design-review

Name: design-review
Author: fusengine

21stars3forks

How to install design-review?

fusengine/agents/design-review

$npx -y skills add fusengine/agents --skill design-review

Installs into the current project.

Summary

The final quality gate before any design deliverable is reported done — two ISOLATED assessments (deterministic detection never in the same head as the subjective/visual read), P0-P3 findings with NO aggregate score ever, a bounded visual loop (max 2 fix cycles), and a mandatory

Files · 1

View on GitHub

SKILL.md

1	## Design Review — Final Quality Gate
2
3	### When
4	After `design-motion` (web/webapp) or directly after the mockup (`design-ios`/
5	`design-android`). The last step before reporting any design deliverable done — never
6	skipped, never reported as passed if it wasn't actually run
7	(`references/review-procedure.md` Failure Handling).
8
9	### Input
10	- The complete rendered artifact: components/mockup with animations (if any) and
11	interactive states.
12	- `design-system.md` as the audit baseline (tokens, forbidden fonts, contrast floors).
13	- The 3 elements declared in `design-web`/`design-webapp` ("Je reproduis: {el1}, {el2},
14	{el3}") and any declared premium pattern(s); the signature element
15	(`design-method` Step 2) instead gets a Signature Dominance verdict.
16
17	### The two assessments — ISOLATED, never in the same head
18	This is the structural rule the whole procedure hangs on: the mechanical read and the
19	subjective read must not anchor on each other.
20
21	1. Deterministic detection (`references/anti-ai-slop-audit.md` +
22	`references/pre-flight-checklist.md`) — grep/count/compute, zero taste involved. Run
23	by `design-expert` itself; there is nothing to isolate here because there is no
24	judgment call to bias.
25	2. Subjective visual review — screenshots, named elicitation techniques
26	(`references/elicitation-visual.md`), Nielsen heuristics
27	(`references/ux-nielsen.md`), UX laws (`references/ux-laws.md`). This pass is run by
28	`design-expert`, so it is not yet independent — independence is delivered by the
29	mandatory challenger gate below, which judges blind, fresh-context, off the rendered
30	PNG only, never this procedure's own reasoning. Do not treat the elicitation-technique
31	pass alone as "isolated" — it isn't; the challenger is the isolation.
32
33	Never collapse these into one pass "to save a round" — that is exactly how the
34	cream/serif/terracotta cluster shipped through this audit once already (documented in
35	`anti-ai-slop-audit.md` Gate Semantics).
36
37	### Part 1 — Deterministic Checks (mechanical, run first)
38	Full 15-item list is canonical in `references/review-procedure.md` Part 1 — contrast
39	(4.5:1 text / 3:1 UI, light+dark), forbidden fonts, OKLCH-only color format, em-dash
40	crutch (2+ occurrences), token adherence, the anti-AI-slop audit
41	(`references/anti-ai-slop-audit.md`, 14 detectors — clusters 9-11 are
42	FLAG-with-justification, not a block, if declared per `design-method` Step 2), the
43	mechanical pre-flight (`references/pre-flight-checklist.md`), WCAG beyond contrast
44	(`references/ux-wcag.md`), cross-component consistency (`references/consistency-checks.md`),
45	mobile nav functionality, doc↔code animation diff, integrity (no fabricated numbers/false
46	urgency), no-JS baseline, Type-Scale/Body-Size/Measure/Focal-Block floors, and
47	Dark-Elevation Direction. Any Critical/Major finding here is fixed before Part 2 runs.
48
49	### Part 2 — Bounded Visual Review
50	Serve → screenshot per section + one full-page capture, light AND dark via
51	`colorScheme` → cross-viewport (mobile/tablet/desktop) → compare declared elements
52	(binary present/absent, no partial credit; signature element gets Signature Dominance) →
53	localized critique only (name the exact section/element, never "improve the style") →
54	motion verdict if any animation exists (`references/motion-verdict.md`, Before/After/Why
55	table + tiered impact + explicit Block/Approve) → cite at least two named techniques
56	from `references/elicitation-visual.md` (Squint / Subtraction / Competitor Line-up /
57	5-Second / Persona) → fix gaps. Full step-by-step: `references/review-procedure.md` Part 2.
58
59	Bounded loop — maximum 2 fix cycles. The loop exits PASS only once the
60	register-applicable positive floors are actually met (not merely "0 flags"). Issues
61	(or an unmet floor) remaining after cycle 2 stop the loop and get reported, not
62	chased further. A plateau — cycle 2 repeating cycle 1's finding — stops immediately, even
63	at cycle 1 of 2. This is a hard cap, not a target to exceed when "almost there."
64
65	### Findings — P0-P3, NO aggregate score, ever
66	Tag every finding with severity, never sum or average them into a total:
67
68	\| Tier \| Meaning \|
69	\|---\|---\|
70	\| P0 Blocking \| Prevents task

