# Li Design Dna

> Use when a frontend or presentation task needs corpus-backed style, palette, typography or stack guidance, selected design-profile tokens, or static design and observed-color contrast checks.

- Skill: `jokerman89/li-design-dna` (Agent Skill)
- Install (CLI): `npx get-skill add jokerman89/li-design-dna`
- Raw SKILL.md: https://api.skillmd.com/api/skills/jokerman89/li-design-dna/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Design & Media
- Author: jokerman89 (https://skillmd.com/u/jokerman89)
- Updated: 2026-11-15
- Page: https://skillmd.com/skills/jokerman89/li-design-dna

---


> **Lintel on GitHub Copilot.** Generated from `skills/design-dna/SKILL.md`; edit the canonical file, then run
> `li-copilot init`.
> - **Resource root:** `../../..` from this skill's base directory (the Lintel source with `bin/`,
>   `lib/`, `skills/`). Write plans, state and evidence into the working repository's `.claude/`
>   tree, never into the resource root.
> - **Skill-relative paths:** this skill's own `scripts/`, `references/` and `data/` folders (and a
>   `<base>` that the workflow defines as its own directory) mean
>   `../../../skills/design-dna/` in the Lintel source, not this generated folder.
>   `${LINTEL_SKILLS_DIR:-skills}` means the skills root, `../../../skills`. A `bin/li-run` step
>   runs in the working repository, so use `$LINTEL_SKILLS_DIR/design-dna/` there.
> - **Shell steps:** run Bash snippets with Bash (Git for Windows' `bash.exe` on Windows, never
>   `System32\bash.exe`). Save a snippet to a temporary `.sh` file and run
>   `bash "<resource root>/bin/li-run" <file>`; it prepares `LINTEL_SOURCE_ROOT`, `LINTEL_REPO_ROOT`
>   and the profile context.
> - **Tools:** Read=`view`, Write=`create`, Edit=`edit`, Bash=`bash`/`powershell`, Grep=`grep`,
>   Glob=`glob`, AskUserQuestion=`ask_user`, TodoWrite=the plan checklist, Task or a named role=`task`
>   with that custom agent, WebFetch=`web_fetch`.
> - **Other Lintel workflows** are native skills: invoke `/li-<name>` rather than reading their
>   files. Named roles such as `CodeReviewer` are custom agents.

You are the `design-dna` module — Lintel's design knowledge + retrieval layer (ADR-0015/0016).

## What this skill does

Retrieval before generation: instead of free-associating design choices from model memory, every
visual decision starts from a curated, searchable corpus (consumed from UI/UX Pro Max v2.5.0, MIT
— see ATTRIBUTION.md) plus the active design profile (default: `anthropic-default`, the warm
ink-and-paper system). This module makes NO rendering decisions — it feeds the decision layer
(`frontend-*`) and gates the rendering layer (`generate-*`), preserving the L-004 split.

The [direct design contract](../../../skills/design-dna/references/design-contract.md) links retained brief,
profile, corpus, renderer and review data through one schema/helper. Reuse it from
every frontend/web consumer; source parsing or a populated spec is not rendering.

## When to use

- Auto-invoked: `/li-frontend-design` Step 1.5 (required), `generate-web`/`generate-app` stack
  pass, `frontend-design-review` validation and built-UI critique
- Solo: "what style fits a fintech dashboard?", "palette for a healthcare app", "React Native
  list-performance rules", "validate this HTML"

## When NOT to use

- Pure backend/API/infra work with no visual surface
- Brand identity *extraction* from existing artifacts — that's `/li-frontend-style-extract`

## Sub-capability dispatch

`<base>` = this skill's directory — the base dir injected at invocation (in-repo:
`skills/design-dna`). Sibling skills resolve it as `"${LINTEL_SKILLS_DIR:-skills}/design-dna"`
where `LINTEL_SKILLS_DIR` is the plugin's skills root (the invoked skill's own base-dir parent).
All searches: python3, stdlib-only, exit 0 + markdown to stdout.

| Capability | Invocation | Returns |
|---|---|---|
| `search` | `python3 "<base>/scripts/search.py" "<query>" [--domain style\|color\|chart\|landing\|product\|ux\|typography\|icons\|react\|web] [-n 3]` | Top-3 corpus rows, token-budgeted (300-char truncation) |
| `system` | `python3 "<base>/scripts/search.py" "<product> <industry> <keywords>" --design-system [-p "<Project>"] [-f markdown]` | Composed recommendation: pattern + style + palette + typography + reasoning + anti-patterns + checklist |
| `stack` | `python3 "<base>/scripts/search.py" "<query>" --stack <react\|nextjs\|vue\|svelte\|astro\|swiftui\|react-native\|flutter\|nuxtjs\|nuxt-ui\|html-tailwind\|shadcn\|jetpack-compose\|threejs\|angular\|laravel>` | Do/Don't/Code-Good/Code-Bad/Severity rules for the stack |
| `persist` | `system` + `--persist [-p "<Project>"] [--page "<page>"] [-o docs/design-system]` | `MASTER.md` + `pages/<page>.md` with self-describing precedence (page overrides master) |
| `slide` | `python3 "<base>/scripts/search.py" "<emotion\|goal\|keyword>" --slide <strategy\|layout\|layout-logic\|color-logic\|typography\|copy\|background\|chart>` | Presentation decision rows: emotion→color, goal→layout, narrative strategy + Duarte sparkline-beats, slide copy formulas. Consumed by generate-ppt |
| `tokens` | `python3 "<base>/scripts/emit_tokens.py" --profile "<verified-selected-profile-path>" [--out design-tokens.css]` | Three-layer design-tokens.css (primitive → semantic → component) from the selected pack/bundled asset |
| `validate` | `python3 "<base>/scripts/validate_design.py" <file.html> [--profile "<verified-selected-profile-path>"]` | Exit 1 on hard violations (zoom-disable, killed focus, emoji icons…); warnings (off-palette, token discipline) listed |
| `contrast` | `python3 "<base>/scripts/measure_contrast.py" --foreground "<observed color>" --background "<observed solid color>" --text-size <normal\|large>` | Unrounded `ratio`/`text_size` observation; exit 0 meets AA, 1 below threshold, 2 refused/unverified; no browser or clearance claim |
| `profile` | Read `<base>/profiles/<id>.yaml` (resolution below) + `references/token-architecture.md` for the layering doctrine | The active token set + doctrine |

