "Turn a student's recorded music lessons into a practice sheet with Vivu: each point you taught, lesson and time, your checked words and a frame of your demo. Use after a run of lessons."
---

Practice sheet from recorded music lessons

This skill is for a private instrument teacher (piano, violin, guitar, voice, online or in person) who records lessons. The teacher gives Claude the last two to four lesson recordings of one student and the four to six points they want the student to practice ("relax the right wrist", "rushing in bar 12", "let the hand land, then close"). Claude prices the recordings against the teacher's Vivu plan, uploads them to a private Vivu project, runs one precise search per point to find where the teacher explained and showed it, reads the transcript inside each returned window to keep only real explanations, picks a frame of the demonstration from the local file, and writes a practice sheet: point, lesson file, start and end time, the teacher's own words, the demo frame, a link to the recording at that time, and this week's assignment. The teacher sends the sheet to the student and parents themselves.

The value is in what the recording shows. Teachers say "not like this, like this" and "put your hand here"; in our test run the captions of a hand lesson were full of "these bumps" and "right in here", and the caption writer had to add notes like [puts fist into palm] to say what the teacher did. A transcript gives the student the word "this"; the frame shows the hand. The hard part is that everything in a lesson is explanation, so a search for one point can land on the neighbouring point the teacher explained a minute earlier. Every window Vivu returns stays a candidate until the transcript inside it shows the teacher explaining that point.

When to use

Use when a teacher asks "make a practice sheet from my last three lessons with this student", "find where I showed the wrist thing and the bar 12 fix so I can send it to the parents", "turn these lesson recordings into practice notes with timestamps and pictures", or "which minute of the lesson did I explain each of these?".

Not for these:

  1. One question about one lesson ("where did I talk about vibrato on Tuesday?"): search Vivu directly and read the transcript around the hit.
  2. Comparing performances, grading the student or rating progress. The sheet lists what the teacher taught; it judges nobody.
  3. Finding a passage by the sound of the playing ("the take where the run is clean"). Searching by performance sound was not exercised in our test run; this skill searches what the teacher says and shows.
  4. Lessons recorded without the student's consent, or with a minor without a guardian's written consent.

Working principles

  1. Report measured numbers, not estimates. When a number is an estimate, say so.
  2. Nothing is verified until it has been checked against the recording. Every Vivu window is a candidate until the transcript inside it shows the teacher explaining that point; the teacher's words come from the transcript, never from Vivu's reason text, which paraphrases.
  3. Stop and tell the teacher when a required capability or tool is missing. Do not guess around it.
  4. Ask before anything expensive to redo (the recordings and their index minutes, before upload) and before writing the full sheet (one sample row first). The skill sends nothing; the teacher shares the sheet.
  5. A point with no checked window is reported as "not found in these lessons", never as "not taught".

What you need before starting

Check each item at the start of the run and tell the teacher plainly what is missing before doing anything else.

Requirement Why How to check
Vivu connector with write access create a private project, open its upload page, search vivu_get_account shows can_create_projects: true (tool names may carry a server prefix). A write call failing with "has not granted vivu.write" means the teacher reconnects Vivu and allows write access
The lesson recordings as local video files (Zoom or Meet cloud recording downloaded, or the phone video copied to the computer) Vivu indexes files it is given, and every frame is cut from the local file ffprobe -v error -show_entries format=duration -of csv=p=0 "FILE" prints a length for each file
A transcript per lesson: the meeting tool's caption export (.vtt), or a local speech to text tool such as whisper.cpp the transcript inside each window decides whether the teacher is explaining this point, and the teacher's words are quoted from it open the caption file, or run the speech to text tool's version command
A shell with ffmpeg and ffprobe measure lessons, cut frames and contact sheets ffmpeg -version, ffprobe -version
An upload path move the files into Vivu vivu_open_upload_page plus the teacher's own browser, or a browser tool that can attach local files
The student's consent to recording, and a guardian's written consent for a minor lessons show and record the student the teacher confirms it in the chat

This skill needs Claude Code on the teacher's computer (the terminal or the Code tab of Claude Desktop), because it reads local files and runs ffmpeg; Claude on the web and cloud sessions cannot reach the files. It downloads nothing from the internet, so no residential IP is needed. Nothing recurs, so no scheduler is involved; run it again for the next batch of lessons.

