菜单
Staj's research manual — the RFC template with exit criteria, the competitive landscape map (styling engines and UI kits), the CSS platform watchlist, citation discipline, and the teardown format. Consult before starting any research task, RFC, or competitive analysis.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Step 3 of the create-recipe chain. Writes the recipe TypeScript file(s) at theme/src/recipes/<name>/, the per-recipe barrel index, the root barrel update, and Vitest tests. Uses the approved design.md to decide color/variant-style patterns, compound-variant strategy (flatMap or map-per-style), and whether the recipe needs a setup callback (e.g., for @keyframes). Tests are required by default. Writes .context/recipe-<name>/implementation.md.
Step 6 (final) of the create-recipe chain. Runs pnpm typecheck, pnpm --filter @styleframe/theme test, and pnpm lint to verify the recipe is correct. Reports failures with file:line references and suggested fix locations. Does NOT attempt auto-fix. Standalone-capable — runs against any existing recipe without prior artifacts. Writes .context/recipe-<name>/verification.md.
Maître's front-of-house manual — the positioning canon (single source of truth for outbound claims), developer-marketing rules, channel playbooks, the launch checklist, and templates for announcements and copy briefs. Consult before writing any outbound piece, planning a launch, or briefing another seat.
Open a Styleframe pull request in the house style — conventional-commit title, a What/Why/How-verified body, a linked issue, and a changeset when a publishable package changed. Enforces the hard rule that PR descriptions never @mention agents (PRs are public; agent handles must not leak), and routes the review-request mention to the Multica issue comment instead. Invoke with /create-pr whenever you are about to open a PR.
Institutional knowledge for the Styleframe documentation site (apps/docs, Nuxt Content) — content structure and numbered categories, link/route resolution quirks, component-doc anatomy and reference pages, runnable-example rules, voice, framework switcher, and the no-changeset rule for docs. Consult before every apps/docs task.
Institutional knowledge for building Styleframe theme work — the /create-recipe chain, recipe pitfalls unique to this codebase (modifier-registry selectors, utility decomposition, variant-axis ordering, eager ref resolution), shared builders, naming conventions, colocated tests, and Storybook conventions. Consult before EVERY theme/** or storybook task.
| name | rfc-method |
| description | Staj's research manual — the RFC template with exit criteria, the competitive landscape map (styling engines and UI kits), the CSS platform watchlist, citation discipline, and the teardown format. Consult before starting any research task, RFC, or competitive analysis. |
Research that does not end in a decision is trivia. Every output of this seat is written to be decided: one decision per document, a recommendation the reader can accept or reject in ten minutes, and an exit criterion that makes the recommendation falsifiable.
# RFC: <decision to be made, phrased as a question>
Status: draft | decided(accepted/rejected) · Date: <YYYY-MM-DD> · Author: Staj
## Problem
What hurts or what opportunity closes if we do nothing. Grounded in THIS repo:
name the modules/recipes/docs affected.
## Prior art
How Tailwind vN / Panda vN / StyleX vN / the platform solve it — each claim
cited with version, date, and link. Primary sources only.
## Proposed design
Sketch level: API shape, which packages change (engine/core? transpiler
pipelines? theme?), migration story. Enough to estimate, not to implement.
## Alternatives
2–3, each with named tradeoffs. "Do nothing" is always listed, with its cost.
## Adoption & migration
Breaking? Codemod-able? What does the changeset/release train look like?
## Exit criterion
The observable outcome that would prove this decision wrong, and when to check.
## Recommendation
One paragraph. Take a side.
File as a Multica issue, @-mention @mise — the only @ in the document. An RFC names seats constantly (feasibility, landing zones, spikes); every one of those is a plain name (Roux, Mère), because an @handle in a posted issue wakes that agent to start working mid-read. Depth goes in appendices; the body stays a ten-minute read. Two decisions = two RFCs.
testing/benchmark via @etoile, or spec a spike issue for the owning seat stating exactly what question the spike must answer. You do not write code — not even spikes.Styling engines: Tailwind v4 (oxide engine, CSS-first @theme config — the mindshare rival), Panda CSS (nearest architectural neighbor: static extraction + recipes), StyleX (Meta; compile-time atomic), vanilla-extract (zero-runtime TS pioneer), CVA / tailwind-variants (the recipe-pattern incumbents), Open Props (token distribution).
UI kits (recipe-parity yardstick): shadcn/ui, Radix, Nuxt UI, Chakra, Mantine — track their component lists against our 39 recipes and their theming/token stories against @styleframe/theme.
Interchange: DTCG spec evolution, Style Dictionary, Tokens Studio — tooling/dtcg must stay conformant.
Styleframe's differentiators to defend: typed dual output (CSS + TS + DTS from one AST), design-token composables as a library (not config), 9-bundler unplugin reach, free bidirectional Figma sync, recipes with a ~1.4KB runtime.
Track Baseline status; when a feature crosses "newly available," evaluate whether it deserves first-class Styleframe support (modifier, utility, token type, or transpiler output) versus staying an escape hatch: container queries (+units), @property, anchor positioning, view transitions, :has(), @scope, scroll-driven animations, light-dark(), color-mix(), nesting, @starting-style, popover/dialog primitives.
Map each candidate to its landing zone: modifiers → theme/src/modifiers/, utilities → theme/src/utilities/, token types → engine/core + all three transpiler pipelines (see engine-craft's end-to-end checklist — feasibility questions go to @roux).
Scope (versions, date) → feature/DX matrix vs Styleframe → what they do better (honest, specific) → what we do better (evidence, not pride) → top-5 gaps ranked by adoption impact → handed to @mise as issue candidates. A teardown that finds nothing to steal was not adversarial enough — rivals' good ideas are gifts.