Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Read a Canva design and return structured, actionable design feedback — visual hierarchy, copy/messaging, layout & spacing, consistency, readability, and accessibility. Read-only; makes no changes to the design. Use when the user asks to \"review my design\", \"give me feedback on this\", \"critique my deck/poster/flyer\", \"how can I improve this design\", or \"what's wrong with this slide\".
.claude/skills/kunanonj-cursor-plugin-canva-get-design-feedback/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 258% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 20% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 56% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 14% | 0% |
| case-19 | ✗→✓ | ▲ Improved | 36% | 0% |
Act as a design reviewer: read the design as it actually appears, then return concrete, prioritised feedback the user can act on. This skill is read-only — it never edits the design. When the user wants the changes made, hand off to canva-edit-design or canva-implement-feedback.
Canva:get-design-content returns text (richtexts) only — good for copy, headings, and wording, but it does NOT include colors, fonts, sizes, or element positions.Canva:get-design-thumbnail gives you the rendered image — this is how you "see" layout, hierarchy, balance, color, and contrast. Always pull this; visual critique depends on it.Canva:start-editing-transaction, inspect the returned richtexts/fills, then Canva:cancel-editing-transaction (never commit — you are not changing anything). Use this for layout/spacing/alignment detail.Short link → Canva:resolve-shortlink; full URL → extract the ID; raw D... ID → use directly; otherwise ask.
Canva:get-design for title and page count.Canva:get-design-thumbnail (and/or Canva:get-design-pages) to see each page.Canva:get-design-content for the text.Assess the design against these lenses. Skip any that don't apply to the design type:
Organise findings by page, each with a severity and a concrete fix:
## Feedback — "<design title>" (N pages)
### Top priorities
1. [High] Page 2 — Title competes with the body text (same size/weight).
Fix: bump the title to ~1.5× and bold it so it reads first.
2. [High] Page 4 — White caption over a light photo is hard to read.
Fix: darken the image or add a scrim; or move the caption to a solid band.
### Page-by-page
**Page 1** — [Med] Three different accent colors; pick one. [Low] "recieve" → "receive".
**Page 2** — ...
### What's working
- Consistent margins; strong cover image.Use severities High / Med / Low. Lead with the few highest-impact items, then the per-page detail. Be specific and located (page + element), not generic ("make it pop").
End by offering to implement the API-fixable items via canva-edit-design, and note which items need manual work in Canva (e.g. font-family or background changes the API can't touch — see canva-edit-design for the full CANNOT list).
cancel-editing-transaction.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 6,203 | 4,476 | -28% | 1 | 1 | 0% | 1,166 | 1,408 | +21% | 0 | 0 | — |
case-02 | fail→fail | 15,188 | 5,503 | -64% | 1 | 1 | 0% | 2,123 | 1,484 | -30% | 0 | 0 | — |
case-03 | fail→fail | 5,452 | 4,274 | -22% | 1 | 1 | 0% | 954 | 1,389 | +46% | 0 | 0 | — |
case-04 | fail→pass | 4,384 | 10,090 | +130% | 1 | 1 | 0% | 852 | 3,054 | +258% | 0 | 0 | — |
case-14 | fail→fail | 8,996 | 4,975 | -45% | 1 | 1 | 0% | 1,672 | 1,937 | +16% | 0 | 0 | — |
case-05 | fail→fail | 3,636 | 3,467 | -5% | 1 | 1 | 0% | 658 | 1,679 | +155% | 0 | 0 | — |
case-06 | fail→pass | 8,104 | 4,682 | -42% | 1 | 1 | 0% | 1,665 | 1,990 | +20% | 0 | 0 | — |
case-07 | fail→fail | 4,866 | 5,383 | +11% | 1 | 1 | 0% | 827 | 1,507 | +82% | 0 | 0 | — |
case-08 | pass→pass | 4,265 | 2,064 | -52% | 1 | 1 | 0% | 753 | 1,458 | +94% | 0 | 0 | — |
case-09 | fail→fail | 10,463 | 5,884 | -44% | 1 | 1 | 0% | 1,769 | 1,515 | -14% | 0 | 0 | — |
case-10 | pass→pass | 8,757 | 2,826 | -68% | 1 | 1 | 0% | 1,292 | 1,611 | +25% | 0 | 0 | — |
case-11 | pass→pass | 8,206 | 3,556 | -57% | 1 | 1 | 0% | 1,541 | 1,651 | +7% | 0 | 0 | — |
case-12 | fail→pass | 6,432 | 4,490 | -30% | 1 | 1 | 0% | 1,173 | 1,834 | +56% | 0 | 0 | — |
case-13 | fail→pass | 9,923 | 4,498 | -55% | 1 | 1 | 0% | 1,659 | 1,897 | +14% | 0 | 0 | — |
case-15 | pass→pass | 5,578 | 4,680 | -16% | 1 | 1 | 0% | 949 | 1,903 | +101% | 0 | 0 | — |
case-16 | fail→fail | 4,304 | 5,475 | +27% | 1 | 1 | 0% | 784 | 1,494 | +91% | 0 | 0 | — |
case-17 | fail→fail | 7,027 | 7,698 | +10% | 1 | 1 | 0% | 1,245 | 1,662 | +33% | 0 | 0 | — |
case-18 | fail→fail | 10,091 | 6,068 | -40% | 1 | 1 | 0% | 1,883 | 1,608 | -15% | 0 | 0 | — |
case-19 | fail→pass | 6,458 | 2,876 | -55% | 1 | 1 | 0% | 1,182 | 1,607 | +36% | 0 | 0 | — |
case-20 | pass→fail | 9,789 | 5,534 | -43% | 1 | 1 | 0% | 1,821 | 1,411 | -23% | 0 | 0 | — |
case-21 | pass→fail | 12,926 | 5,279 | -59% | 1 | 1 | 0% | 1,816 | 1,467 | -19% | 0 | 0 | — |
case-22 | pass→pass | 6,207 | 2,860 | -54% | 1 | 1 | 0% | 1,273 | 1,621 | +27% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 12 counted toward the lift figure. The other 10 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +14 percentage points is the difference between those two pass rates over the 12 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.