Inputs to collect

Ask for anything missing, most important first.

  1. POINTS: the four to six practice points, in the teacher's own words, with the lesson each came from if the teacher remembers. Required.
  2. LESSONS_DIR: the folder with this student's lesson files (two to four lessons). Required.
  3. ASSIGNMENTS: what the student should do this week for each point ("6 slow drops each hand, daily"). Default: left blank for the teacher to fill.
  4. TRANSCRIPTS: where the caption files are, or which speech to text tool to run. Default: a .vtt with the same name as each video in LESSONS_DIR.
  5. LINK_PATTERN: how the teacher shares recordings (a Drive or YouTube unlisted link with a time parameter, such as https://youtu.be/VIDEO_ID?t=SECONDS). Default: no link; the sheet gives the file name and time.
  6. STUDENT_KEY: a short label for the working folder and the Vivu project, not the student's full name. Default: initials and the month.

Files and state

Keep everything in one working folder next to LESSONS_DIR:

practice-STUDENT_KEY/
  config.json          points with their query, lessons, Vivu project id, link pattern
  lessons.csv          file, listed_name, lesson_date, duration_s, video_id, transcript_file
  results/             one JSON per point: search results without the result page link
  sheets/              contact sheets cut around kept windows
  frames/              the demo frame chosen for each point
  candidates.csv       every window Vivu returned, with the transcript verdict and a reason
  practice_sheet.csv   one row per point
  practice_sheet.md    the printable sheet for the student and parents
  message_draft.txt    a short note the teacher can paste when sending the sheet
  state.json           steps done, files uploaded, points searched with job ids, windows checked

Run the commands below from the folder that holds both LESSONS_DIR and practice-STUDENT_KEY/, so relative paths resolve. In the commands, FILE is one lesson file name inside LESSONS_DIR, POINT_KEY is the point's short key from config.json (for example wrist_location), TRANSCRIPT is that lesson's caption or transcript file, and SECONDS is the time of the chosen tile (START plus k divided by 2). A rerun reads state.json first and skips finished work: an uploaded file is not uploaded again, a point with a saved result is not searched again, and a window with a verdict is not checked again. For next week's sheet, add the new lessons to the same project and search only the new points.

Step 1: Check the Vivu connector and the setup

Goal: confirm every requirement before spending anything.

  1. Call vivu_get_account. If the tool does not exist, tell the teacher to add the Vivu connector in Claude (https://mcp.vivu.ai/mcp) and stop. If the account does not show can_create_projects: true, or a later write call fails with "has not granted vivu.write", ask the teacher to reconnect Vivu and allow write access, then stop until they have.
  2. Run ffmpeg -version and ffprobe -version.
  3. List LESSONS_DIR, run the ffprobe length command on one file, and open its caption file (or run the speech to text tool's version command).
  4. Ask the teacher to confirm the consent items in Compliance. Without consent for a recording, leave that lesson out.

In our test run the transcripts were the recordings' caption files. Running a local speech to text tool was not exercised in our test run, so when the teacher has no caption export, test the tool on one lesson before going on.

Done when vivu_get_account shows can_create_projects: true, ffmpeg and ffprobe print their versions, a lesson file has a length and a transcript source, and the teacher has confirmed consent.

Step 2: Turn the practice points into queries

Goal: one search query per point, written the way the lesson sounds and looks.

