agentleFS
jqaisystemsjqaisystems/jqai-ai-skills/skills/thumbnail-legibility-check/SKILL.md

thumbnail-legibility-check

Skill3 starsChanged 6 months agoMITView on GitHub

Judge thumbnail or cover-image candidates at the size viewers actually see them, using measured pixels plus glance, grayscale, real-feed and decoy tests, and pick a winner through pairwise duels. Use when choosing between thumbnail options, when an image looks great full size but weak in a feed, when checking whether text or a face survives at small size, or when deciding which cover art to publish.

  • Installs packages

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 4 are distinctive, and 0 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.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  • could be automatic: a setting could enforce it
  1. Distinctive:

    Run each candidate through all four.

    line 50
  2. Distinctive:

    Real feed: Place the candidate in the newest slot of the actual feed it will live in, surrounded by genuinely published neighbours.

    line 56
  3. Distinctive:

    Ask "which would you click", never "which is better".

    line 74
  4. Distinctive:

    Test in the real feed, not against a blank background.

    line 90

≈ “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.

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

Thumbnail Legibility Check

Thumbnails are designed at 1280 pixels wide or larger and seen at about 168. Almost every bad thumbnail decision comes from judging it at the wrong size.

This skill measures what actually survives, then puts the candidates through the conditions they will really face.

When to use

  • Choosing between several thumbnail, cover, or preview-image candidates.
  • An image that looks strong at full size and disappears in a feed.
  • Checking whether on-image text or a face reads at small size.
  • Comparing a new candidate against what already performs.

Step 1: measure

python scripts/legibility.py path/to/candidates --recursive

Requires Pillow (python -m pip install Pillow). Add --json for machine output. Every candidate is cover-cropped and downscaled to 168x94, then measured:

MetricWhat it meansRisk below or above
brightMean luminanceUnder 45 sinks into a dark feed; over 200 washes out on white
contrMean per-channel spreadUnder 40 reads as flat at small size
popDistance between strongest and weakest channelUnder 12 means muted, no colour signal in a feed
glanceContrast surviving a heavy blurLow means it turns to mush during a fast scroll
grayContrast with colour removedLow means the design leans on hue alone for separation

These are risk signals, not verdicts. A dark thumbnail can be the right choice when the channel's whole look is dark. Use the numbers to know what you are choosing, not to choose for you.

Step 2: look at it small

Render each candidate at real display size and actually look. The measurement catches what is measurable; your eye catches a face cropped at the jaw, or two candidates that are indistinguishable at that size.

Step 3: the stress tests

Run each candidate through all four. They fail differently.

  • Glance blur. Blur heavily. What is still readable is what a scrolling viewer gets. If nothing survives, the composition has no focal hierarchy.
  • Grayscale. Strip colour. Structure should hold. If it collapses, the design is doing its work with hue alone, which fails against every background.
  • Real feed. Place the candidate in the newest slot of the actual feed it will live in, surrounded by genuinely published neighbours. This is the only test that catches "it looks like everything else we publish" and "it clashes with the three above it".
  • Decoy feed. Hide all candidates among unrelated images and ask someone to find yours. If they cannot, it does not stand out, whatever the numbers say.

Also check both light and dark interface modes, and all three placements: the large player-adjacent size, search results, and the narrow sidebar.

Step 4: pick with duels, not a lineup

Looking at seven options at once produces a tie. Instead, show two at a time and ask which one you would click. Six or so duels across seven candidates produces both a clear winner and a full ranking.

Two rules that keep this honest:

  • Ask "which would you click", never "which is better". People answer the second question with their taste and the first with their behaviour.
  • Show the candidates styled the way the platform will show them, with the title text next to them. A thumbnail is never seen alone.

Step 5: keep the results

Record which layout and which colour direction won, every time. After a handful of decisions the pattern is real data about your audience, and future candidate sets can lead with the combination that keeps winning. Without the record, every choice starts from zero.

Rules

  • Judge at display size, never at design size.
  • Numbers flag risk; they never pick the winner.
  • Test in the real feed, not against a blank background.
  • Pairwise duels, phrased as "which would you click".
  • Report the flagged candidates and why, rather than a single score.

Output

A table of measurements per candidate, the flagged risks with plain-language reasons, the results of the four stress tests, and a ranked order with the winner first.

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.