@ryuxdsgn/design-intelligence 0.0.0-stage → 2.6.1

package/dist/intelligence/critique.js

@@ -0,0 +1,171 @@
+import { CAPTURE_TABLE } from "./capture.js";
+// RYUX Critique: the review workflow. It orchestrates Analyze, the knowledge skills, and RYUX
+// Knowledge into findings. Source of truth for skills/ryux-critique/SKILL.md (group "critique").
+export const CRITIQUE_DESCRIPTION = "RYUX Critique - senior UX and UI design critique for Figma, pen.dev, websites, screenshots, and built interfaces. Analyzes first, runs a Design Read across nine dimensions, then reports prioritized findings with severity, evidence, impact, recommendation, confidence, and source, plus what to keep. Use when asked to critique, review, or audit a UI, screen, or flow.";
+export const CRITIQUE_BODY = `# RYUX Critique
+
+> Get a senior design critique before your users do.
+
+Critique answers "does this interface make sense, and what should change first?" It judges
+decisions; it is not Analyze (what exists) or QA (does the build match its design). It is an
+orchestrator, not a pile of rules: it analyzes the design, reasons with the RYUX knowledge modules,
+pulls evidence from RYUX Knowledge, and reports findings someone can act on.
+
+\`\`\`
+Critique
+     ↓
+Analyze                   what is actually there
+     ↓
+knowledge modules         ux, ui, interaction, forms, content, accessibility,
+                          responsive, design-system, edge-cases, anti-slop
+     ↓
+RYUX Knowledge            real product screens (search_screens, heuristic_eval)
+     ↓
+findings
+\`\`\`
+
+## Principles
+
+1. **Review what you can see.** Capture the real design first. Never review from a description
+   alone when the source is reachable.
+2. **Read-only.** Do not edit the Figma file, the pen.dev document, or the site.
+3. **Evidence over taste.** Every finding names its source; severity 3 or more needs a real
+   comparison \`screen_id\` when the ryux MCP is connected.
+4. **Honest scope.** A static frame shows no hover, focus, loading, or narrow widths. Say what was
+   not tested, and lower the confidence of anything inferred.
+5. **Concise and prioritized.** At most 12 findings; the ones that hurt users most come first.
+
+## Steps
+
+### 1. Context
+
+Who the user is, what task happens on this screen, the platform, and the point in the flow. If the
+request does not say, infer it from the design and list it as an assumption.
+
+### 2. Analyze first
+
+Run Analyze (\`ryux-analyze\`: capture plus inventory) or reuse an analysis that already exists. Its
+labeled inventory (Measured, Observed, Inferred) is what the findings point at. For a quick
+critique, capture the design with this table and note what you saw:
+
+${CAPTURE_TABLE}
+
+### 3. Design Read
+
+Rate each dimension **Strong**, **Adequate**, or **Weak**, with one sentence of reason.
+
+| Dimension | Question |
+| --- | --- |
+| Clarity | Can the user understand what this screen is for? |
+| Hierarchy | Can the user identify what matters first? |
+| Coherence | Does it feel like part of the same product? |
+| Density | Is the information density right for how often and how carefully it is used? |
+| Confidence | Does the user have enough information to act safely (cost, consequence, status)? |
+| Efficiency | Can the task be done without unnecessary work? |
+| Specificity | Does it feel designed for this product, or could it belong to any generic SaaS app? |
+| Recoverability | Can users recover from mistakes and failures? |
+| Accessibility | Can people with different abilities and input methods operate and understand it? |
+
+### 4. Fact check
+
+List every product fact the design states: names of features, tools, and commands, integrations,
+supported platforms, numbers, dates, prices, people, and claims. Mark each one sourced (the brief,
+the repo, the product's docs, a screen_id) or unsourced. Every unsourced fact is a finding under
+RX-PR-02 or RX-AS-01, even when it looks plausible, and even inside a sample or example panel.
+
+### 5. Evaluate by category
+
+Read the knowledge modules that apply and walk the design with their rules. [Required] rules are the
+gate, [Preferred] rules need a reason when broken, and [Contextual] rules count only when their
+situation is present. Use their "Not when" lines so a rule is not applied mechanically.