  1. Rewrite each point as a description of the teacher explaining it and what the teacher does while explaining. Template: "the teacher explains TOPIC while DOING_WHAT". Example: "let the hand land, then close" becomes "the teacher drops a juggling ball or a fist into the open palm and has the student let it land before closing the hand". The search matches what is said and shown in the lesson, so naming the action (pointing at the wrist, turning the hand on the thigh, holding the bow arm) separates this point from the one explained a minute earlier.
  2. Keep the teacher's own wording in config.json next to the query; the sheet shows the teacher's wording.
  3. Show the point to query table to the teacher.

In our test run we wrote the points ourselves from the captions; the teacher's approval of the query table was not exercised in our test run.

Done when the teacher approves the queries and config.json holds them.

Step 3: Measure the lessons and price them

Goal: an indexed set the teacher's plan can pay for, approved before upload.

  1. Run the ffprobe length command on every file in lessons.csv and write duration_s. Add listed_name: the file name with spaces and punctuation replaced by underscores (hyphens and the dot before the extension stay), because that is how Vivu lists names.
  2. Call vivu_get_usage and show one table:
These lessons Plan allowance Remaining this month
Index minutes sum of duration_s / 60 from vivu_get_usage from vivu_get_usage
Search credits 5 per point, 30 for six points (estimate), plus 5 for each rewording from vivu_get_usage from vivu_get_usage

Plan facts: Free is $0 a month with 20 index minutes a month and 50 search credits a month; Premium is $30 a month with 180 index minutes and 500 search credits. A precise search uses 5 credits in total. Three 40 minute lessons are about 120 index minutes (estimate), which fits Premium for one student a month; Free does not cover one such lesson.

  1. If the lessons do not fit, offer levers in this order: index only the lessons where the points were taught (the teacher usually remembers), then split the points across two sheets, and only then move to a larger plan. Never drop a lesson the teacher asked for without saying which one, and never trim or recompress a file to save minutes.

Our test account returned no allowance figures from vivu_get_usage, so the plan comparison and the teacher's approval were not exercised in our test run.

Done when the teacher approves the lesson list and its index minutes, and config.json records both.

Step 4: Upload and index

Goal: every approved lesson indexed in a private Vivu project.

  1. Call vivu_list_projects and reuse a project named "Lessons STUDENT_KEY" if one exists. Otherwise call vivu_create_project with that name and visibility "private". The default visibility is organization, which shows the project to everyone in the teacher's Vivu organization (a music school account, for example); lesson recordings show the student.
  2. Call vivu_open_upload_page with the project ID immediately before uploading. The link expires in 180 seconds and works once, so never post or store it. Give it to the teacher to open in their own browser and select the files, or open it in a browser tool that can attach local files. Claude in Chrome accepts at most 10 MB per upload call and a lesson recording is usually far larger, so lessons normally go through the teacher's own browser or the Vivu web app. Never split or recompress a file to fit.
  3. Poll vivu_list_videos until every file shows ready. Match each Vivu file name to lessons.csv by listed_name; when a name does not match, match by length (duration_ms against duration_s). Record each video_id in lessons.csv and state.json.

In our test run both lessons showed ready about 14 minutes after the upload started, and the names Vivu listed matched the local names. Uploading through the teacher's own browser was not exercised in our test run.

Done when every file in lessons.csv shows ready and has a video_id.

Step 5: Search each point

Goal: a saved list of candidate windows for every point.

One precise search per point, because each point is its own row on the sheet. Every search is precise: only precise returns a time window, and fast returns whole files with an empty reason, which the teacher's lesson list already gives. The chain behind each row: the precise search suggests windows, the transcript inside each window decides whether the teacher is explaining this point or a neighbouring one (Step 6), and a frame from around the kept window shows the hand, arm or posture the transcript only calls "this" (Step 7). Step 7 is the flip from what is said to what is shown.

