qualia-ux-prompts
Soba-noodl/qualia-ux-prompts
Qualia's production UX & accessibility audit prompts + a drop-in Claude Code skill. MIT.
Instalação
npx convez-ai Soba-noodl/qualia-ux-prompts1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
Qualia UX audit prompts
The production prompts that powered Qualia — an AI tool for fast, consistent UX & accessibility audits — extracted verbatim and annotated so you can read, learn from, and reuse them.
Qualia was shut down commercially in 2026 and open-sourced. The full application lives at Soba-noodl/qualia; this repo pulls just the prompt engineering out of the codebase, where it was otherwise buried inside 1,000-line TypeScript files.
The idea
Most "AI design review" is cosmetic nit-picking. Qualia's prompts are built around a different stance: act as an elite Strategic Product Design Lead and surface business-critical blockers and conversion killers, with findings that are scored and falsifiable — every verdict must be a specific claim that could be proven wrong. That posture is what these prompts encode.
Validation
These prompts are not theoretical. Their wording, scoring bands, and finding filters were tuned against real usage:
- 160+ real audits run by real users
- 43 in-depth user interviews
- 5 cycles of quantitative research across 100 users
- countless hands-on tests by the maker
That body of feedback is what calibrated what you see here.
What's here
prompts/ — image-based audits
The live product audited screenshots and designs. Each file is one production prompt, with its inputs and a link back to source.
| Prompt | Audits |
|---|---|
single-screen | one screen or design (the core prompt) |
user-flow | a multi-step flow as an ordered sequence |
auto-crawl | screenshots auto-captured from a live product |
figma-prototype | frames traversed from a Figma prototype |
synthetic-users | a role-played persona reacting to a product |
_node-map-anchoring | shared fragment: anchor issues to Figma layer IDs |
engines/ — code-based audits
A second system that audits a route's source code through six lenses, then aggregates.
| Engine | Lens |
|---|---|
| C | Cognitive load & visual hierarchy |
| D | Senior designer judgment |
| H | Usability heuristics & navigation |
| I | Interaction cost |
| SL | State, feedback, error handling, flow logic |
| X | Cross-sectional coherence (whole codebase) |
skills/ — drop-in Claude Code skill
q-auditing-qualia — a Claude Code skill that applies these prompts as Claude's own rubric, so Claude is the auditor. It is self-contained: the prompts it uses are bundled inside the skill folder (skills/q-auditing-qualia/prompts/), so it works in any project once copied in.
How to use it
Path A — in Claude Code (no setup beyond copying a folder).
Copy skills/q-auditing-qualia/ into your project's .claude/skills/, then run /q-auditing-qualia on a screen, flow, or route. Claude reads the bundled rubric and audits.
Path B — raw, with any vision LLM.
Open any file in prompts/, copy the block under ~~~text, fill in the {placeholders} (each file lists what they mean), and send it to a vision-capable model alongside your screenshot(s). The first image is the audit target; later images are context only.
The {placeholders} are slots the application filled at call time (the product's mission, the target persona, the screen description, etc.). They're documented per-file; supply your own values.
License
MIT — see LICENSE. The prompts are provided as-is for reference and reuse.