design-system-drift-detector — Claude Code skill
The design-system-drift-detector skill for Claude Code identifies discrepancies between implementation and design tokens. It detects ungoverned literals, semantic conflicts, and off-scale values.
MIT ★ 529 updated 2026-11-17
npx skills add ShuaiMXu/happyhands --skill design-system-drift-detector -a claude-codeWhat design-system-drift-detector does
This skill scans the repository to compare implementation values against a defined design foundation. It identifies three types of divergence: ungoverned values, semantic conflicts, and off-scale values.
The tool requires a design token foundation and access to repository sources to function. Because static analysis cannot resolve computed styles, utility classes, or runtime themes, findings are marked as evidence for human review. If rules are not clearly defined, the tool returns a needs-evidence status.
What it helps with
- find code values missing tokens
- identify semantic slot conflicts
- calculate distance to nearest token steps
- compare implementation against design foundation
Example prompts for the agent
- Check for component drifts using design-system-drift-detector
- Analyze code against design tokens
- Find off-scale values in
src/styles/
How to install design-system-drift-detector
1Skills CLI (Vercel)
npx skills add ShuaiMXu/happyhands --skill design-system-drift-detector -a claude-code2Manually: put the files in place
~/.claude/skills/design-system-drift-detector/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/design-system-drift-detector/
1Skills CLI (Vercel)
npx skills add ShuaiMXu/happyhands --skill design-system-drift-detector -a codex2Manually: put the files in place
.agents/skills/design-system-drift-detector/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add ShuaiMXu/happyhands --skill design-system-drift-detector -a cursor2Manually: put the files in place
.cursor/skills/design-system-drift-detector/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install ShuaiMXu/happyhands skills/product-design/design-system-drift-detector2Manually: put the files in place
.github/skills/design-system-drift-detector/SKILL.md1Skills CLI (Vercel)
npx skills add ShuaiMXu/happyhands --skill design-system-drift-detector -a gemini-cli2Platform CLI
gemini skills install https://github.com/ShuaiMXu/happyhands --path skills/product-design/design-system-drift-detectorManually: put the files in place
.devin/skills/design-system-drift-detector/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about design-system-drift-detector
How do I install design-system-drift-detector in Claude Code?
Run the command npx skills add ShuaiMXu/happyhands --skill design-system-drift-detector -a claude-code.
What does this skill do?
It detects implementation divergences from your design system, identifying semantic conflicts and off-scale values.
What is required to run design-system-drift-detector?
You need a repository with a design token foundation and project source files.
Source text
SKILL.md · MIT
--- name: design-system-drift-detector description: "Detect implementation values that diverge from the design foundation as ungoverned values, semantic conflicts or off-scale steps." --- # Design System Drift Detector Find where the implementation and the design foundation no longer agree, and say what kind of disagreement it is. ## Required evidence - a design foundation with tokens - current repository sources - changed files when available ## Three mechanisms Drift is not one thing. Each mechanism needs its own check, because a check built for one is blind to the others. **Ungoverned value.** A literal that no token represents. Ranked by how many times and how many files it reached, because a value that spread across dozens of files is a token candidate, while one that appears twice is probably a local mistake. Report the count and the distribution; do not treat frequency as approval. **Semantic conflict.** One semantic slot — danger text, success surface, warning border — carrying more than one value. This is the mechanism a frequency check cannot see: a wrong danger colour used once in a single notice never clears any threshold, and it is exactly what an audit has to surface. Group by role and by property slot, list every value with where it came from, and do not pick a winner. **Off-scale value.** A value that sits between the steps of a token scale: radius 8 against a 7/9/11 scale, a half-step font size, a 300ms transition against a 130/170/300/600 band. Report the nearest step and the distance. For motion, also report whether snapping to the nearest step would visibly change how the interface moves — snapping a duration is not a rename, it changes behaviour. ## Decisions - Report exact divergence with its mechanism, count and sources. - Report the direction as unresolved. A value that already shipped is often the truth and the token file is what has to change; the opposite is just as common. A person decides which side moves and records why. - Separate a deprecated usage from a legitimate extension by asking for review, never by replacing a value automatically. - Distinguish registering a value from applying it. Adding a token to the foundation changes the foundation only; the implementation still holds a literal until someone changes it to reference the token. - Report how much of the implementation already reads tokens, so the next alignment pass has a number to move. ## Boundary Approved rules must exist before drift can be claimed. With a foundation whose entries are only observed, the same divergences are review targets and the result stays `needs-evidence`. A static scan cannot follow computed styles, utility classes, runtime theming or a `var(--token)` reference to the value it resolves to. Semantic slots are inferred from nearby keywords, so an unusual naming convention can hide a real conflict. Return structured evidence, findings, proposed updates and limitations. Use `needs-evidence` when required inputs or professional review are missing. Never invent an Experience Score or promote a project observation into a universal rule.
© ShuaiMXu · MIT · ShuaiMXu/happyhands