reconcile-paper-sync
millionco/skills · Agent Skill
檢查和修復Paper-to-code的實現與其標準渲染paper-sync的差異,包括字型、精確的token和截圖比較。
前往安裝 ↓總覽
| 發布者 | |
|---|---|
| 儲存庫 | millionco/skills ↗ |
| 技能資料夾 | reconcile-paper-sync |
| 星 | 34 |
| 分支 | 2 |
| 附帶檔案 | 1 |
| 授權 | MIT |
| 最近更新 | 2026年6月30日 |
| 連結 | 技能原始碼 ↗ · 儲存庫 ↗ |
Mozgi 檢查
我們在技能檔案中發現的內容。詳細資訊需訂閱。
安裝
技能就是一個含 SKILL.md 的資料夾。選擇你的代理,在專案根目錄的終端機(macOS、Linux 或 Windows 的 Git Bash)執行指令。
把提示貼到 Claude、Codex 或其他助理:它會自行檢查並安裝技能。
Install the “reconcile-paper-sync” skill from mozgi.io (https://mozgi.io/skills/millionco-reconcile-paper-sync/) for me. 1. Download the archive: https://mozgi.io/skills/millionco-reconcile-paper-sync/download If it fails, take the skill folder from the source: https://github.com/millionco/skills/tree/main/skills/reconcile-paper-sync 2. Read SKILL.md and the other files. Check the skill is safe: no hidden commands, no sending data elsewhere, no scripts doing something other than described. If anything looks suspicious, stop and tell me. 3. Put the “reconcile-paper-sync” folder into your agent's skills folder: .claude/skills/ for Claude Code, .agents/skills/ for Codex, .cursor/skills/ for Cursor, .gemini/skills/ for Gemini CLI. Inside the project if we are in one, otherwise in the home folder (~/.claude/skills/ etc.). 4. Make sure <skills folder>/reconcile-paper-sync/SKILL.md exists. 5. Briefly tell me what this skill does and offer to try it on my task. If you cannot do a step yourself, tell me what to click or type.
git clone --depth 1 --branch main https://github.com/millionco/skills.git /tmp/skills mkdir -p .claude/skills cp -r /tmp/skills/skills/reconcile-paper-sync .claude/skills/reconcile-paper-sync
重新啟動 Claude Code — 技能就會出現在清單中,手動呼叫:/reconcile-paper-sync。要套用到所有專案,請放在 ~/.claude/skills。
git clone --depth 1 --branch main https://github.com/millionco/skills.git /tmp/skills mkdir -p .agents/skills cp -r /tmp/skills/skills/reconcile-paper-sync .agents/skills/reconcile-paper-sync
Codex 會從目前資料夾往上到儲存庫根目錄,在 .agents/skills 中尋找技能;用 /skills 或 $reconcile-paper-sync 呼叫。所有專案:~/.agents/skills。
git clone --depth 1 --branch main https://github.com/millionco/skills.git /tmp/skills mkdir -p .cursor/skills cp -r /tmp/skills/skills/reconcile-paper-sync .cursor/skills/reconcile-paper-sync
Cursor 啟動時會載入技能;在 Agent 聊天輸入「/」呼叫。Cursor 也會讀取 .claude/skills 與 .agents/skills。所有專案:~/.cursor/skills。
git clone --depth 1 --branch main https://github.com/millionco/skills.git /tmp/skills mkdir -p .gemini/skills cp -r /tmp/skills/skills/reconcile-paper-sync .gemini/skills/reconcile-paper-sync
任務符合時,Gemini CLI 會提議啟用技能,由你確認。所有專案:~/.gemini/skills。
安裝前請先檢視 SKILL.md 與相關檔案。點擊檔案即可預覽。
SKILL.md 內容
作者原文 · 唯讀Reconcile a Built App with paper-sync
You make a running app match the Paper design it was built from, using the project's paper-sync output as the canonical reference render.
This is the verification-and-repair phase that follows paper-to-code-components. The conversion produces maintainable code; this skill confirms the result still looks like the design and closes the gaps where it drifted.
When this applies
- A Paper-derived app (e.g. a Next.js site) renders close to the design but "slightly off."
- A
paper-syncproject exists (a.paper-sync/<fileId>/directory and usually a dev server such aslocalhost:6006). - You want to verify a fresh Paper-to-code implementation before calling it done.
If there is no paper-sync project, fall back to the Paper screenshot as the oracle and apply the same method.
RED/GREEN Intent
This skill prevents these reconciliation failures:
- Declaring parity from reading code instead of comparing rendered pixels
- Leaving the app on
system-ui/Geist/Georgia fallbacks while the design uses real licensed fonts - Collapsing the design's several typefaces into one font token
- "Fixing" a difference that is actually newer design work in paper-sync by inventing data to match it
- Dropping small decorations (code-chip backgrounds, hairlines, dashed outlines, badges) during conversion and never noticing
GREEN outcome: the running app, screenshotted at the frame's width, matches the paper-sync render on typeface, type scale, color, spacing, and per-element decoration — and any remaining differences are explicitly attributed to newer design iterations, not conversion bugs.
1. Find the paper-sync canonical
paper-sync mirrors the Paper file into the repo. For each Paper file it keeps:
1. .paper-sync/<fileId>/.canonical/*.tsx — the verbatim Paper export per frame, the source of truth for exact values. Frame files are named page-1--frame-<id>-0.tsx and carry a header comment with the Paper URL (.../<FILE>/1-0/<FRAME>).
2. .paper-sync/<fileId>/paper.css — @font-face declarations and the design's CSS.
3. .paper-sync/<fileId>/public/paper-assets/fonts/*.otf — the real font files.
4. A dev server with a preview route: ?preview#<FRAME> renders a single frame alone (the frame id is the uppercase hash, e.g. N66-0).
