agentsclimarketplace

Youtube copy

Skill AlemTuzlak/skills/skills/youtube-copy

Use when the user wants YouTube metadata for a video — a click-worthy title, an SEO/above-the-fold description, tags, and timestamped chapters. Accepts a transcript (ideally timestamped SRT), a topic, a PR, or a freeform description. Used standalone and by /produce-video. Triggers on "youtube title", "youtube description", "youtube chapters", "youtube tags", "youtube metadata", "youtube-copy".From its SKILL.md

Install
npx -y skills add AlemTuzlak/skills --skill youtube-copy

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.

SKILL.md

5.2 KB, ~1.2k tokens by cl100k_base, as published. Nobody here has run it

YouTube Copy

Write the four pieces of YouTube metadata that drive clicks and rank: title, description, tags, chapters. Ground everything in the provided source (transcript / topic) — never invent claims the source doesn't support. For chapters you need a timestamped source (SRT or words.json); without timestamps, ask for them or skip chapters.

Output format

Write youtube.md (or print the same structure) exactly like this:

# YouTube

**Title**

<title>

**Description**

<description>

**Tags**

#tag1 #tag2 #tag3 ...

**Chapters**

00:00 - <chapter>
01:15 - <chapter>
...

(When /produce-video calls this skill, write youtube.md into the output folder.)

Writing rule (non-negotiable)

Never use em-dashes (— or –). Use a hyphen - instead. This applies to the title, description, chapter separators, everything.

Title (the single highest-leverage element)

  • 60-70 characters (hard max 100; anything past ~70 truncates on most surfaces).
  • Primary keyword near the front so YouTube and viewers see the topic immediately.
  • Do all three: spark curiosity/emotion (surprise, urgency, FOMO, benefit), include the keyword, and set honest expectations (a title that oversells tanks watch-time and is clickbait — avoid it).
  • Use a number when natural (numbers attract the eye) and emotive/power words ("actually", "finally", "stop", "the right way", "type-safe", "without").
  • Use the audience's real lingo, not generic phrasing.
  • Optional: 1-2 relevant hashtags only if they add discovery; usually skip in the title.
  • Avoid: misleading bait, ALL CAPS spam, profanity, >70 chars.
  • Useful formula starters (pick what fits, don't force): "How to X (without Y)", "The X that Y", "Stop doing X - do this instead", "X explained in N minutes", "I tried X so you don't have to", "N ways to X".

Description

Structure, front-loaded (the first ~100-160 chars show above the fold / in search):

  1. Hook (first 1-2 sentences) containing the primary keyword — answer "what is this and why watch it?". Promise the outcome.
  2. A short paragraph (2-4 sentences) expanding what the viewer learns.
  3. Chapters/timestamps (see below) for videos > ~10 min (or whenever there are clear sections).
  4. Optional links/CTA (main link, "watch next", subscribe) - only if the user provided them; don't fabricate URLs.
  5. 2-3 relevant hashtags at the very end.
  • Length: tight and useful, roughly 150-300 words / under ~1000 characters. Write for humans; weave related keywords naturally, never keyword-stuff.

Tags

  • 5-15 tags, comma/space free per tag (no spaces inside a multi-word tag is not required, but keep them clean).
  • Primary keyword/phrase first, then a mix of broad (e.g. react, webdev) and specific (e.g. react router loader data, unstable_useRouteLoaderData).
  • Quality over quantity; all tags must genuinely describe the content. No unrelated/spam tags, no competitor names.
  • Output them prefixed with # in youtube.md for readability (they're entered without # in YouTube's tag box).

Chapters (need a timestamped source)

YouTube's hard rules: first chapter must be 00:00, at least 3 chapters in ascending order, each chapter >= 10 seconds long.

  • First chapter (00:00) answers "why did they click?" - name it the payoff/topic (e.g. "Type-safe loader data", "The final result"), NOT "Intro".
  • Titles: 3-8 words, one clear noun phrase (mobile crops long ones). Replace useless labels ("Intro", "Demo", "Conclusion", "Q&A") with the actual topic of that section.
  • Frequency: roughly every 2-4 minutes for tutorials, 5-10 minutes for long-form; aim for 5-15 chapters total, never > ~20.
  • Derive each MM:SS start from the timestamped source (the moment that section actually begins). Use MM:SS (or H:MM:SS past an hour). Separator is - (a hyphen), never an em-dash.

Process

  1. Read the source. If it's an SRT/words.json, use the timestamps for chapters; if it's plain text or a topic, write title/description/tags and note that chapters need a timestamped source.
  2. Identify the primary keyword/topic (what someone would search) and the payoff (why they'd click).
  3. Draft 2-3 title options, pick the strongest against the rules above, then write description, tags, chapters.
  4. Self-check: title <= 70 chars and keyword-forward; description hook has the keyword in the first sentence; tags 5-15 and relevant; chapters start at 00:00, ascending, >=3, descriptive; no em-dashes anywhere.

References

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.