figma-workspace-architect
A Claude Code skill for designing, building, auditing, and migrating Figma workspaces via
the figma-console-mcp server. Sets the bar at "a workspace where a designer
can find what they need without asking" — correct Variable scopes, Variants matching the
engineering prop API, Modes carrying theming instead of duplicate components, predictable
naming.
What it does
The skill activates when a user asks about Figma file architecture, design tokens, components, or a workspace decision. It picks one of four modes based on the user's intent, then loads only the references it needs — references are scoped, not bundled. Sub-modes handle the narrower cases that benefit from a more focused recipe.
Build
triggers on: create, set up, bootstrap, scaffold, build, generate, add, refactorDiscovery → Decide → Map tools → Execute → Validate → Document. Never write to Figma without first reading what is there.
One-shot starter: Cover + Tokens + Components pages with placeholder content. For when the user wants a working skeleton in one turn.
Audit
triggers on: audit, review, check, assess, what's wrong with, how good isTwo layers: Design System Dashboard (breadth: A11y, Coverage) followed by the architectural deep-audit (depth: 5 categories with severity tags). Pre-flight pins the snapshot (git SHA + Figma lastModified) and runs cross-source greps so findings are grounded in current state.
Re-checks an existing audit report by running only the per-category verify queries. Emits still-open / auto-resolved / state-shifted per finding so stale findings drop off the priority list automatically.
Decide
triggers on: either/or questions, "should I…?", "is it better to…?"Single-pass recommendation with reasoning in 1–2 sentences plus the alternative path with its trade-off. No essays.
Migrate
triggers on: rename, split, restructure, refactor, deprecate, retire, replaceThe additive coordination protocol for breaking changes (add new alongside old → wait one cycle → migrate consumers → remove old). Recipes for variable renames, mode adds, variant set splits, primitive→semantic promotions, library splits.
Post-flight Storybook + browser-automation recipe to confirm Figma↔code parity in Light + Dark after a token or variant change. Catches the bug class where Figma and code each look correct but their handoff is broken.
Install
Claude Code skills live as directories under ~/.claude/skills/ (user-global)
or .claude/skills/ (project-local). Both options below land the skill in the
canonical layout — Claude Code auto-discovers it on the next session.
From the published zip
# Download the runtime payload (SKILL.md + references/ + assets/)curl -sL "https://atelier.pieper.io/.well-known/agent-skills/figma-workspace-architect.zip" \ -o /tmp/figma-workspace-architect.zipunzip -o /tmp/figma-workspace-architect.zip -d ~/.claude/skills/From a repo clone
# Or install from a clone of the repogit clone https://github.com/DominikPieper/atelier-ui.gitcp -r atelier-ui/skills/figma-workspace-architect ~/.claude/skills/The skill needs figma-console-mcp configured to do anything useful — it is the underlying MCP server the skill drives. See the Figma Setup page for the connection guide.
When the skill loads
The skill's description field is tuned so Claude Code picks it up on architectural
/ audit / build prompts about Figma — even when the user only says "Figma" without naming
the file or library. It explicitly does not trigger on Figma → code translation
requests, which are out of scope.
# Triggering examples — any of these will load the skill
"audit our Figma library""set up token tiers in Figma""should I use a Variant or a Component Property here?""rename color/primary to color/brand-primary safely""check Figma against the spec"References
Loaded on demand by the active mode. Every reference is self-contained — opening one does not pull the others into context.
| File | Summary |
|---|---|
| token-architecture.md | Three-tier model (Primitive / Semantic / Component), Modes, Variable Scopes, anti-patterns. |
| component-design.md | Variants vs. Component Properties vs. Instance Swap, atomic composition, slot patterns. |
| naming-and-file-structure.md | Slash naming, page taxonomy, _ and . prefixes for sub-components, library boundaries. |
| tool-map.md | figma-console-mcp tools grouped by purpose with ready-to-use figma_execute payloads. |
| build-workflow.md | Build mode in depth: discovery checklist, decide gates, validation, documentation requirements. |
| audit-checklist.md | Five architectural categories with severity tags. Pre-flight: snapshot pin + cross-source grep. |
| audit-verify-queries.md | Re-verify sub-mode. One verify query per audit category with auto-resolve signals. |
| decision-heuristics.md | Decision trees for the recurring forks (Variant vs. Property, Token tier, Mode vs. Collection). |
| migration-playbook.md | Refactoring an existing file safely. Coordination protocol, recipes, removal-tombstone gotcha. |
| code-verify.md | Code-side visual verification recipe. Storybook flow for Figma↔code parity in Light + Dark. |
| code-sync.md | Keeping Figma Variables in lockstep with code-side tokens. Drift sources and mode-mapping pitfalls. |
| iconography.md | Icon-system architecture: size stops, color binding, library-split signals. |
| scaffold-payload.md | Scaffold sub-mode of Build. Three-call recipe producing Cover + Tokens + Components pages. |
Test scenarios
Each scenario has an input.md (the user prompt) and an expected.md
(the mode + references + first tool call + regressions to flag). Scenarios are well-formedness
tested in CI; behavioural grading is manual or LLM-judged.
| Scenario | Mode | Summary |
|---|---|---|
audit-existing-system | Audit | Two-year-old library, prioritised audit report. |
build-bootstrap-tokens | Build | Bootstrap a fresh token system end-to-end. |
build-scaffold-starter | Build (Scaffold) | One-shot starter file with three-call recipe. |
decide-variant-vs-property | Decide | Variant vs. Component Property fork. |
migrate-rename-variable | Migrate | Variable rename with code-side coordination. |
re-verify-stale-audit | Audit (Re-verify) | Re-check a 3-day-old audit before acting. |
code-verify-token-rename | Migrate (post-flight) | Browser confirmation after token consolidation. |
out-of-scope-code-gen | Out-of-scope | Figma → code translation request — agent declines and points elsewhere. |
Source
- SKILL.md
- https://atelier.pieper.io/.well-known/agent-skills/figma-workspace-architect/SKILL.md
- Repository
- https://github.com/DominikPieper/atelier-ui/tree/main/skills/figma-workspace-architect
- Discovery index
- /.well-known/agent-skills/index.json
- License
- Same as the repository (MIT). The skill is
private: trueon npm — distribution is via the discovery URL only.