Check design system
Audit a live product against its design system on Stunt Double: typography, colour usage, spacing, component and iconography rules, checked on rendered pages with evidence, and re-checked automatically on library publishes.
When to use
- You have a design system (tokens, component rules, a Figma library) and want to know where the live product drifts from it
- Screens shipped independently and visual consistency has eroded
- You want design-system adherence enforced when the library is published or files change
Instructions
-
Resolve context:
list_workspaces()→ pick the workspacelist_projects(workspace_id)→ match the URL to a project, or letcreate_checklistfind/create it
-
Set up a design-guardian actor and record the rules:
list_actors(workspace_id), elsecreate_actor(...)described as a design-system reviewerlist_workspace_guidelines(workspace_id)first, then either attach an existing rule withset_project_guideline(project_id, guideline_id)or write one withadd_project_guideline(project_id, category: "design_system", content, title). Token values and component dos and don'ts belong here: guidelines reach every run and design review for the projectupdate_workspace_guideline(workspace_id, guideline_id, apply_to_design_reviews: true)when the standard should also hold for design reviews raised from Slack or Linear, which carry no project- If no rules were supplied, ask for the key tokens and component rules, or derive a draft from the most polished screens and confirm it
-
Create a checklist per surface:
- Actors see the rendered product the way a user does, so phrase checks visually, not in terms of code or class names:
- "Body text uses the product font stack, no serif fallbacks anywhere"
- "Primary CTAs use the brand primary colour and sentence case"
- "Form fields show a visible focus state"
- "Empty states use the illustration style, not stock icons"
- "Spacing between sections is consistent across pages"
- Actors see the rendered product the way a user does, so phrase checks visually, not in terms of code or class names:
-
Run and collect evidence:
run_checklist(checklist_id)→ pollget_checklist_run(run_id)until terminal; collect evidence for each deviation
-
Report deviations by page:
- With screenshots or extracted evidence and the rule each one breaks
-
Wire it into the design workflow:
- If the workspace has Figma connected,
create_workflowwith afigma_eventtrigger so the audit re-runs when the design library is published or files change - Otherwise offer a schedule or deploy trigger (see
setup-guardrails)
- If the workspace has Figma connected,
Example output
Design-system audit: app.acme.com
Dashboard:
[FAIL] Card titles render in Georgia (serif): should use Inter. Font stack is missing on /dashboard.
[PASS] Primary buttons use brand primary #4F46E5
[FAIL] Section spacing 12px here vs 24px on Settings: inconsistent vertical rhythm
Settings:
[PASS] Focus states visible on all inputs
[FAIL] Empty "no integrations" state uses a stock plug icon, not the house illustration set
4 deviations. If Figma is connected, offer a figma_event workflow so this re-runs on library publish.Tips
- Phrase checks visually. The actor cannot see class names or tokens: only the rendered result. "Body text is the product sans-serif" works; "uses
--font-body" does not. - Store rules as actor knowledge so every run enforces the same system.
- A
figma_eventtrigger closes the loop: the audit re-runs the moment the library changes.
Check continuity
Check continuity across surfaces on Stunt Double: pricing, terminology, feature names, and promises should match between marketing site, product, docs, and emails. Actors walk each surface and the mismatches surface with evidence.
Create actor panel
Create and configure a Stunt Double actor (AI persona) with knowledge entries for realistic user simulation.