Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Apply a project's DESIGN.md (Google Stitch format + OmD v0.1 brand-philosophy layer) as authoritative brand context when generating or modifying UI. Use whenever the user asks for UI work — new components, styling changes, layout, microcopy, empty/error states, motion — and a DESIGN.md exists at the project root. Also use when the user asks to "apply the brand", "match the design", "follow the design system", or references OmD, DESIGN.md, or the brand explicitly.
.claude/skills/kwakseongjae-omd/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 22% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 51% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -38% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -7% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -4% | 0% |
ROUTER_ONLY_NO_DESIGN_MD_WRITES
This skill preserves old omd entrypoints while the active workflows live in omd:apply, omd:autopilot, and omd:init. It classifies the available design context, transfers the original request to one canonical workflow, and then stops. Do not duplicate a target workflow here.
Inspect these files without mutating them:
.omd/system/manifest.json with format: design-md-core, format_version: 2.0.0, profile: portable-core, and authority.canonical: system-graph. Only after the target workflow validates the manifest, graph, projection, and exact bidirectional hashes is .omd/system/graph.json canonical. Root DESIGN.md remains its complete, vendor-neutral portable projection.
profile: migration-candidate is non-authoritative. Keep the sourceDESIGN.md named by its authority record canonical while the candidate is reviewed. Never use the candidate as a Bound System.
DESIGN.md independently for the rules it actually expresses. It does not require sidecars, an OmD install, or this skill to be useful. Do not claim graph authority or a conformance level that has not been validated.
DESIGN.md remains a read-only compatibilityinput. If root DESIGN.md is absent, .stitch/DESIGN.md may be used only as the same read-only fallback. Apply declared rules, omit unknown values at the smallest boundary, and never silently rewrite or upgrade either file.
A stale binding invalidates graph authority, not the readable Markdown. Never promote reference observations, placeholders, or generic defaults into project facts.
bounded product work -> transfer the unchanged request and authority classification to omd:apply.
autonomous design-system-plus-implementation work -> transfer to omd:autopilot.
refactor, or refresh without product implementation -> transfer to omd:init.
omd:harness; do not convert that request into Autopilot.
Use the runtime's skill dispatch when available. Otherwise locate the selected skill in the active channel, read its SKILL.md completely, and follow it. Pass through the user's original brief, preferences, and known authority facts so the user is not asked to repeat them.
This compatibility shim may inspect and route only. It must not create, edit, copy, move, rename, delete, or project DESIGN.md; it must not edit .omd/system; and it must not run a writer or migration command itself. The selected canonical workflow owns every write, checkpoint, migration proof, and validation step.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 9,851 | 6,903 | -30% | 1 | 1 | 0% | 1,666 | 2,031 | +22% | 0 | 0 | — |
case-02 | fail→fail | 5,350 | 7,117 | +33% | 1 | 1 | 0% | 280 | 1,234 | +341% | 0 | 0 | — |
case-03 | fail→pass | 8,731 | 12,293 | +41% | 1 | 1 | 0% | 1,497 | 2,263 | +51% | 0 | 0 | — |
case-04 | fail→pass | 12,602 | 3,360 | -73% | 1 | 1 | 0% | 2,090 | 1,305 | -38% | 0 | 0 | — |
case-05 | fail→pass | 11,176 | 5,985 | -46% | 1 | 1 | 0% | 1,909 | 1,781 | -7% | 0 | 0 | — |
case-06 | fail→pass | 10,204 | 4,460 | -56% | 1 | 1 | 0% | 1,538 | 1,481 | -4% | 0 | 0 | — |
case-07 | fail→pass | 35,695 | 3,049 | -91% | 1 | 1 | 0% | 8,226 | 1,319 | -84% | 0 | 0 | — |
case-08 | fail→pass | 11,434 | 5,620 | -51% | 1 | 1 | 0% | 1,873 | 1,826 | -3% | 0 | 0 | — |
case-09 | pass→pass | 14,404 | 3,536 | -75% | 1 | 1 | 0% | 2,232 | 1,346 | -40% | 0 | 0 | — |
case-10 | fail→pass | 11,938 | 5,078 | -57% | 1 | 1 | 0% | 1,906 | 1,703 | -11% | 0 | 0 | — |
case-11 | pass→pass | 8,583 | 3,325 | -61% | 1 | 1 | 0% | 1,448 | 1,327 | -8% | 0 | 0 | — |
case-12 | fail→pass | 11,181 | 5,940 | -47% | 1 | 1 | 0% | 1,748 | 1,831 | +5% | 0 | 0 | — |
case-13 | pass→pass | 12,505 | 8,001 | -36% | 1 | 1 | 0% | 1,991 | 1,314 | -34% | 0 | 0 | — |
case-14 | fail→pass | 20,915 | 9,111 | -56% | 1 | 1 | 0% | 5,038 | 2,451 | -51% | 0 | 0 | — |
case-15 | fail→pass | 10,930 | 7,483 | -32% | 1 | 1 | 0% | 2,026 | 1,400 | -31% | 0 | 0 | — |
case-16 | pass→pass | 7,658 | 2,359 | -69% | 1 | 1 | 0% | 1,282 | 1,184 | -8% | 0 | 0 | — |
case-17 | pass→pass | 2,187 | 7,096 | +224% | 1 | 1 | 0% | 288 | 2,000 | +594% | 0 | 0 | — |
case-18 | pass→pass | 10,587 | 5,999 | -43% | 1 | 1 | 0% | 1,913 | 1,767 | -8% | 0 | 0 | — |
case-19 | pass→pass | 9,225 | 3,830 | -58% | 1 | 1 | 0% | 1,495 | 1,546 | +3% | 0 | 0 | — |
case-20 | fail→pass | 10,735 | 4,048 | -62% | 1 | 1 | 0% | 1,815 | 1,583 | -13% | 0 | 0 | — |
case-21 | pass→pass | 2,903 | 3,865 | +33% | 1 | 1 | 0% | 556 | 1,533 | +176% | 0 | 0 | — |
case-22 | pass→pass | 2,463 | 3,237 | +31% | 1 | 1 | 0% | 458 | 1,378 | +201% | 0 | 0 | — |
case-23 | pass→pass | 2,429 | 2,846 | +17% | 1 | 1 | 0% | 449 | 1,249 | +178% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted, and 22 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +52 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/20/2026 | +36% |
Other measured skills in the registry, with their headline benchmark lift.