A Claude Code skill that gate-reviews Figma implementation plans before canvas work begins, then learns from outcomes after.
- Phase A (Pre-Build): Audits your design system, checks component selection, variable bindings, structure, and project-specific lessons before you write a single
use_figmacall - Phase B (Post-Build): Captures what went wrong, writes project lessons to CLAUDE.md, and evolves cross-project rules in the skill itself
The #1 cause of Figma MCP rework is planning from assumption instead of audited inventory. This skill forces an inspection pass that catches:
- Building from primitives when a component exists
- Hardcoding colors/spacing when variables exist
- Wrong layout structure (3 columns vs 4 columns = full rebuild)
- Oversized DS components used where custom builds are needed
- Duplicated components (composite components that already contain sub-components)
Copy to your Claude Code skills directory:
cp -r figma-plan-review ~/.claude/skills/Or as a user-level skill in any Claude Code project.
The skill triggers on:
"review the plan","check this approach","plan review"- Before any multi-step
figma-use/figma-generate-designwork - Post-build:
"how did that go","what did we learn"
figma-plan-review/
SKILL.md # Main skill definition
references/
review-checklist.md # Detailed sub-checks for Phase A
README.md # This file
The skill accumulates cross-project rules in the ## Learned Review Rules section of SKILL.md. These are written by the skill itself during Phase B, gated by a strict filter: each rule must prevent a real rebuild, not be redundant, fit in one sentence, and not be something a competent agent would figure out anyway.