360 Secure SkillHub
360 安全技能中心

sync-figma-token

该 Skill 用于代码设计令牌与 Figma 变量的对齐:规范化令牌行,按策略检测缺失、值、别名、类型、模式、作用域等漂移,输出 dry-run 摘要与明细,要求用户明确批准后再按顺序写入,并持久化报告。默认归档而非删除,冲突按指定策略处理。

firebenders
热度 0.00
最近更新 2026-10-18 23:35:32
分类:开发依赖:MCP 服务 / 读写文件应用场景:工作流 / 数据处理 / 报告撰写

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add firebenders/sync-figma-token-skill/sync-figma-token
Metadata
name
sync-figma-token
description
Sync design tokens between code and Figma variables with strict drift reporting, mandatory approval gate, safe delta apply, and persisted reports.
disable-model-invocation
true
metadata
{"mcp-server":"figma, figma-staging"}

sync-figma-token

Use this skill for token parity workflows (code tokens vs Figma variables).

MANDATORY prerequisite: load figma-use before every use_figma call.

Non-negotiable safety rule

After producing dry-run output, you MUST STOP and ask for approval.

  • Do NOT run any write use_figma calls in the same turn as dry-run output.
  • Ask a normal confirmation question (example: "Apply these changes? (yes/no)").
  • Only proceed on explicit affirmative approval.
  • If the response is unclear or negative, do not apply writes.

Standard source formats (required)

Prefer real token sources in this order:

  1. Design Tokens JSON (tokens.json, tokens/*.json, DTCG-style)
  2. Style Dictionary input JSON
  3. Platform theme sources (Compose/Kotlin/TS) only when JSON source is unavailable

If source format is non-standard, explicitly state assumptions in dry-run output.

Required policies before writes

  • direction: code_to_figma (default), figma_to_code, bidirectional
  • deletePolicy: default archive_only (NOT delete)
  • conflictPolicy: prefer_code, prefer_figma, manual_review
  • namingPolicy: token key normalization strategy
  • modePolicy: code mode <-> Figma mode mapping

Never delete by default. Deletion requires explicit user instruction.

Normalization rules

Normalize both sides to canonical rows:

  • key (canonical token name)
  • type (COLOR, FLOAT, STRING, BOOLEAN)
  • modeValues (light/dark/etc.)
  • aliasTarget
  • scopes
  • codeSyntax

Name normalization examples:

  • color.bg.primary <-> color/bg/primary
  • Neutral10 <-> Neutral/10 only if explicitly mapped by naming policy

Value validation (required)

Dry-run must validate values, not only presence/type.

  • COLOR: compare RGBA with tolerance epsilon = 0.0001
  • FLOAT: strict numeric comparison unless tolerance is configured
  • STRING/BOOLEAN: strict equality
  • Aliases: compare canonical alias targets

Drift categories

Each drift item must include one of:

  • missing_in_figma
  • missing_in_code
  • value_mismatch
  • alias_mismatch
  • type_mismatch
  • mode_mismatch
  • scope_mismatch
  • code_syntax_mismatch
  • broken_alias

Dry-run output format

Always return:

  1. Headline summary:
{
  "create": 0,
  "update": 0,
  "aliasFix": 0,
  "scopeFix": 0,
  "syntaxFix": 0,
  "archive": 0,
  "delete": 0
}
  1. Detailed drift list with token keys and before/after values.

Then ask:

Dry-run complete. Apply these changes? (yes/no)

Report persistence (required)

Persist report JSON every run:

  • /tmp/sync-figma-token-dry-run-{runId}.json
  • /tmp/sync-figma-token-final-{runId}.json

If file persistence fails, mention that explicitly in output.

Conflict handling

When conflicting data is found (type/mode/alias ambiguity):

  • If conflictPolicy=manual_review, list conflicts and STOP.
  • If conflictPolicy=prefer_code, update Figma to source values/types.
  • If conflictPolicy=prefer_figma, keep Figma and emit drift as informational.

Apply order

Apply deltas in this order:

  1. Ensure collections/modes
  2. Create missing primitives
  3. Create/update semantic aliases
  4. Apply value updates
  5. Apply scopes and code syntax
  6. Archive stale tokens per deletePolicy

Never parallelize write use_figma calls.

Success condition

After apply, run a fresh diff. Success = unresolved drift is zero, or only explicitly approved exceptions remain.