skills

# lroolle/skills

> Opinionated agent skills for Claude Code, Codex, and any
> agent that reads SKILL.md folders: UX design, animation, HTML
> artifacts, decision spread, goal contracts, knowledge vaults,
> llms.txt maps, and the craft of making skills themselves. Every
> protocol opens with a gate that can say no. Apache-2.0. Each skill
> is one self-contained folder under skills/<name>/.

Install: `npx skills@latest add lroolle/skills` -- or copy any
skills/<name>/ folder into ~/.claude/skills/ (or your agent's
skill root). No build step, nothing to install -- the wiki-it and
llms-txt checkers are stdlib-only Python.

## Skills

One line per skill: what fires it. Fetch the raw SKILL.md to use it.

- [design-system](https://raw.githubusercontent.com/lroolle/skills/main/skills/design-system/SKILL.md): give a repo a design system -- install the material contract (DESIGN.md tokens, TASTE.md scars, runnable bans), then build/judge/move against it; choose fonts, icons, and references by function
- [kiln](https://raw.githubusercontent.com/lroolle/skills/main/skills/kiln/SKILL.md): design or restyle a frontend surface; audit an interface; rescue generic AI-looking output
- [taste](https://raw.githubusercontent.com/lroolle/skills/main/skills/taste/SKILL.md): judge whether a design change is actually better -- verdict on a UI diff or redesign; decompose "make it like X" references into mechanisms and acceptance criteria; record rejections as reusable scars
- [animate-it](https://raw.githubusercontent.com/lroolle/skills/main/skills/animate-it/SKILL.md): implement or fix web animation -- easing, duration, springs, tool choice; review animation code
- [htmlize](https://raw.githubusercontent.com/lroolle/skills/main/skills/htmlize/SKILL.md): technical explanations that beat markdown -- one-page artifacts or multi-page briefing sites with question-led navigation, source bundles, diagram inspection, and integrity checks
- [parallax](https://raw.githubusercontent.com/lroolle/skills/main/skills/parallax/SKILL.md): decisions where the obvious answer being wrong is expensive -- spread genuinely different approaches, surface traps, commit
- [goldengoal](https://raw.githubusercontent.com/lroolle/skills/main/skills/goldengoal/SKILL.md): turn fuzzy intent into a goal contract for /goal or agentic loops; sharpen a draft goal against five gates
- [skillize](https://raw.githubusercontent.com/lroolle/skills/main/skills/skillize/SKILL.md): crystallize this session's workflow into a reusable skill
- [distill](https://raw.githubusercontent.com/lroolle/skills/main/skills/distill/SKILL.md): distill a book, transcript, or doc set into the few skills it actually earns
- [skillbun](https://raw.githubusercontent.com/lroolle/skills/main/skills/skillbun/SKILL.md): bundle skills plus their dependencies into one .skill archive
- [drop](https://raw.githubusercontent.com/lroolle/skills/main/skills/drop/SKILL.md): publish a deliverable to a live review URL and pull human verdicts and annotations back into the loop
- [burnish](https://raw.githubusercontent.com/lroolle/skills/main/skills/burnish/SKILL.md): audit, renovate, instrument, and promote a repo until real work reads as real in 30 seconds
- [wiki-it](https://raw.githubusercontent.com/lroolle/skills/main/skills/wiki-it/SKILL.md): build or operate a knowledge vault -- compile sources into reviewed, self-describing OKF bundles; orient, ingest, query, lint; work the review queue
- [llms-txt](https://raw.githubusercontent.com/lroolle/skills/main/skills/llms-txt/SKILL.md): generate, refresh, or check llms.txt -- the curated map a site, docs dir, repo, or vault publishes so agents orient in one fetch
- [compost](https://raw.githubusercontent.com/lroolle/skills/main/skills/compost/SKILL.md): doc hygiene by blind trial -- a subagent reading only the code tests each doc's claims; restatement dies, survivors become ADRs, glossary, or roads, savings feed code legibility
- [handoff](https://raw.githubusercontent.com/lroolle/skills/main/skills/handoff/SKILL.md): pass the baton between sessions -- compress live working state into HANDOFF.md when leaving; verify against the repo, then resume in motion when arriving

## Doctrine

- [CLAUDE.md](https://raw.githubusercontent.com/lroolle/skills/main/CLAUDE.md): the craft rules every skill here is built by -- gate-first shape, one-trigger-per-branch descriptions, leading words, routes vs credit
- [template/SKILL.md](https://raw.githubusercontent.com/lroolle/skills/main/template/SKILL.md): the starter skill matching the doctrine