video-to-mastery
Skill1 starsChanged 3 months agoMITView on GitHub
Turns a long educational video (YouTube tutorials, lectures, courses, programming/AI/ML/robotics/electronics walkthroughs) into a short, verified, up-to-date learning experience instead of a summary. Use this whenever the user shares a video URL and wants to learn from it, asks "teach me this video," "is this tutorial still relevant/outdated," "update this video to [year]," "should I watch this," or gives a video link plus a learning goal or time budget (e.g. "I have 20 minutes, extract what I need"). Also use for comparing multiple videos on the same topic into one curriculum. Do not use this for simple "summarize this video" requests where the user explicitly just wants a recap of what was said — this skill is for reconstructing verified, current knowledge, not reproducing the video.
In this repository2 other files it gives its agents.
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 2 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 18
Never just summarize. A summary compresses words.
- Very specific: line 42
If none of goal/level/time is given, don't block on asking — proceed with sensible defaults (assume intermediate-appropriate depth, FAST …
- Very specific: line 48
Search for the video/transcript if the page doesn't yield enough (e.g.
- Distinctive: line 49
If a transcript truly isn't obtainable, work from whatever metadata, description, and chapter list you *do* have, and clearly tell the user …
- Very specific: line 51
Read references/technical_verification.md for how to extract concepts and technical claims once you have the content.
- Very specific: line 61
Read references/teaching_framework.md for the teaching structure (big picture → mental model → essential concepts → updated knowledge → …
- Very specific: line 67
Use the structure in references/output_formats.md (the "🎥 Video-to-Mastery" template) unless the user's request matches one of the special …
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Video-to-Mastery
What this skill is for
A video teaches something. That something may be true, half-true, or years out of date — and even when it's still true, the video wastes the learner's time with intros, repetition, and filler. This skill's job is to answer "what do I actually need to know after this video?", not "what did the speaker say?"
The equation to hold in mind throughout:
VIDEO → KNOWLEDGE → VERIFICATION → UPDATE → COMPRESSION → TEACHING → PRACTICE → MASTERYNever just summarize. A summary compresses words. This skill reconstructs knowledge — checking it against the present, discarding what's no longer useful, and teaching the rest efficiently.
Core principles (read these before doing anything else)
- Don't trust the video by default. Confidence in a video's delivery says nothing about whether the API, command, or best practice it shows is still current. Software, ML frameworks, and hardware tooling move fast — check dates and versions before teaching anything as fact.
- Preserve concepts, update implementations. The idea behind a technique usually outlives the specific method used to demonstrate it. When something is outdated, say what stays true (the concept) and what needs replacing (the implementation) — don't just discard the whole section.
- Classify, don't summarize. Sort content into essential / important / useful / optional / outdated / incorrect / redundant / skip, and spend your effort on essential + important. This is what makes the output learning-optimized rather than a recap.
- Compress time honestly. Aim to teach the essential material faster than watching the video, but never invent an unrealistic number — if a subject genuinely needs 40 minutes to actually land, say that instead of forcing it into 15.
- Never hallucinate video content. Don't invent timestamps, quotes, code, or claims from a video you couldn't actually access. If you can't get a transcript or reliable metadata, say exactly what you do and don't know instead of pretending to have watched it. This matters more than looking thorough.
Workflow
Step 1 — Understand the request
Figure out from the user's message:
- The video URL(s) (single video, "update this," "compare these," etc. — see Modes below)
- Their goal, if stated (e.g. "learn ArduPilot and configure a flight controller")
- Their skill level, if stated — if not stated, infer a reasonable starting point from context and say what you assumed
- Their time budget, if stated (see
references/teaching_framework.mdfor how depth maps to time)
If none of goal/level/time is given, don't block on asking — proceed with sensible defaults (assume intermediate-appropriate depth, FAST mode ~10-20 min) and state the assumption up front so the user can redirect you.
Step 2 — Get the video content
Try, in order:
- Fetch the video page (
web_fetchon the URL) for title, channel, publish date, description, and chapters if listed. - Search for the video/transcript if the page doesn't yield enough (e.g.
web_searchfor"<title>" transcriptor channel + topic). - If a transcript truly isn't obtainable, work from whatever metadata, description, and chapter list you do have, and clearly tell the user which sections of the analysis are based on transcript vs. metadata-only inference. Never fabricate what was said.
Read references/technical_verification.md for how to extract concepts and technical claims once you have the content.
Step 3 — Detect outdatedness
For every technical claim, command, API, or version mentioned, ask: is this still current? Read references/outdatedness_detection.md for the domain-specific checklist (software, AI/ML, programming languages, robotics/electronics, security). Then use web_search/web_fetch to verify against current sources — prioritize official docs, release notes, and repos over blogs or forum posts. Read references/source_quality.md for how to rank and cite sources, and how to handle disagreement between sources.
Produce a rough freshness score and per-category breakdown (concepts / commands / APIs / installation / best practices) as qualitative estimates — don't fabricate false precision like "73.4% current."
Step 4 — Compress and teach
Read references/teaching_framework.md for the teaching structure (big picture → mental model → essential concepts → updated knowledge → implementation → common mistakes → practice → knowledge check → next step) and for how learning depth (QUICK/FAST/STANDARD/DEEP) maps to time budgets.
Adapt depth and jargon to the user's stated or inferred skill level. Explain formulas with intuition, not bare notation (see references/teaching_framework.md#formulas). For code, don't reproduce obsolete snippets verbatim — explain what changed and give the modern equivalent.
Step 5 — Output
Use the structure in references/output_formats.md (the "🎥 Video-to-Mastery" template) unless the user's request matches one of the special modes below, in which case use the matching template file instead. Always end with a short knowledge check and a next-step recommendation.
Modes
Most requests are the standard mode above. A few phrasings call for a different shape — check templates/ for the full structure of each:
| User says something like... | Mode | Template |
| "Analyze this video and teach me..." | Standard | templates/standard_analysis.md |
| "I have 10 minutes, just the essentials" | Quick | templates/quick_mode.md |
| "Give me a deep dive / I have 2 hours" | Deep | templates/deep_mode.md |
| "Update this video to 2026" / "what's changed since" | Update | templates/update_mode.md |
| "Is this still worth following?" / video is badly outdated | Replacement | templates/replacement_mode.md |
| (end of any mode) | Knowledge check | templates/quiz.md |
For multiple videos on the same topic, don't run the pipeline independently per video — compare them (overlapping vs. unique vs. outdated content, which has the best explanation vs. best hands-on demo) and synthesize one curriculum that references which video (if any) is worth watching for which section. See references/teaching_framework.md#multi-video.
If the user has continuity context available (things they've told you or you already know they've learned), connect the new material to it explicitly and skip what's redundant — but only claim they "already know" something you actually have grounds for, not as a default assumption.
What good output feels like
The user should finish reading and think: "I didn't need to watch the whole thing. I know what's changed since it was published, I know the modern way to do it, and I can actually go use this." If instead the output reads like a transcript recap with a few dates sprinkled in, it hasn't done its job — go back and re-classify content by value and currency rather than by chronology.
See examples/ for full worked outputs across programming, ML, robotics, and electronics videos, including one deliberately outdated tutorial handled end-to-end.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

