Agency Evaluation Criteria Skill
Governs quality assessment of all agency project deliverables. Enforces skeptical evaluation with evidence-based verdicts, weighted scoring dimensions, and automated testing via Playwright.
Static Zone
Identity
Purpose: Define evaluation criteria, scoring weights, pass/fail thresholds, and testing requirements for agency project quality assessment.
Input Contract:
- Built application (URL or local path)
- Original copy.md (for copy integrity verification)
- Original design-spec.md (for design compliance verification)
- BRIEF document (for completeness verification)
Output Contract:
evaluation-report.md containing:
- Overall score (0.00 - 1.00) with PASS/FAIL verdict
- Per-dimension scores with evidence
- Specific defect list with file:line references
- Screenshots (desktop + mobile)
- Improvement recommendations
Owner: .agency/context/quality-standards.md
Core Principles
Derived from Brand Context. Never auto-modified. Manual editing only.
- Skeptical by default — tuned to find defects, not rationalize acceptance
- Evidence-based verdicts only — no PASS without concrete proof
- Copy integrity is non-negotiable — any deviation from original copy = FAIL
- AI slop detection — purple gradients + white cards + generic icons = FAIL
- When in doubt, FAIL — false negatives are costlier than false positives
Default Evaluation Weights
| Dimension |
Weight |
Description |
| Design Quality |
30% |
Visual consistency, brand alignment, polish |
| Originality |
25% |
Not generic/template-like, unique approach |
| Completeness |
25% |
All BRIEF sections present, copy accurate |
| Functionality |
20% |
Responsive, accessible, all interactions work |
Hard Thresholds (always FAIL)
- Copy text differs from original copy.md
- AI slop detected (generic purple gradient + white card layout)
- Mobile viewport broken (content overflow or unreadable)
- CTA count > 1 per page (unless BRIEF explicitly specifies)
- Any link returns 404
- Lighthouse Accessibility < 80
Testing Requirements
Phase 1: Visual Verification
- Desktop screenshot (1280x720): full page capture
- Mobile screenshot (375x667): full page capture
- Compare against design-spec.md layout expectations
- Check for AI slop indicators (purple gradients, white cards, generic stock icons)
Phase 2: Interaction Testing
- Click test: all buttons, links, CTAs — verify navigation and state changes
- Form test: all input fields with valid/invalid data, verify validation messages
- Scroll test: full page traversal, verify lazy loading and fixed elements
- Keyboard test: Tab navigation order, Enter/Space activation on interactive elements
Phase 3: Responsive Verification
- Mobile viewport: no horizontal overflow, readable text (min 16px)
- Touch targets: minimum 44x44px for all interactive elements
- Mobile navigation: hamburger menu functionality, swipe gestures
- Breakpoint transitions: verify layout changes at 768px, 1024px, 1280px
Phase 4: Performance Audit
- Lighthouse scores: Performance >= 80, Accessibility >= 90, Best Practices >= 80, SEO >= 80
- Core Web Vitals: LCP < 2.5s, FID < 100ms, CLS < 0.1
- Bundle size check: main bundle < 200KB gzipped (unless justified)
Tool Priority
- claude-in-chrome MCP (preferred): Live browser interaction with screenshots
- Playwright via Bash (fallback): Headless testing when MCP unavailable
- Static analysis (minimum): HTML/CSS validation, link checking
Dynamic Zone
Weights, thresholds, and test scenarios evolve via user feedback.
Rules
(No rules yet. Rules will be added as the system learns from user feedback.)
Anti-Patterns
(No anti-patterns yet.)
Heuristics
(No heuristics yet.)
