agentleFS
whinstudiowhinstudio/skills/game-look/SKILL.md

game-look

Skill0 starsChanged 3 days agoMITView on GitHub

Use when the art of a Unity 6 mobile game must look like one game - new or mismatched characters, backgrounds, icons or UI art, a restyle, or a request to make the visuals consistent or to match a reference image.

The file

We split this skill into the rules it gives an agent

11 rules. About 0 are common, 5 are distinctive, and 6 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 6

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use the image tool you have to generate assets.

    line 21
  2. Very specific:

    Run game-smoke (pass the platform and the screens to cover).

    line 26
  3. Very specific:

    If the project has no image for the thing the request names, say so first.

    line 48
  4. Very specific:

    When no user can answer in this run, stop after this step and report BLOCKED: need <image or confirmation>.

    line 53
  5. Very specific:

    Write design/look.md, one screen at most:

    line 57
  6. Very specific:

    Every new or reworked asset is generated with the reference image attached and the text of look.md in the prompt.

    line 77
  7. Very specific:

    Put the asset in the game, keeping the file or prefab references that exist.

    line 92
  8. Distinctive:

    Do the new assets match the reference, in the same frame as the old ones?

    line 102
  9. Distinctive:

    Do touch targets and text stay clear of the art, with enough contrast (at least 4.5:1 for text, 3:1 for large text)?

    line 103
  10. Distinctive:

    After the third, only name the change and report what is left.

    line 114
  11. Distinctive:

    Only the user can judge how it feels.

    line 124

≈ “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

Game look

One reference image, one page of rules, every new asset made from both, and a device screenshot before and after. Without a fixed reference each asset drifts from the last one.

A visual task is not done without a screenshot. The editor's Game view does not count. The screenshot comes from game-smoke (step 4). If it returns Visual: NOT DONE on a requested platform, this skill's work is not done either; say so and name the blocker.

Done means all three questions of step 4 are yes and game-smoke shows no new FAIL against the before run. A FAIL present in both the before and after runs does not block done for the art task; this overrides the Caller: line of game-smoke for that case only. Report that FAIL by name.

Image tools

Use the image tool you have to generate assets. If none is available, the loop still works with images the user supplies: ask for them, and run the same steps on those.

1. Before shots

Run game-smoke (pass the platform and the screens to cover). Keep the PNG paths it cites as the "before" set. Read the Art direction line of design/brief.md when it exists.

The next game-smoke run deletes the test results, so copy the before run's now (a missing file is fine):

mkdir -p qa/look
cp Builds/editmode.xml qa/look/before-editmode.xml 2>/dev/null
cp Builds/playmode.xml qa/look/before-playmode.xml 2>/dev/null
true

2. Fix one reference image

Exactly one image defines the look. Either the user names one, or you pick one existing asset (or generate one) and show it. Do not average several. Ask the user to confirm it when the choice is open; do not go on with two candidates.

A white or tinted glyph (the UI kit's icons) is not a reference; use a coloured image or a screenshot of the intended look, or ask.

If the project has no image for the thing the request names, say so first. Art enters through the template's skin hooks (README, "Look"): Resources/Branding/menu-background.png and Resources/GameSkin.json (backgrounds.game, backgrounds.menu). A character drawn in code needs a code change, so ask whether that is wanted.

When no user can answer in this run, stop after this step and report BLOCKED: need <image or confirmation>. Continue with stand-ins only if told to, and name every stand-in in look.md and in the report.

Write design/look.md, one screen at most:

  • Reference: the image's path in the project
  • Palette: hex values. One pixel of an image as hex: ffmpeg -i <img> -vf crop=1:1:x:y -f rawvideo -pix_fmt rgb24 - | xxd -p
  • Shape language: for example rounded and chunky, or angular and thin
  • Line: weight and colour, or none
  • Light: direction and softness
  • Perspective: for example flat side view, top-down
  • Detail level: how much texture and small shapes
  • Never: what must not appear (other styles, text in art, a second outline weight)

The template's style family (theme: bright, neon, pastel, bold, nature, clean) is the UI's look. Start from the applied family (theme.family in .cmpose/generation.json); name another only when the reference clearly belongs to it. Make the palette in look.md agree with the family. The family is described in "Look: style family, fonts and the UI kit" of the template's README, and .cmpose/themes.json. If none fits, say so in look.md instead of bending the art.

3. Make assets from the reference and the page

Every new or reworked asset is generated with the reference image attached and the text of look.md in the prompt. Same style and size family as the reference; the canvas is the screen's aspect for a background and the sprite's own for a character. Check each asset against the reference side by side with a contact image you can read:

mkdir -p qa/look
ffmpeg -y -i ref.png -i new.png -filter_complex \
  "[0:v]scale=-2:512[a];[1:v]scale=-2:512[b];[a][b]hstack" qa/look/contact-<asset>.png

Both sides are scaled to the same height. Read the PNG and note what differs in palette, line, light and detail. Fix the asset, not the page, unless the page was wrong; then change the page and say so. Without ffmpeg, sips or ImageMagick convert +append after resizing does the same.

Put the asset in the game, keeping the file or prefab references that exist. A sprite needs Sprite import (textureType: 8 in its .meta) and a path under a Resources folder when code loads it, or Resources.Load<Sprite> returns null without a message. A test that goes red because of the asset is a result: report it by name; do not edit the test.

4. Play-test

Each round is step 3 then step 4. Run game-smoke again for the same screens. Read the new PNGs and judge, at phone size:

  • Is the character, or the item the player must find, readable against the background?
  • Do the new assets match the reference, in the same frame as the old ones?
  • Do touch targets and text stay clear of the art, with enough contrast (at least 4.5:1 for text, 3:1 for large text)?

Compare the before and after runs check by check, not by the Verdict line alone: a FAIL verdict can stay while a new failure hides in it. For tests, compare failing test names. The smoke report may give only counts, so read the names from the <test-case> elements with result="Failed": compare qa/look/before-editmode.xml and qa/look/before-playmode.xml (step 1) with the new Builds/editmode.xml and Builds/playmode.xml. With no before copy of a test XML, every failing test in the new XML counts as new. A FAIL present in both is not caused by the art: say so and name it. A new FAIL, or a new failing test, is caused by it.

Then pick one change for what is worst. It is the next round's step 3. Maximum three rounds. After the third, only name the change and report what is left; do not loop "until it looks right".

Report

  • the reference image path and design/look.md
  • a BLOCKED: line from step 2, when one was reported
  • the rounds run and the one change of each
  • the before and after device screenshots, by path
  • what is left, and anything game-smoke marked NOT RUN or NOT ASSESSED

Only the user can judge how it feels; say so in one line. Do not commit the screenshots or push.

Out of scope

3D worlds and model or animation pipelines. Gameplay and UI layout changes beyond what the art needs.

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.