Run vivu_search_videos (project_id, query, mode "precise", maximum_results 5) for each point. Five is enough for a point taught once or twice in each of two to four lessons, and small enough that the search does not fill up with the neighbouring points of the same lesson; it is also the recall ceiling, so a point the teacher came back to in every lesson needs a higher value. vivu_search_videos returns a job ID; call vivu_get_search_results until complete is true. Each status call can wait up to 45 seconds, so a pending search is not a stalled one. Save the results to results/POINT_KEY.json without the result page link. Show the result page link only in the live reply: it expires after four hours, so it never goes into the sheet, the message draft or state.json.

The queries from our test run, on a guitar lesson about hands and arms and a second lesson about sound, show the pattern. Adapt the topic and the action to your own lessons.

Field Query Mode maximum_results
wrist_location the teacher shows where the wrist really is, pointing to the two bumps at the end of the arm and explaining that the fingers start further back than the knuckles precise 5
thumb_unit the teacher explains the opposable thumb, the pad of the thumb meeting the pads of the fingers, and that the four fingers work together as one unit precise 5
forearm_muscles the teacher puts a hand on the forearm while wiggling the fingers to show that the muscles that move the fingers are in the forearm, and says not to stiffen it precise 5
ball_into_palm the teacher drops a juggling ball or a fist into the open palm and has the student let it land before closing the hand precise 5
forearm_rotation the teacher asks what the student rotates the hand around, has him try turning it around the thumb, and explains that the turn comes from the elbow and the little finger side gives stability precise 5
sound_every_direction the teacher says the sound goes everywhere, up and down, side to side, front and back, and that the player's sound also shines out behind them precise 5

In our test run the forearm_rotation wording above came after 2 rewordings on the same corpus, and the sound_every_direction wording after 1 rewording. The first rotation wording returned the neighbouring forearm explanation and two other moments of the same lesson. A rewording that described only the physical steps (hand flat on the thigh, turn it, hold the elbow) returned 0 results. What worked was describing the exchange as it sounds: the teacher's question, the student's try, and the explanation. The first sound wording found the moment the teacher turned his back but missed the opening and closing explanations of the same point; naming the words the teacher repeats ("up and down, side to side, front and back") found all three.

Done when every point has a saved result file and its job id is in state.json.

Step 6: Check every window against the transcript

Goal: each window labeled this point, neighbouring point, or passing mention, before anything reaches the sheet.

This is the step that keeps the wrong minute off the sheet. In a lesson everything is explanation, so a window for one point can sit on the point taught a minute earlier.

  1. For each window, convert start_ms and end_ms to seconds and read the transcript lines whose times fall inside it. In a .vtt or .srt file the times are on the line above the text, so grep -n "^00:MM:" TRANSCRIPT finds the line numbers for minute MM (use ^01:MM: past the hour); read from the window's start line to its end line.
  2. Label the window:
    • this point: for most of the window the teacher is explaining or showing this point, or the student is trying it with the teacher's feedback;
    • neighbouring point: the teacher is explaining a different point of the same lesson;
    • passing mention: the point is named in a sentence or two inside something else (a recap, feedback on another exercise, a story).
  3. Only "this point" windows go on to Step 7. Write every window, with its label and a one line reason, into candidates.csv, and count the other two labels as false positives for the point. Vivu's reason text is a paraphrase and can describe a different point in the words of your query, so it never decides the label.
  4. When a point has several "this point" windows, keep the clearest one for the sheet (often the first full explanation) and list the others on the sheet as "also in".
  5. A point with no "this point" window gets a row saying "not found in these lessons" and goes back to the teacher, who can name the lesson or reword the query once (one rerun is one more precise search).
Worked example from our test run

The corpus was public lesson recordings: 2 videos, 37.59 minutes in all, from one teacher's published guitar lesson series (an online close-up introduction, then a lesson in a room with a wide fixed camera and classmates watching). We wrote 6 practice points from the captions and marked every place each point was taught before searching.

