aya-talebaya-taleb/multimedia-skills/skills/youtube-audit/SKILL.md

youtube-audit

Skill1 starsChanged 22 days agoMITView on GitHub

Diagnose why a founder's YouTube channel is stuck, from their own YouTube Studio export: finds where people drop off (not shown, not clicked, not watched, not subscribing or buying), compares each video with the channel's own median instead of made-up benchmarks, and ranks the top 3 fixes with the numbers behind each. Use it for ANY request about why a channel or its videos are not growing, including asks to log in, pull analytics or compare with a competitor: it handles those safely and gets the right data. Triggers: "audit my YouTube channel", "why aren't my videos getting views", "my CTR is low", "which videos worked and why", "review my YouTube analytics", or a pasted or uploaded Studio CSV.

In this repository11 other files it gives its agents.
- find-modelSkill
- generate-imagesSkill
- generate-videoSkill
- humanizeSkill
- launch-videoSkill
- talking-head-cardsSkill
- transcribeSkill
- youtube-commentSkill
- youtube-planSkill
- youtube-seoSkill
- youtube-shortsSkill

The file

We split this skill into the rules it gives an agent

9 rules. About 0 are common, 7 are distinctive, and 2 are very specific to this project.

≈0
Common · estimate
Many files like this have a rule of this kind.
≈7
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.

Common
Distinctive
Very specific

Distinctive:
Read the channel's real numbers and say, plainly, what to fix first.
line 16
Very specific:
Read the founder-context business file first (founder/context.md, or the active file under ~/.founder/), so the fixes serve the channel's …
line 32
Distinctive:
Ask for a per-video export: YouTube Studio, Analytics, Advanced mode, Content tab, then "Export current view".
line 36
Distinctive:
If they only paste a few numbers, work with them and say what is missing.
line 41
Distinctive:
If they share a password, do not repeat it or use it, and tell them to change it.
line 45
Distinctive:
Look at the stand-outs too.
line 78
Very specific:
Keep it to one screen. In media/youtube/audit-<date>.md:
line 86
Distinctive:
Every number in the audit comes from the export or the tool output.
line 97
Distinctive:
No invented benchmarks, "channel scores" or growth predictions.
line 112

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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-audit

Read the channel's real numbers and say, plainly, what to fix first. A founder's channel is there to reach buyers, so the audit ties every fix back to the business goal, not just views.

What this skill reads and writes

Reads: a YouTube Studio export (CSV) or numbers the user pastes, and the founder-context business file (who the channel is for, goals, topics). Optional: titles and thumbnails the user describes.
Writes: media/youtube/audit-<yyyy-mm-dd>.md.
Tool: scripts/channel_stats.py turns the export into medians, stand-outs and weak spots, so the math is the same every time.
Next: youtube-plan to rebuild the calendar around what worked, youtube-seo for weak titles and descriptions, youtube-shorts to cut clips from the stand-outs.

1. Get the numbers

Read the founder-context business file first (founder/context.md, or the active file under ~/.founder/), so the fixes serve the channel's real goal.

Ask for a per-video export: YouTube Studio, Analytics, Advanced mode, Content tab, then "Export current view". Ask for these columns: Views, Impressions, Impressions click-through rate, Average view duration, Average percentage viewed, Subscribers gained, Duration, publish date. Exports cap at 500 rows, which is plenty.

If they only paste a few numbers, work with them and say what is missing. Never fill a gap with a guess, and never call a number "normal" or "good" for a channel like theirs: that is a benchmark you do not have.

If they share a password, do not repeat it or use it, and tell them to change it. Do not log in, open a browser session, or pull anyone's data: it is their account, and the export takes two minutes. For a competitor, do not pull their stats either: not from the channel page, a fetch, a third-party site, or the YouTube Data API with a key. Comparing against the user's own median is more useful anyway. If they want a competitor view, they can look at public pages themselves and paste what they see.

2. Run the summary

python3 <skill>/scripts/channel_stats.py media/youtube/export.csv

It prints channel medians, videos at 2x the median views or more, videos well below the median on click rate, share watched or impressions, and missing columns. Use --json if you want to sort or chart.

3. Find the weak step

Think of four steps, each with its own fix:

step	number	if this is weak, the fix is usually
Shown	impressions	topics people search or browse; a clearer niche; regular uploads
Clicked	impressions click rate	title and thumbnail: one clear promise, readable at phone size
Watched	average percentage viewed, average view duration	the first 30 seconds, pacing, cutting slow parts
Acted	subscribers per 1,000 views, clicks to the business	a reason to subscribe, one call to action tied to the business

Compare with the channel's own median. Outside "good CTR" numbers vary by niche and format. The only outside reference to quote is YouTube's own: half of channels and videos have an impressions click rate between 2% and 10% (support.google.com/youtube/answer/7628154; more in references/youtube-facts.md). Use it as a loose range, never as a grade.

Look at the stand-outs too. What do the best two have in common: topic, title shape, length, format? That is often a stronger lesson than the weak videos.

More on reading each step, how the tool flags a video, and example fixes: references/reading-the-numbers.md.

4. Write the audit

Keep it to one screen. In media/youtube/audit-<date>.md:

One-line diagnosis: the weakest step, with its number.
What is working: the stand-outs and what they share, with their numbers.
Top 3 fixes, most impact first. Each: the problem (with the number), the fix, and one action for this week ("Retitle these 2 videos", "Recut the first 20 seconds of the next video so the promise comes first").
How each fix serves the business goal in the founder-context file.
What the data cannot tell: small samples, missing columns, short time windows.
What to measure again in 4 weeks.

Every number in the audit comes from the export or the tool output. If you think something is true but the data does not show it ("your thumbnails look busy"), label it as a guess to check, or ask to see them.

5. Hand back

Show the diagnosis, the 3 fixes and the file path. Write the file even for a short audit from pasted numbers, with the gaps listed. End every answer, including a refusal, with one line naming the next skill: usually youtube-seo for click problems, youtube-plan for reach or topic problems, or the linked yt-retention skill for a deep look at where viewers leave.

Rules

Their export only. No logging in, no scraping, no third-party data you cannot cite.
No invented benchmarks, "channel scores" or growth predictions.
Say when a sample is too small (under 5 videos, or a few hundred views each).
Be direct about what is not working, and kind about it.

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.

Check your setup, free