check-thumbnail
Skill2 starsChanged 28 days agoMITView on GitHub
Check thumbnail candidates before an upload - a code gate for format, 16:9, size, empty dark area, clutter, text rules and title repetition, plus a feed-size preview to choose from. Use at the Package station (before upload-checker), for "check my thumbnail", "which thumbnail is better", or when setting the thumbnail in upload.json.
In this repository13 other files it gives its agents.
- channel-pulseSkill
- check-renderSkill
- check-voiceoverSkill
- find-topicSkill
- localizeSkill
- make-shortsSkill
- morph-clipsSkill
- produce-videoSkill
- research-factsSkill
- retention-by-sceneSkill
- setupSkill
- upload-privateSkill
- write-scriptSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 74
After publishing, test the thumbnail first, then the title, never both at once.
≈ “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.
Files
Check the thumbnail
The title and the thumbnail are one pairing: the viewer sees them together, so they say different things and together make one promise that the opening of the video keeps. Thumbnails are judged at feed size, never full size. Code measures what it can; you and the creator judge the rest from the preview.
$SKILL below is this skill's folder. Rules and their reasons: $SKILL/references/thumbnail-rules.md.
1. Make 2-3 real candidates
Real alternatives: a different picture or idea, not one frame with three wordings. Keep the text short (up to 3
lines, about 12 characters each) and draw it with code in the channel's font. If an image model drew the text,
check every letter. Every number on a thumbnail comes from facts.json, and nothing on it is made up: no fake
screenshot, headline, chart or quote.
2. Run the check
python $SKILL/scripts/thumb_check.py thumbs/a.jpg thumbs/b.jpg thumbs/c.jpg --project <video> \
--text "SOLD AT|THE TOP"Lines are split by |. Give one --text for all images or one per image, in order. The title comes from
upload.json (or --title). It prints the candidates side by side and writes qa/thumbnail-check.json and
qa/thumbnail-preview.png.
| Check | Fails when | Warns when |
| Format | not JPG/PNG, or the bytes do not match the extension | the PNG has transparency |
| Shape and size | over 5% off 16:9, under 640 px wide, over the upload limit | under 1280x720, slightly off 16:9 |
| Dark area | - | over 50% of the frame near-black at feed size (--dark-max) |
| Clutter | - | more than 14 colour clusters at feed size (--clutter-max) |
| Text | - | over 3 lines, a line over 12 characters (--max-line-chars), longest line over 2x the shortest, abbreviations like $6BN |
| Title overlap | - | the thumbnail text shares content words with the title |
Exit 1 means a candidate failed: fix or drop it. Warnings do not block. Each one is a question for step 3.
3. Judge at feed size
Open qa/thumbnail-preview.png. Every candidate appears at about 210, 320 and 630 px wide, first on YouTube's
dark background and then on the light one. Judge the 210 and 320 px columns:
- Can you read the text in one glance, and does it add to the title instead of repeating it?
- Is there one subject that wins, or several small things competing?
- Does a background that looked rich at full size turn into mud?
- Does the opening of the video show what this thumbnail promises in its first seconds?
Show the creator the preview and your pick with one line of reasons. The creator chooses.
4. Record the choice
Save the chosen file as the project's thumbnail and fill upload.json:
"thumbnail": "thumbnail.jpg",
"thumbnail_text": ["SOLD AT", "THE TOP"]Then run the check once more on the final pair:
python $SKILL/scripts/thumb_check.py --project <video>qa/thumbnail-check.json now carries verdict, thumbnail and its sha256. Next comes the upload-checker
agent. A thumbnail changed after its UPLOAD-GATE review needs a new review, because the upload binds to its
hash.
After publishing, test the thumbnail first, then the title, never both at once. Read the testing rules in
references/thumbnail-rules.md before starting a test.
Report back
The candidates with their verdicts and warnings in one table, the preview path, your pick and why in one line, and the next step: "creator picks", or "run upload-checker".
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.