Map the frame you are reconciling to its canonical file by matching visible text (grep -l "We introduce React Bench" .canonical/*.tsx). The on-disk .canonical can lag the running server; trust the live ?preview#<frame> render for what the user sees, and the canonical .tsx for exact numeric values.
2. Fonts first — the biggest lever
Substituted fonts are the most common cause of "slightly off." Do this before chasing spacing.
1. List the families the canonical frame uses: grep -ohE "font-\['[^']+'" <frame>.tsx | sort | uniq -c | sort -rn. Expect several distinct families by role (body, serif display, UI/label, numeric/mono, code).
2. For each family, confirm the app self-hosts the real font via @font-face and maps it to its own token. If it instead falls back to Geist/Georgia/system, that is the bug.
3. Port any missing fonts from public/paper-assets/fonts/*.otf into the app (e.g. public/fonts/), add @font-face per weight/style, and point each token at the real family with the system stack only as fallback. Keep distinct typefaces as distinct tokens — do not collapse them.
See paper-to-code-components §8 for the porting rules.
3. Extract exact values from the canonical
The .canonical/*.tsx holds the real numbers. For any element that looks off, read its canonical class list and match:
1. Type: text-[15px]/6 is 15px / line-height 24px; replicate size AND line-height, not just size.
2. Color: color(display-p3 ...) and hex — keep the exact gamut/value.
3. Decoration: backgrounds, borders, dashed outlines, underlines, radii, small markers.
Do not read sizes or colors off a screenshot when the canonical source is available.
4. Screenshot-diff the running app against the frame
1. Render both at the same width. A headless browser screenshot works without installing Playwright: drive a cached Chromium, e.g. chrome-headless-shell --hide-scrollbars --force-device-scale-factor=2 --virtual-time-budget=6000 --window-size=<w>,<h> --screenshot=out.png <url>. Allow time for @font-face to load before the shot.
2. Capture the app (localhost:3000) and the frame (localhost:6006/?preview#<FRAME>), then compare side by side.
3. Reconcile element stacking order (e.g. nav above vs below a header band), content width and text measure (same paragraph → same line count?), typeface, type scale, color, spacing, and every decoration. A flow/responsive layout will not be pixel-identical to absolute Paper coordinates — judge structural, typographic, and decorative fidelity, not coordinate equality.
5. Separate conversion drift from version drift
A difference has two possible causes; treat them differently:
- Conversion drift — the app dropped or substituted something the source frame has (a font, a code-chip background, a hairline). Fix it to match.
- Version drift — the paper-sync frame is a newer design than the export the app was built from (the frame header date / frame id differs; new elements like error bars or bullets appear that need data the app does not have). Do NOT invent data to fake it. Flag it to the user and let them decide whether to re-sync the app to the newer frame.
State which differences you fixed and which you attributed to a newer frame.
Escape Hatches Closed
| You Think | Do This Instead |
|---|---|
| "The code matches the design, so it's fine." | Screenshot the build against the paper-sync frame and compare pixels. |
| "The font has a system-ui fallback, that's close enough." | The fallback is not the design. Self-host the real .otf from public/paper-assets/fonts and @font-face it. |
| "I'll match the new error bars / bullets in the frame." | Check the frame date/id first. If it's newer than the built export, that's version drift — flag it, don't invent data. |
| "I'll eyeball the sizes from the screenshot." | Read exact text-[..]/.. and color(display-p3 ..) values from the .canonical/*.tsx. |
| "No Playwright installed, so I can't screenshot." | Drive a cached chrome-headless-shell with --screenshot; no install needed. |
Before Marking Complete, You MUST:
1. Identify the canonical frame (file id + frame hash) the app section corresponds to
2. Confirm every typeface the frame uses is self-hosted via @font-face and mapped to its own token — no design text left on a fallback font
3. Match exact type size + line-height, colors, and decorations from the .canonical/*.tsx, not from a screenshot
4. Screenshot the running app against ?preview#<frame> at matched width and reconcile the differences
5. Classify each remaining difference as conversion drift (fixed) or version drift (flagged to the user)
6. Re-run the app's lint/build after changes and report the screenshot comparison
常見問題
reconcile-paper-sync 技能有什麼用途?
檢查和修復Paper-to-code的實現與其標準渲染paper-sync的差異,包括字型、精確的token和截圖比較。
如何安裝 reconcile-paper-sync?
打開本頁的「安裝」區塊並選擇你的代理 — Claude Code、Codex、Cursor 或 Gemini CLI:那裡有把技能資料夾複製到正確位置的現成指令。也可以複製「透過助理」分頁中的提示,讓助理自行檢查並安裝技能。
reconcile-paper-sync 適用哪些 AI 模型?
技能是純 Markdown 指示(Agent Skills 格式),因此 reconcile-paper-sync 適用任何讀取 SKILL.md 的代理:Claude Code、Codex、Cursor、Gemini CLI、GitHub Copilot、OpenCode 等。只有在任務符合技能描述時,模型才會載入指示。
reconcile-paper-sync 是免費的嗎?
此技能由 millionco 公開發布,採用 MIT 授權。技能本身免費 — 你只需向 AI 供應商支付實際使用的 token 費用。
安裝技能安全嗎?
技能是代理會遵循的指示,安裝前請閱讀 SKILL.md 及相關檔案。技能資料夾中的腳本只有在代理執行時才會運行 — 留意它的動作,不要授予多餘權限。
技能和 MCP 伺服器或外掛有何不同?
MCP 伺服器或外掛為模型提供新工具 — 會執行並回傳結果的程式碼。技能提供的是知識與流程:如何處理任務、要遵循哪些步驟、好的成果長什麼樣子。它不需要伺服器或 API 金鑰。