Compile screenshots, URLs, Figma exports, or an existing UI into a project-local StyleSeed output grammar with evidence, tokens, confidence, anti-patterns, and a validation screen.
SkillScore86Maintained
Automatic
◆Design & UX Review pack
Last verified 8/18/2026
What it changes
Passed 19 of 22 test cases with the skill, and 12 of 22 without it.
Without skill55%
With skill86%
Checks it passed that the model alone missed
Executes /ss-score during validation
Updates STYLESEED.md to select reference:<slug>
Executes /ss-verify when the screen can be rendered
and 9 more in the full report
7 fixed12 passed either way3 still failing0 broke
Pass-rate lift
+32 pp
Tokens
-6%
Agent turns
0%
Response time
-32%
Model gemini-3.6-flashMeasured 8/18/2026v219 of 22 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 7 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 7
Prompt
We are configuring a registry-based project setup containing project.json and artifacts/index.json in our .styleseed directory. When resolving a specific artifact ID like dashboard-v1, should the system pull from the legacy bundle or read strictly from the artifact bundle and manifest paths?
What was checked, without and with the skill
Checked
Without skill
With skill
Reads strictly from .styleseed/bundles/<artifact-id>.md and .styleseed/manifests/<artifact-id>.json
Missed
Passed
Refuses fallback to global legacy bundle when registry configuration files exist
Passed
Passed
case-04 · gemini-3.6-flash · 8/18/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md32 linessingle file
Compile references into a design grammar
Registry-first artifact boundary
When .styleseed/project.json and .styleseed/artifacts/index.json exist, resolve the requested artifact ID first, then read only .styleseed/bundles/<artifact-id>.md and .styleseed/manifests/<artifact-id>.json. Never fall back to the global legacy bundle for a registry project. Legacy projects may use .styleseed/effective-rules.md only when no registry exists.
Do not merely imitate the supplied screen. Read PRODUCT-PRINCIPLES.md, RULESETS.md,
ADAPTERS.md, and REFERENCE-COMPILER.md, then execute the compiler pipeline in full.
Workflow
Ingest every supplied reference and assign evidence IDs (R1, R2, ...). If a URL can be
rendered, capture it. If an image cannot be seen, stop claiming visual analysis for it.
Separate references by surface adapter and job. Do not merge a marketing homepage with an app
dashboard unless the user explicitly wants both languages connected.
Observe and measure the twelve axes. Every rule cites evidence IDs and a confidence level.
Resolve contradictions instead of averaging them. Explain any material choice briefly.
Choose the nearest built-in fallback grammar and adapter, then compile the project artifacts required by
REFERENCE-COMPILER.md under .styleseed/rulesets/<slug>/.
Update STYLESEED.md to select reference:<slug> while preserving bounded brand choices.
Validate transfer: apply the grammar to one representative screen not present in the source
set, run /ss-score, then /ss-verify if renderable. Fix and repeat to the gate floor.
Report what was learned, what remains low-confidence, where the artifacts live, and the
validation result.
Rules
Visible evidence before adjectives. “Calm” is not a rule; the spacing, type, color, hierarchy,
and motion decisions that create calm are rules.
Never copy protected assets, brand marks, or prose.
One reference produces a provisional grammar. Say so and keep confidence low where warranted.
The generated grammar is project-local. Do not edit RULESETS.md or publish it as built-in.
Core invariants and accessibility remain non-negotiable.