The first wordings returned 12 windows: 9 real and 3 false, and missed 2 segments. All the false windows came from the rotation point: the forearm muscles explanation a few minutes earlier, a passing mention in the closing feedback, and a story between points. For the first of those, the reason text described the forearm muscles window as rotation. The missed segments were the opening and closing explanations of the sound point.

After 2 rewordings of the rotation point and 1 rewording of the sound point on the same corpus, the final wordings returned 11 windows: 11 real, 0 false, and 0 missed. Per point, the windows averaged 23 to 62 seconds. The transcript check caught the false windows; the reason text would have let one through. The rewordings were tuned on the same small corpus and one teacher's way of talking, so on your lessons expect neighbouring points again and keep this step.

Done when every window in results/ has a label and a reason in candidates.csv, and each point has either a kept window or a "not found" note.

Step 7: Pick the demo frame

Goal: for each kept window, one frame where the student can see what the teacher's hand, arm or body is doing.

The demonstration is not always inside the window. The teacher often says "like this" a few seconds before or after the sentence the search matched, so cut the contact sheet from 15 seconds before the window to 15 seconds after it.

  1. Cut a contact sheet with one frame every half second (a hand demonstration lasts a second or two):
ffmpeg -v error -y -ss START -t 50 -i "LESSONS_DIR/FILE" -vf "fps=2,scale=240:-1,tile=10x10" -frames:v 1 practice-STUDENT_KEY/sheets/POINT_KEY_START.png

START is the window start in seconds minus 15. One sheet covers 50 seconds; repeat with START plus 50 until you pass the window end plus 15. Tile k (counting from 0 at the top left, row by row) is at START plus k divided by 2 seconds. 2. Pick the tile where the hand or posture of this point is clearest: the teacher's hands in frame and not blocked, the camera not moving. Prefer a close-up (an online lesson, or the phone placed near the instrument) to a wide room shot. 3. Extract that moment as a full frame and look at it:

ffmpeg -v error -y -ss SECONDS -i "LESSONS_DIR/FILE" -frames:v 1 -q:v 3 practice-STUDENT_KEY/frames/POINT_KEY.png
  1. If no tile shows the demonstration, read the transcript a minute either side of the window for "like this", "watch" or "try it", sheet that stretch, and pick from there. If there is still no clear frame, leave demo_frame blank and say so; never use a frame that shows a different point.

In our test run the demonstration was inside the kept window for most points, but not for the ball and sound points: the fist version of the ball exercise came after the window, and the moment the teacher turned his back to show that sound goes behind him was before every window of the final sound wording. In the wide room shot the teacher's pointing hand is small; for the wrist point the close-up introduction gave the clearer frame. Choosing frames is judgment from the pictures, so the teacher checks every frame on the sheet.

Done when every kept window has a frame in frames/ or a blank demo_frame with a reason.

Step 8: Write the practice sheet

Goal: a sheet the teacher can send, with one sample row approved first.

  1. Show the teacher one sample row and the field mapping, and wait for a yes before writing the rest:
point,lesson_file,start,end,teacher_words,demo_frame,video_link,checked,practice_assignment
"Let the ball land in your palm, then close",lesson_2_hands.mp4,13:09,13:32,"Put it six times in here and then close. Try not to anticipate its getting there",frames/ball_into_palm.png,LINK_PATTERN with t=789,yes,"6 slow drops each hand before playing, daily"
Field Source If unavailable
point the teacher's wording from config.json none
lesson_file lessons.csv none
start, end the kept window, as MM:SS none
teacher_words one or two sentences copied from the transcript inside the window NOT VERIFIED when there is no transcript
demo_frame the frame from Step 7 blank, with the reason
video_link LINK_PATTERN with the start in seconds lesson file and time only
checked yes when Step 6 labeled the window from the transcript no
practice_assignment ASSIGNMENTS left blank for the teacher
  1. Write practice_sheet.csv and practice_sheet.md. The Markdown sheet has one section per point: the point, the lesson and time (plus "also in" times), the teacher's words, the frame, and "This week:" with the assignment. Address it to the student; it lists what to practice and never grades the playing.
  2. Write message_draft.txt, a few lines the teacher can paste when sending the sheet, with STUDENT_OR_PARENT_NAME and TEACHER_NAME left for the teacher.
  3. Tell the teacher the counts: points with a kept window, points not found, windows set aside as neighbouring points or passing mentions.
  4. The skill sends nothing. The teacher shares the sheet and the frames with the student and the guardian from their own email or messaging app.

