Token导航 LogoToken导航

Agent Skill · 研究检索

fix-design-system-finding(修复设计系统发现)

用于辅助界面设计、视觉规范、排版、配色、布局和交互体验优化。它适合让 Agent 根据产品场景整理页面结构、生成 UI 方案、检查视觉一致性或改进组件层级。使用时需要结合现有品牌、设计系统和用户任务,不应只堆装饰元素;涉及真实页面改动时,应通过截图或浏览器预览检查文本溢出、对齐和响应式表现。

fix-design-system-finding 是研究检索类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为调用外部服务;许可证信息还不明确,第三方扫描已通过,按安装量在研究检索类的 73,675 个 Skill 里排第 46,467。

总安装 776周安装 33GitHub Stars 31下载 272核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
unknown

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:fix-design-system-finding(修复设计系统发现)
来源仓库:https://github.com/edenspiekermann/skills
仓库路径:skills/fix-design-system-finding
安装命令:
npx skills add https://github.com/edenspiekermann/skills --skill fix-design-system-finding
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/edenspiekermann/skills --skill fix-design-system-finding

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

fix-design-system-finding 用于辅助界面设计和视觉规范优化。适合整理页面结构、生成 UI 方案或检查视觉一致性。通过 npx skills add 命令从指定仓库安装,需确认权限与维护状态。使用时需结合现有品牌和设计系统,避免仅堆装饰元素。

建议结合原始 README 进一步核验具体用法和功能边界。

来源仓库内路径:skills/fix-design-system-finding

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/edenspiekermann/skills --skill fix-design-syst…;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要调用外部服务,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

fix-design-system-finding 怎么安装?

在终端执行 npx skills add https://github.com/edenspiekermann/skills --skill fix-design-system-finding。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

fix-design-system-finding 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

fix-design-system-finding 需要什么权限?

权限级别是调用外部服务。该 Skill 可能调用第三方服务、云服务或外部模型 API,使用前需要确认账号、额度、数据发送范围和服务条款。

fix-design-system-finding 是免费的吗?

公开资料没有标注明确许可证,收费和商用限制都无法确认。正式使用前请到来源仓库核对授权说明。

fix-design-system-finding 用的人多吗?

公开数据是累计安装 776 次,近一周安装 33 次,下载 272 次,GitHub 31 星。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

fix-design-system-finding 安全吗?

第三方扫描结果为通过,但这只覆盖已知的静态检查项,不能替代自己读一遍源码和安装命令。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。
  • 第三方扫描已通过,常见的高风险行为没有命中。

先想清楚的地方

  • 许可证没有明确标注,能不能商用、能不能再分发都不确定。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。

不太适合

  • 需要明确授权才能用到公司项目里的场景。
  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
fix-design-system-finding

修复设计系统发现 · 本页

77631调用外部服务未明确通过

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

Fix One Design-System Finding

Use this skill after audit-design-system has already identified a concrete issue.

This skill is intentionally narrow:

  • one finding at a time
  • one offending node or local cluster at a time
  • minimal read scope
  • minimal write scope

Load these capabilities first:

  • Figma MCP read access for the normal inspection tools
  • a figma-use-style helper before any use_figma write call, when your environment requires one

Use apply-design-system instead when the task is a broader screen-wide reconnect, multiple sections need coordinated changes, or the user explicitly wants a full pass across the frame.

Expected Input

Prefer one of these:

  • a single JSON finding from audit-design-system
  • the full review JSON plus the specific finding to address
  • a paraphrased issue plus a Figma URL or fileKey and nodeId

If the finding came from audit-design-system, extract the target from:

  • code_location.absolute_file_path = /figma/<fileKey>/nodes/<nodeId>

Treat the review result as the scope contract. Do not silently widen from one finding into a full cleanup pass.

Scope Rule

Stay pinned to the selected finding until one of these is true:

  • fixed: the targeted issue is corrected
  • blocked: the fix requires missing library assets, inaccessible components, or broader product decisions
  • needs-follow-up: the targeted node is improved, but a separate adjacent issue remains and should be reviewed independently

If the user pastes multiple findings, either address the one they selected or pick the highest-priority finding and say that is the one being fixed now.

Required Workflow

1. Normalize the Finding

Extract and restate:

  • fileKey
  • offending nodeId
  • finding title
  • why the reviewer flagged it

If the review output does not identify a specific node, stop and ask for either the raw finding JSON or a direct Figma node.

2. Run A Local Compatibility Check When Replacement Choice Is Ambiguous

Before backing anything up or editing the canvas, decide whether a clean fix is actually possible.

