# Customize /taste for Your Workspace

Paste this entire file into Claude Code. Then paste the contents of `taste.md` directly below it. Send.

---

I'm forking the `/taste` skill from Russ Gardner's Jack's Community case study pack. The skill body is below this prompt. Before installing it, ask me 6 questions and customize the constitution so it grades against my taste, not his.

## Questions to ask (one at a time, wait for my answer before moving to the next)

1. **Your brand voice in 5-10 words.** Examples: "terse, technical, no marketing fluff." "Warm, narrative, restaurant editorial." "Playful, irreverent, gaming-adjacent." "Clinical, evidence-first, healthcare." This tunes the Content Quality domain in the constitution.
2. **Two reference URLs whose visual taste you align with.** What do you find beautiful and well-built. The skill uses these to anchor what "good" looks like for you.
3. **Two anti-references whose visual taste you reject.** What looks generic, dated, or AI-slop. The constitution explicitly bans the patterns these sites exhibit.
4. **Punctuation or patterns you never use.** Banned list. Examples I use: em dashes (U+2014), en dashes, smart quotes, the word "innovative", any sentence starting with "Just." If you don't have a list, the skill defaults to banning em dashes since that's the most reliable AI-generation tell.
5. **Top 5 binary checks you care about most.** From this list (or write your own):
   - Typography rhythm (heading sizes, line-heights, measure)
   - Contrast (dark theme legibility, hero overlay opacity)
   - Density (whitespace, breathing room, section padding)
   - Motion (easing curves, animation restraint, choreography)
   - Restraint (accent usage, color discipline)
   - Accent count (how often the accent color appears)
   - Asymmetry (one section breaks the grid)
   - Photo quality (real photos vs gradient placeholders)
   - Composition variety (no two adjacent sections the same)
   - Signature moment (one mid-page beat that breaks the rhythm)
   
   Pick 5. These get weighted highest in the report.
6. **The viewport sizes you ship to.** Default: 1440 desktop, 375 mobile. If you build for 4K displays, kiosks, watches, or tablets primarily, give those numbers. The skill takes screenshots at exactly the viewports you specify.

## What to do with my answers

After I've answered all 6, do the following:

1. **Read the skill body below** (everything after the `---` separator).
2. **Replace `{{DESKTOP_VIEWPORT}}` and `{{MOBILE_VIEWPORT}}`** with my answers from question 6.
3. **Update the constitution principles** based on my answers:
   - Question 1 (voice): tune the Content Quality domain. Edit principles 24-34 to reflect my voice. If I said "terse declarative", strengthen specificity rules. If I said "warm narrative", relax some of the "no comma chains" rules and add ones about emotional pacing.
   - Question 2 (positive references): inject my URLs as anchor references in a new "Anchor References" section near the top. The skill consults these when grading borderline cases.
   - Question 3 (anti-references): replace the "Never default-AI palette" placeholder with the actual patterns my anti-references exhibit. Be specific. Bad: "avoid generic AI sites." Good: "avoid the {anti-ref-1} pattern of three identical centered hero sections in a row" or "avoid the {anti-ref-2} pattern of full-bleed gradient backgrounds with no surface depth."
   - Question 4 (banned punctuation): replace the FILL IN block in Step 4 → "Banned Punctuation" with my exact list.
   - Question 5 (top 5 checks): mark these in the T1-T10 table with a `**` weight indicator and tell the report generator to flag them prominently. If my top 5 don't map to the existing T1-T10 list, add new evals (T11+) that match what I picked.
4. **Strip principles I disagree with.** Ask me at the end: "Are there any of the 34 constitution principles you want to strip or rewrite? Reading them now." Then list the principles by number and one-line summary. Wait for me to flag any. Strip or rewrite as instructed.
5. **Save the customized skill** to `~/.claude/skills/taste/SKILL.md`. Create the directory if it doesn't exist. If a file is already there, write the new version to `~/.claude/skills/taste/SKILL.md.new` and tell me to compare before overwriting.
6. **Show me the diff** between the template and my customized version. Highlight which principles got edited, which placeholders got replaced. Confirm I can run `/taste path/to/file.html` or `/taste https://example.com` in any future session.

## Bias toward action

Make sensible defaults if my answers are vague. Don't loop on clarifying questions.

If I'm new to taste auditing, my voice answer will be hand-wavy. That's fine. The constitution still works, and as I run the skill against real builds and give corrections, those corrections become candidate principles. Tell me to keep a `lessons.md` somewhere and append corrections as they come up. Over time those become the next round of customize.

---

(Skill body follows below the next separator. Read it, then run the steps above.)

---