agentleFS
qwwiwiqwwiwi/agentos-skills-public/skills/youtube-producer/SKILL.md

youtube-producer

Skill4 starsChanged 4 months agoMITView on GitHub

YouTube producer for a creator's channel. Generates 5 title variants + thumbnail concept based on an idea from a tweet/article/video, using proven formulas from top YouTube creators (e.g. Nate Herk: 680K subs, 116K avg views). Also provides a hook for the first 30 seconds and a video structure using his patterns. ALWAYS use this skill when: (1) user shares a tweet, article, or another video and says "make a video on this topic", (2) asks for "video title", "come up with a video name", "YouTube title", (3) asks for "thumbnail variants", "cover concept", "thumbnail for a video", (4) asks for "video hook", "how to start the video", "first seconds of video", (5) preparing YouTube content — even if user doesn't name this skill explicitly. Triggers: YouTube, video, title for video, thumbnail, cover, hook.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 4 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If source is unclear — ask: "Is this news or an experiment?"

    line 50
  2. Very specific:

    Use references/title-formulas.md.

    line 54
  3. Very specific:

    No filler words: tutorial, how to (at the start), guide, tips, abstractions

    line 65
  4. Distinctive:

    Write verbatim text of the opening line — 3-5 short sentences, 25-60 words.

    line 95
  5. Distinctive:

    No greeting, no "hey everyone".

    line 101
  6. Very specific:

    Add a new one with the transcript skill (Loore yt_transcript).

    line 128
  7. Distinctive:

    One strong verb in the title: Not three.

    line 142

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

YouTube Producer

YouTube video producer. Learn from top creators, adapt topics from social media, output in target language.

When to use

User brings:

  • Tweet / X post with AI news
  • Link to someone else's YouTube video / reel
  • New release (Claude/OpenAI/Google/etc.)
  • Article or changelog
  • Their own experiment ("I built / tested / spent time with X")

Skill outputs:

  1. Video type — shock claim / demo-first / contrarian / authority-experiment
  2. 5 titles using different formulas with keyword emphasis
  3. Thumbnail concept — composition + text + color
  4. Hook (30 sec) — verbatim text of the opening line
  5. Video structure — 60-90 sec × 7 blocks, 10-14 min total

Workflow

Step 1. Classify the source idea

SourceDefault video type
Tweet about a fresh release (Anthropic, OpenAI, Google)Shock claim (type A)
Someone else's demo / workflowDemo-first (type B)
Controversy / community debateContrarian (type C)
Creator ran their own experimentAuthority / experiment (type D)

If source is unclear — ask: "Is this news or an experiment?"

Step 2. Generate 5 titles

Use references/title-formulas.md. Output exactly 5 variants, each using a different formula:

  1. Authority hijack — big name + verb + tool
  2. Transformation / I Turned — personal transformation story
  3. Shock number — percentage/number + superlative
  4. Just Destroyed — destruction verb + subject
  5. Absolutism / Contrarian — never/forever/everyone

Title rules:

  • In the creator's target language
  • Under 70 characters (fits mobile thumbnail)
  • No filler words: tutorial, how to (at the start), guide, tips, abstractions
  • Short dash only — no em dashes
  • No emoji

Output format:

1. [Formula] — Title with KEYWORD emphasis
   Why it works: 1 sentence about the trigger

2. ...

Step 3. Thumbnail concept

In one block describe the composition. Use references/thumbnail-patterns.md.

Format:

Thumbnail:
- Left: [creator face, emotion]
- Right: [specific object / logo / screenshot]
- Text: "[1-2 words, CAPS]" — color: yellow/red/white
- Accent: [arrow / circle / cross]

Image generation prompt:
[ready prompt for Midjourney/GPT Image]

Step 4. Hook (first 30 sec)

Write verbatim text of the opening line — 3-5 short sentences, 25-60 words.

Requirements:

  • First phrase ≈ title (verbatim).
  • Trigger words: just, destroyed, never again, in 3 seconds.
  • Last phrase — bridge: "No long intro — let's go."
  • No greeting, no "hey everyone".

Step 5. Video structure

Output 7 blocks × 60-90 sec:

0:00–0:30   Hook
0:30–2:00   Context (what's the release/idea)
2:00–2:30   Mid-CTA #1 (subscribe / join channel)
2:30–7:00   Live demo / screen-share
7:00–10:00  Edge cases / gotchas
10:00–12:00 Honest verdict
12:00+      Closing CTA

Under each block — 1-2 sentences on what exactly to show.

Reference library

  • references/title-formulas.md — 5 title formulas with metrics
  • references/scenario-patterns.md — hook/flow/retention/CTA patterns
  • references/thumbnail-patterns.md — composition + generation prompt
  • references/nate-corpus.md — table of 30 videos with views/likes/comments/ER

Transcript corpus

data/transcripts/<video_id>.txt — transcripts of reference creator videos (plain text + metadata header). Add a new one with the transcript skill (Loore yt_transcript); fresh stats for the corpus come from yt-research (Loore or the official YouTube Data API v3). Never scrape YouTube for this data.

Use when:

  • Finding a live quote for a technique
  • Checking frequency of trigger words
  • Seeing how two blocks were connected

Load only the specific transcript needed — files are 10-50KB.

Golden rules

  1. Idea first, format second. Formulas are packaging. First understand what's uniquely valuable (news / transformation / insight).
  2. Don't copy verbatim. Adapt to your channel and audience language.
  3. One strong verb in the title. Not three.
  4. Always show WHY a formula will work for the specific source. Don't output templates without grounding.
  5. If source is weak — say so. Better to skip a topic than package emptiness as clickbait.

Antipatterns (never do)

  • Titles with tutorial / guide / overview / how to do.
  • Long titles (>70 chars) — cut off on mobile.
  • Questions in parentheses ("Really?", "Or not?") — signal of insecurity.
  • Thumbnails without the creator's face — recognition is lost.
  • Hook with "hey everyone" — tired, avoid.

Working cycle "idea → video"

[tweet/link/article]
   ↓
[Step 1: classify]
   ↓
[Step 2: 5 titles]  ← USER PICKS ONE
   ↓
[Step 3: thumbnail for chosen title]
   ↓
[Step 4: hook + structure]
   ↓
[creator records]

User chooses the title from 5 — only then write thumbnail and hook (they depend on the chosen title).

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.