In our test run the sheet has six rows, one per point, each checked against the captions. Our recordings were public videos, so the links point to the public video at the window start. The teacher's approval of the sample row and the assignments were not exercised in our test run, and nothing was sent.

Done when practice_sheet.csv and practice_sheet.md have one entry per point (kept or "not found"), message_draft.txt exists, and the teacher has approved the sample row.

Compliance

  1. Record and index a lesson only with the student's consent. When the student is a minor, the teacher needs a guardian's written consent before Step 4; without it the skill is not used for that student.
  2. Lesson recordings show the student. Create the project as private (the default is visible to the whole Vivu organization) and keep one project per student.
  3. The videos stay in the teacher's Vivu project until the teacher deletes them. Delete a project or video only when the teacher asks, and confirm first.
  4. The sheet goes only to the student and their guardian, and the teacher sends it. The skill writes nothing outside the working folder.
  5. The sheet lists what the teacher taught. It does not grade the student, rate their level or compare them with anyone.
  6. The skill identifies no one: no face or voice recognition, no search by the sound of the playing, no naming of other people who appear in a recording.
  7. Recordings of someone else's lessons (a public masterclass, another teacher's video) are for the teacher's own study under the platform's terms, not for redistribution.

Known failure modes

Symptom Cause Fix
(observed) a window for one point is the neighbouring point of the same lesson everything in a lesson is explanation; the reason text called the forearm muscles window "rotating along the pinky axis" label it neighbouring point in Step 6; reword the query to describe the exchange as it sounds (the teacher's question, the student's try, the explanation)
(observed) a window is a passing mention in the closing feedback or a story between points the point's words come back in a recap label it passing mention; keep the window where the point is taught
(observed) a rewording returns nothing it described only the physical steps; the search "returned 0 results with an empty result_page_url" describe what the teacher says while doing it, then rerun once
(observed) a point taught three times is found once the first wording matched the demonstration and not the spoken explanations name the words the teacher repeats for that point; list the extra windows on the sheet
(observed) the demonstration is outside every window the teacher says "like this" before or after the matched sentence; the moment the teacher turned his back "is outside every window of this wording" widen the contact sheet past both ends of the window, then read the transcript around it for "like this" (Step 7)
(observed) the hand is too small to see in the frame a wide room camera; "the pointing hand is small" prefer a close-up recording of the same point, or ask the teacher to film demonstrations closer next time
(observed) a caption file belongs to a different recording "the creator caption file for 360 Degree Sound is the Deep Cup caption file, word for word" check that the first caption lines match the first words of the video before Step 6; otherwise use auto captions or a speech to text tool
the teacher's words on the sheet do not match the lesson they were copied from the reason text, which paraphrases quote only from the transcript
no caption export and no speech to text tool the meeting tool did not save captions stop and tell the teacher; without a transcript Step 6 cannot tell this point from the next
upload page asks to sign in or shows an error the one time upload link expires after 180 seconds request a new link right before opening it
a lesson file is rejected by the browser upload tool Claude in Chrome accepts at most 10 MB per upload call the teacher opens the upload link in their own browser or adds the file in the Vivu web app
"has not granted vivu.write" Vivu connected read only the teacher reconnects Vivu with write access

Connect Vivu, then add the skill.

The skill runs through the Vivu connector. Add it to Claude from the connector directory, then install the skill.