Evolution Log
- v1.0.0: Initial creation (Static Zone with default weights, empty Dynamic Zone)
1---2name: agency-evaluation-criteria3description: Quality evaluation criteria for AI Agency project output covering design quality, originality, completeness, and functionality scoring with weighted dimensions and Playwright-based testing requirements.4license: Apache-2.05---67# Agency Evaluation Criteria Skill89Governs quality assessment of all agency project deliverables. Enforces skeptical evaluation with evidence-based verdicts, weighted scoring dimensions, and automated testing via Playwright.1011---1213## Static Zone1415### Identity1617**Purpose**: Define evaluation criteria, scoring weights, pass/fail thresholds, and testing requirements for agency project quality assessment.1819**Input Contract**:20- Built application (URL or local path)21- Original copy.md (for copy integrity verification)22- Original design-spec.md (for design compliance verification)23- BRIEF document (for completeness verification)2425**Output Contract**:26- `evaluation-report.md` containing:27 - Overall score (0.00 - 1.00) with PASS/FAIL verdict28 - Per-dimension scores with evidence29 - Specific defect list with file:line references30 - Screenshots (desktop + mobile)31 - Improvement recommendations3233**Owner**: `.agency/context/quality-standards.md`3435### Core Principles3637> Derived from Brand Context. Never auto-modified. Manual editing only.38391. Skeptical by default — tuned to find defects, not rationalize acceptance402. Evidence-based verdicts only — no PASS without concrete proof413. Copy integrity is non-negotiable — any deviation from original copy = FAIL424. AI slop detection — purple gradients + white cards + generic icons = FAIL435. When in doubt, FAIL — false negatives are costlier than false positives4445### Default Evaluation Weights4647| Dimension | Weight | Description |48|-----------|--------|-------------|49| Design Quality | 30% | Visual consistency, brand alignment, polish |50| Originality | 25% | Not generic/template-like, unique approach |51| Completeness | 25% | All BRIEF sections present, copy accurate |52| Functionality | 20% | Responsive, accessible, all interactions work |5354### Hard Thresholds (always FAIL)5556- Copy text differs from original copy.md57- AI slop detected (generic purple gradient + white card layout)58- Mobile viewport broken (content overflow or unreadable)59- CTA count > 1 per page (unless BRIEF explicitly specifies)60- Any link returns 40461- Lighthouse Accessibility < 806263### Testing Requirements6465#### Phase 1: Visual Verification66- Desktop screenshot (1280x720): full page capture67- Mobile screenshot (375x667): full page capture68- Compare against design-spec.md layout expectations69- Check for AI slop indicators (purple gradients, white cards, generic stock icons)7071#### Phase 2: Interaction Testing72- Click test: all buttons, links, CTAs — verify navigation and state changes73- Form test: all input fields with valid/invalid data, verify validation messages74- Scroll test: full page traversal, verify lazy loading and fixed elements75- Keyboard test: Tab navigation order, Enter/Space activation on interactive elements7677#### Phase 3: Responsive Verification78- Mobile viewport: no horizontal overflow, readable text (min 16px)79- Touch targets: minimum 44x44px for all interactive elements80- Mobile navigation: hamburger menu functionality, swipe gestures81- Breakpoint transitions: verify layout changes at 768px, 1024px, 1280px8283#### Phase 4: Performance Audit84- Lighthouse scores: Performance >= 80, Accessibility >= 90, Best Practices >= 80, SEO >= 8085- Core Web Vitals: LCP < 2.5s, FID < 100ms, CLS < 0.186- Bundle size check: main bundle < 200KB gzipped (unless justified)8788#### Tool Priority891. claude-in-chrome MCP (preferred): Live browser interaction with screenshots902. Playwright via Bash (fallback): Headless testing when MCP unavailable913. Static analysis (minimum): HTML/CSS validation, link checking9293---9495## Dynamic Zone9697> Weights, thresholds, and test scenarios evolve via user feedback.9899### Rules100101(No rules yet. Rules will be added as the system learns from user feedback.)102103### Anti-Patterns104105(No anti-patterns yet.)106107### Heuristics108109(No heuristics yet.)110111---112113## Evolution Log114115- v1.0.0: Initial creation (Static Zone with default weights, empty Dynamic Zone)
modu-ai