Domain auto-detects from the query when `--domain` is omitted. Multi-dimensional queries work
best: product + industry + tone + density ("entertainment social vibrant content-dense").

**Slide controlled vocabulary** (query with the exact word so BM25 surfaces the row): emotions =
`curiosity frustration fear relief trust urgency confidence warmth evaluation hope narrative clarity interest`;
goals = `hook problem agitation solution proof social comparison traction cta team pricing demo vision timeline features`.

## Profile resolution

Use P07's already verified profile reference and explicit source/target/profile
roots. `design_contract.profile_asset(record, config)` resolves the selected
asset and records its bytes. Required-policy load/drift errors block; never source
a personal-home helper or silently choose anthropic-default after an error.

Packs override by declaring `design.profile` + shipping `profiles/<id>.yaml` in the pack dir
(checked first), falling back to `<base>/profiles/`. The pack contract is untouched — the field
is additive-by-convention; the neutral `_default` pack declares nothing and gets anthropic-default.
The fallback is to the **same selected name**. An absent/invalid custom asset is
not permission to switch brands. `load_design` verifies the pin and asset hash again.

Profile precedence vs corpus: **brief > profile > corpus search hit.** The profile is the house
default; a corpus palette/style hit replaces profile tokens only when the brief asks for something
the profile doesn't cover (e.g. a product-specific palette for a customer build). When the
operator pins a direction ("EXACTLY brand X"), the brief wins over everything — record the chosen
tokens in the persisted MASTER.md so the decision sticks.

## Degradation: python3 absent

The corpus is plain CSV — readable without the engine. Map the need to its file and Read/Grep it
directly: styles→`data/styles.csv`, palettes→`data/colors.csv`, products+reasoning→
`data/products.csv`+`data/ui-reasoning.csv`, fonts→`data/typography.csv`, UX→
`data/ux-guidelines.csv`, charts→`data/charts.csv`, landing→`data/landing.csv`, icons→
`data/icons.csv`, React perf→`data/react-performance.csv`, app interface→`data/app-interface.csv`,
stacks→`data/stacks/<stack>.csv`. Pick max 3 relevant rows; respect the same anti-patterns and
severity columns. Validation falls back to the review checklist in `frontend-design-review`.

## The non-negotiables (carried into every consumer)

1. No emoji as icons — SVG only (Lucide, Heroicons)
2. Body-text contrast >=4.5:1 (large text >=3:1), both themes, verified not assumed;
   use [measured text contrast](../../../skills/design-dna/references/design-contract.md#measured-text-contrast)
   on actual observed colors, not a palette label or static-validator success
3. Visible keyboard focus — never `outline: none` without replacement
4. Micro-interactions 150-300ms; transform/opacity only; reduced-motion respected
5. Touch targets >=44px; `cursor: pointer` on clickables
6. Hover states must not shift layout
7. One primary CTA per screen; accents never carry body text (profile rule)

## Integration

**Reads:** `data/*.csv`, `profiles/*.yaml`, active pack via `resolve_pack_field design.profile`
**Writes:** `docs/design-system/MASTER.md` + `pages/*.md` in the target repo (persist), audit:
`.claude/runtime/audit/design-dna-runs.jsonl`
**Consumed by:** `/li-frontend-design` (Step 1.5), `/li-frontend-typography`,
`/li-frontend-motion`, `/li-generate-web`, `/li-generate-app`, `/li-frontend-design-review`

## Reusable patterns

Follow the [reusable pattern consumer contract](../../../skills/pattern/references/consumer-contract.md).
Retrieval, validators and the non-negotiables above are unchanged. When a consumer supplies a
ready resolution, its mandatory clauses bound the choices and its defaults fill only choices the
brief did not decide, ahead of the profile and corpus. Profile and corpus files are never
modified to implement a pattern. A pure palette or token read needs no cycle.

## Status protocol

- **DONE** — search/compose/validate returned, results surfaced
- **DONE_WITH_CONCERNS** — search returned 0 hits (query too narrow — retry with different keywords) or validate passed with warnings
- **BLOCKED** — validate exit 1 (hard violations listed; fix before ship)

