shawn-sandy/style-tune

/style-tune

Route a natural-language aesthetic intent ("warmer button", "softer card", "deeper accent") to either a theme-role OKLCH edit or a component SCSS token edit. The full workflow for each layer lives in the parent skill; this router handles intent parsing and dispatch.

Step 0 — Exit plan mode

Call ExitPlanMode before parsing intent. Both downstream paths write to disk — /theme-update edits theme CSS, component-layer edits SCSS in place, and OKLCH scripts run via Bash. Plan mode blocks all of these.

Stay in plan mode only when the user explicitly asked for a preview ("show me the deltas", "don't apply yet"). In that case, narrate the resolved (modifier, token-family, layer) tuples from Step A without invoking Write/Edit/Bash, and wait for approval.

Step A — Resolve intent

A0. Load the intent vocabulary

Read ${CLAUDE_PLUGIN_ROOT}/skills/style-tune/references/intent-vocabulary.md. Each row maps a modifier (and synonyms) to a token family + canonical delta + "var-only fallback" route.

A1. Tokenize the prompt

Parse the prompt for vocabulary matches. Each match yields a (token-family, delta, layer-hint) tuple. Record all matches before dispatching.

A2. Resolve the subject and dispatch

| Subject in prompt | Layer | Follow | |---|---|---| | primary, accent, danger, warning, info, success, brand, theme, app | theme | styles/SKILL.md — Style-Tune Mode | | button / btn, card, alert, dialog, input, nav, form | component | components/SKILL.md — Style-Tune Mode | | bare this / it / the component / everything | ambiguous | AskUserQuestion |

"This button" / "the button" / "buttons" always map to a global SCSS edit — never inline style={...} props.

A3. Confirm low-confidence intents

Use AskUserQuestion (≤ 4 options) when:

• The prompt is a single bare adjective with no subject ("warmer").
• A modifier has multiple plausible token families ("louder" → border vs shadow vs both).
• Modifiers contradict each other ("calmer but bolder").
• Subject is "alert" and modifier is a colour modifier — ask whether to tune base tokens only or include all four severity variants.

For unambiguous prompts, skip A3 and dispatch immediately.

Step F — Summary

After the layer workflow completes, print:

Layer:       <theme | component | both>
Files:       <list>

Modifier         Token / Role               Old           New           Status
warmer           --color-primary            #2563eb       #3265ec       accepted
warmer           --color-primary-hover      #1e4dc7       #294fc8       accepted
softer           --btn-radius               0.375rem      0.5625rem     accepted
calmer           --color-danger             #dc2626       #d8413b       reverted (contrast)

Notes:
  - <any var-only fallback routing notes or drift warnings from the layer workflow>

Next:
  - Try "now go a touch sharper" to dial back radius.
  - Or use /theme-update for explicit hex values.

Always include the "Next" hint so users know iteration is cheap and bounded.