Run a local compatibility check when:

  • the finding implies a DS swap but does not identify the exact component
  • more than one library component looks plausible
  • override behavior or sizing is unclear
  • the user explicitly wants extra research before writing

Do not continue into writes until the compatibility check returns one of:

  • safe-to-apply
  • blocked
  • needs-human-choice

The compatibility check should answer:

  • which exact component or variant is the leading candidate
  • whether text, icon, state, and size overrides are supported
  • whether the replacement can preserve the node's layout contract

If the result is blocked, stop and report the blocker.

If the result is needs-human-choice, stop and ask the minimal clarifying question instead of guessing.

3. Gather Only the Evidence Needed

Before writing, inspect the target with the smallest useful set of reads:

  1. get_screenshot for the offending node or closest enclosing section
  2. get_metadata for structural context
  3. get_variable_defs when the finding mentions tokens, raw values, color, spacing, or typography
  4. get_code_connect_map when the finding suggests an existing mapped component
  5. search_design_system only if you still need to locate the replacement primitive

Do not re-review the whole board unless the finding clearly points to a repeated pattern and the repeated pattern is required to choose the fix.

4. Classify the Fix

Choose exactly one remediation mode:

  • swap-instance: replace an ad-hoc or wrong instance with the correct library component or variant
  • compose-from-primitives: rebuild the local custom wrapper from published library parts
  • bind-tokens: replace raw values with variables or styles
  • align-variant: move an existing instance to the correct variant or property state
  • blocked: no valid library-backed fix is currently possible

If more than one mode seems plausible, prefer the smallest change that satisfies the original finding.

5. Back Up Only the Affected Area

Before destructive edits, duplicate only the offending node or the smallest enclosing local section.

Name the backup clearly, for example:

  • Backup - Fix finding 1
  • Backup - Custom nav

Make the backup in its own write call and return the created node ID.

6. Apply the Fix Incrementally

Do not rebuild the whole screen.

Work in small steps:

  1. inspect current node IDs and component linkage
  2. import or locate the needed library component or variable
  3. replace, compose, or retokenize only the targeted area
  4. return all created and mutated node IDs

Prefer swapComponent() when it preserves the node's position and valid overrides.

Prefer rebuilding beside the original when:

  • the current node is a local wrapper with mixed custom structure
  • the review finding points at a custom composition, not a single primitive
  • you need visual comparison before replacing the original

If the first validation pass shows that the recommended component regresses layout, text visibility, icon behavior, or width contract, stop. Report blocked or needs-follow-up instead of continuing to force the same primitive into place.

7. Validate the Finding, Not the Entire Screen

After the write:

  1. screenshot the changed node or its immediate section
  2. confirm the specific review complaint is no longer true
  3. confirm the result is library-backed or token-bound when that was the goal
  4. confirm spacing and text overrides still make sense

Only expand validation beyond the local section if the fix touched a global pattern such as navigation.

Decision Heuristics

  • If the finding says a primitive was recreated as a frame, bias toward swap-instance.
  • If the finding says repeated local wrappers should collapse into a shared pattern, fix one representative node unless the task explicitly asks to roll the fix across all siblings.
  • If the finding is about raw values beside tokenized peers, bias toward bind-tokens.
  • If the finding is about a locally edited instance with the wrong size or state, bias toward align-variant.
  • If the library lacks the required component family or the import fails, mark it blocked instead of improvising a fake connection.
  • If a DS component exists but cannot preserve the node's text/icon/sizing contract, mark it blocked or needs-human-choice instead of forcing the swap.

Writing Rules

  • Keep the write scope tightly aligned to the selected finding.
  • No write before a safe-to-apply compatibility outcome when replacement choice is ambiguous.
  • Do not claim the whole screen is "on system" after fixing one finding.
  • Prefer exact component keys and variable IDs over names.
  • Do not continue into unrelated cleanup just because you noticed more drift.
  • If a broader refactor is obviously needed, finish the targeted fix or report the blocker, then recommend apply-design-system for the next pass.

Deliverable Format

When closing the task, report:

  • Finding fixed: what changed and on which node
  • Validation: how you confirmed the original issue is resolved
  • Blocked or Follow-up: only if the finding could not be fully resolved

Keep the report tied to the single finding. Avoid screen-wide summaries unless the user asks for them.

平台分布

Codex

33.24%
按下载量换算90

Claude

30.12%
按下载量换算82

Cursor

18.59%
按下载量换算51

Gemini CLI

9.27%
按下载量换算25

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

调用外部服务

该 Skill 可能调用第三方服务、云服务或外部模型 API,使用前需要确认账号、额度、数据发送范围和服务条款。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills