agentleFS
Sign inStart free
yiyaw-labyiyaw-lab/agent-armor/taste/SKILL.md

taste

Skill1 starsChanged 5 months agoMITView on GitHub

Distill the user's revealed taste from real session transcripts into an enforceable personal rubric (~/.claude/TASTE.md), then apply it as a pre-delivery filter. Mines what they actually corrected and accepted — never what they say they prefer. Modes: induce | apply | audit.

23 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use council, ui-ux-pro-max and 1 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • MicrosoftSkill

    council

    Convene the persona panel — six orthogonal review lenses — on a target. Cold independent fan-out, debate-to-consensus, synthesized verdict with recorded dissent and a roster manifest.

    Also in 3 projects that use this

  • StrayDragonSkill

    ui-ux-pro-max

    UI/UX design intelligence. 50 styles, 21 palettes, 50 font pairings, 20 charts, 8 stacks (React, Next.js, Vue, Svelte, SwiftUI, React Native, Flutter, Tailwind). Actions: plan, build, create, design, implement, review, fix, improve, optimize, enhance, refactor, check UI/UX code. Projects: website, landing page, dashboard, admin panel, e-commerce, SaaS, portfolio, blog, mobile app, .html, .tsx, .vue, .svelte. Elements: button, modal, navbar, sidebar, card, table, form, chart. Styles: glassmorphism, claymorphism, minimalism, brutalism, neumorphism, bento grid, dark mode, responsive, skeuomorphism, flat design. Topics: color palette, accessibility, animation, layout, typography, font pairing, spacing, hover, shadow, gradient.

    Also in 3 projects that use this

  • BlockSkill

    frontend-design

    Create distinctive, production-grade frontend interfaces with high design quality. Use this skill when the user asks to build web components, pages, or applications. Generates creative, polished code that avoids generic AI aesthetics.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

2 rules. 0 are common, 1 is distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 21 edited versions we read keep these.

  • 1

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Guard against contamination: If the user *said* "I prefer X" somewhere but their reactions show they accept Y, record the conflict …

    line 14
  2. Very specific:

    Read ~/.claude/TASTE.md. Review your pending or most recent substantial output against the ACTIVE rules.

    line 19

“Common” is measured: kept by at least 6 in 10 of the 21 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Core law: revealed preferences only. Stated criteria ≠ revealed criteria — rubrics built by asking the user what they like do not predict their actual judgments. Never build or update the rubric from the user's self-description; harvest it from how they actually reacted.

/taste induce [N]

  1. Harvest. Find the newest N (default 10) session transcripts: ~/.claude/projects/**/*.jsonl by mtime. Spawn one subagent per transcript to extract USER-turn reaction moments, each returned as {quote, what_the_agent_had_done, signal}:
    • Corrections: "no", "actually", "instead", "don't", "why did you", "too long/short", partial redos, the user rewriting your output their way.
    • Approvals: "perfect", "yes", "love it", verbatim reuse of your output, immediate build-on-top.
    • Silent rejections: the user re-asks the same thing differently — the first answer missed; figure out what changed between ask one and ask two.
  2. Induce. Cluster moments into candidate rules. A rule needs ≥2 independent evidence moments (different sessions ideally); single moments are anecdotes — record as PROVISIONAL. Each rule: name, the do/don't in one sentence, quoted evidence with session refs, evidence count.
  3. Guard against contamination. If the user said "I prefer X" somewhere but their reactions show they accept Y, record the conflict explicitly and trust the behavior. Absence of objection is NOT preference — never induce a rule from silence.
  4. Write/merge ~/.claude/TASTE.md. Strengthen rules the new evidence confirms (bump n). Retire rules newer evidence contradicts — taste drifts; newest behavior wins. Keep ≤15 ACTIVE rules ranked by evidence count; overflow goes to a PROVISIONAL section. Mark machine-inferred rules (inferred, n=K) and NEVER delete or rewrite lines the user added by hand.
  5. Report: new / strengthened / retired rules, and the 3 most surprising findings — the gaps between what the user would say they want and what they demonstrably reward.

/taste apply (default)

Read ~/.claude/TASTE.md. Review your pending or most recent substantial output against the ACTIVE rules. Fix violations before delivering. Report which rules fired: rule → what changed. If TASTE.md doesn't exist, say so and offer /taste induce.

/taste audit

Grade this session's outputs against TASTE.md: | rule | conformance | example from this session |. The worst offenders are candidates for promotion into CLAUDE.md or a memory file — where they get loaded every session instead of on demand.

Rules

  • Evidence is quoted user words. Never paraphrase a quote into something stronger than what was said.
  • The rubric describes the user, so it belongs to the user: TASTE.md is theirs to edit, and hand edits outrank inferences.
  • This file and TASTE.md stay local. Publishing either is a disclosure decision the user makes explicitly, not a default.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free