+
+| Category | Skill | Look for |
+| --- | --- | --- |
+| Product | ryux-product | stated purpose, one primary action, invented facts |
+| UX | ryux-ux | structure, wayfinding, grouping, disclosure, steps |
+| Interaction | ryux-interaction | before, during, result, recovery; confirm or undo; cost upfront; local payments |
+| Forms | ryux-forms | labels, validation, input kept, submission states |
+| Content | ryux-content | specific labels, errors, terminology, Rupiah, dates and numbers |
+| UI | ryux-ui | hierarchy, scale, color roles, density; each visual's job, composition around the copy, motion level; the result against its Visual Brief when there is one |
+| Design system | ryux-design-system | consistency, tokens, component states |
+| Accessibility | ryux-accessibility | contrast, targets, focus, names, color alone |
+| Responsive | ryux-responsive | what changes, stays, disappears, or stacks across widths |
+| Edge cases | ryux-edge-cases | empty, error, loading, long text, permissions |
+| Anti-slop | ryux-anti-slop | invented numbers or people, fake urgency, unjustified decoration |
+
+### 6. Evidence from RYUX Knowledge
+
+With the ryux MCP connected, find 1 to 3 comparison screens for each major finding with
+\`search_screens\`, then send the findings to \`heuristic_eval\` (\`task_context\` + \`findings\`).
+Its \`heuristic\` field takes the Nielsen codes \`H-01\` to \`H-10\`; map each finding and name the
+RYUX rule in \`issue\`:
+
+| Category | Map to |
+| --- | --- |
+| Edge cases, interaction feedback | \`H-01\` visibility of system status |
+| Content, the user's words | \`H-02\` match with the real world |
+| Exits and wayfinding | \`H-03\` user control and freedom |
+| Design system, accessibility | \`H-04\` consistency and standards |
+| Confirm or undo, forms | \`H-05\` error prevention |
+| Recognition | \`H-06\` recognition rather than recall |
+| Shortcuts | \`H-07\` flexibility and efficiency |
+| UI decoration, anti-slop | \`H-08\` aesthetic and minimalist design |
+| Error messages and recovery | \`H-09\` error recovery |
+| Help | \`H-10\` help and documentation |
+
+The tool rejects major findings without valid evidence, caps the list at 12, and returns PASS or
+FAIL. Without the MCP, say the evidence comes from the design and standards alone.
+
+### 7. Report
+
+Each finding uses this schema:
+
+| Field | Content |
+| --- | --- |
+| ID | C-01, C-02, ... |
+| Severity | 0 not a problem · 1 cosmetic · 2 minor · 3 major · 4 catastrophic |
+| Category | one of the categories above |
+| Finding | where it is and what the problem is |
+| Evidence | what you saw: the capture, a measured value, a comparison screen |
+| Impact | who is affected, in which task, and how badly |
+| Recommendation | a concrete fix |
+| Confidence | High (seen directly), Medium (partly seen, such as one width), Low (inferred) |
+| Source | the RYUX rule ID, plus a screen_id, a standard (WCAG SC), or "this design" |
+
+\`\`\`
+Verdict: one sentence (what works, the biggest risk)
+
+What was reviewed and how
+  Source:      Figma frame "Checkout" (node 1:2) / https://... / pen.dev frame "..."
+  Analysis:    ryux-analyze, desktop 1440 and mobile 390, via Playwright
+  Not tested:  hover and focus states, live data, screen reader
+
+Design Read
+  Clarity       Strong    · ...
+  Confidence    Weak      · ...
+  (all nine)
+
+Findings (most severe first, at most 12)
+  C-01 · severity 3 · Interaction · confidence High
+    Finding:         Payment confirmation shows no status while verifying
+    Evidence:        no loading or progress element in the frame (Observed)
+    Impact:          buyers paying by QRIS may tap twice and pay twice
+    Recommendation:  show status with a time estimate; disable the button while verifying
+    Source:          RX-IX-02; scr_demo_001
+
+What works (keep it)
+  - ...
+
+Fix priorities
+  1. C-01 ...
+\`\`\`
+
+## Don'ts
+
+- Don't review a design you have not captured when it is reachable.
+- Don't edit the source file while reviewing.
+- Don't give a finding without a location, a source, and a reason.
+- Don't raise severity to major without evidence, or claim "app X does this" without a \`screen_id\`.
+- Don't mark an inferred issue High confidence.
+- Don't paste generic advice ("modernize", "add white space") without a real problem.
+- Don't call the result "fully accessible" or "UX optimized"; say what was checked.`;