People who liked this also viewed

6 picks

Type

design-review

21stars3forks

View on GitHub

How to install design-review?

fusengine/agents/design-review

$npx -y skills add fusengine/agents --skill design-review

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	## Design Review — Final Quality Gate
2
3	### When
4	After `design-motion` (web/webapp) or directly after the mockup (`design-ios`/
5	`design-android`). The last step before reporting any design deliverable done — never
6	skipped, never reported as passed if it wasn't actually run
7	(`references/review-procedure.md` Failure Handling).
8
9	### Input
10	- The complete rendered artifact: components/mockup with animations (if any) and
11	interactive states.
12	- `design-system.md` as the audit baseline (tokens, forbidden fonts, contrast floors).
13	- The 3 elements declared in `design-web`/`design-webapp` ("Je reproduis: {el1}, {el2},
14	{el3}") and any declared premium pattern(s); the signature element
15	(`design-method` Step 2) instead gets a Signature Dominance verdict.
16
17	### The two assessments — ISOLATED, never in the same head
18	This is the structural rule the whole procedure hangs on: the mechanical read and the
19	subjective read must not anchor on each other.
20
21	1. Deterministic detection (`references/anti-ai-slop-audit.md` +
22	`references/pre-flight-checklist.md`) — grep/count/compute, zero taste involved. Run
23	by `design-expert` itself; there is nothing to isolate here because there is no
24	judgment call to bias.
25	2. Subjective visual review — screenshots, named elicitation techniques
26	(`references/elicitation-visual.md`), Nielsen heuristics
27	(`references/ux-nielsen.md`), UX laws (`references/ux-laws.md`). This pass is run by
28	`design-expert`, so it is not yet independent — independence is delivered by the
29	mandatory challenger gate below, which judges blind, fresh-context, off the rendered
30	PNG only, never this procedure's own reasoning. Do not treat the elicitation-technique
31	pass alone as "isolated" — it isn't; the challenger is the isolation.
32
33	Never collapse these into one pass "to save a round" — that is exactly how the
34	cream/serif/terracotta cluster shipped through this audit once already (documented in
35	`anti-ai-slop-audit.md` Gate Semantics).
36
37	### Part 1 — Deterministic Checks (mechanical, run first)
38	Full 15-item list is canonical in `references/review-procedure.md` Part 1 — contrast
39	(4.5:1 text / 3:1 UI, light+dark), forbidden fonts, OKLCH-only color format, em-dash
40	crutch (2+ occurrences), token adherence, the anti-AI-slop audit
41	(`references/anti-ai-slop-audit.md`, 14 detectors — clusters 9-11 are
42	FLAG-with-justification, not a block, if declared per `design-method` Step 2), the
43	mechanical pre-flight (`references/pre-flight-checklist.md`), WCAG beyond contrast
44	(`references/ux-wcag.md`), cross-component consistency (`references/consistency-checks.md`),
45	mobile nav functionality, doc↔code animation diff, integrity (no fabricated numbers/false
46	urgency), no-JS baseline, Type-Scale/Body-Size/Measure/Focal-Block floors, and
47	Dark-Elevation Direction. Any Critical/Major finding here is fixed before Part 2 runs.
48
49	### Part 2 — Bounded Visual Review
50	Serve → screenshot per section + one full-page capture, light AND dark via
51	`colorScheme` → cross-viewport (mobile/tablet/desktop) → compare declared elements
52	(binary present/absent, no partial credit; signature element gets Signature Dominance) →
53	localized critique only (name the exact section/element, never "improve the style") →
54	motion verdict if any animation exists (`references/motion-verdict.md`, Before/After/Why
55	table + tiered impact + explicit Block/Approve) → cite at least two named techniques
56	from `references/elicitation-visual.md` (Squint / Subtraction / Competitor Line-up /
57	5-Second / Persona) → fix gaps. Full step-by-step: `references/review-procedure.md` Part 2.
58
59	Bounded loop — maximum 2 fix cycles. The loop exits PASS only once the
60	register-applicable positive floors are actually met (not merely "0 flags"). Issues
61	(or an unmet floor) remaining after cycle 2 stop the loop and get reported, not
62	chased further. A plateau — cycle 2 repeating cycle 1's finding — stops immediately, even
63	at cycle 1 of 2. This is a hard cap, not a target to exceed when "almost there."
64
65	### Findings — P0-P3, NO aggregate score, ever
66	Tag every finding with severity, never sum or average them into a total:
67
68	\| Tier \| Meaning \|
69	\|---\|---\|
70	\| P0 Blocking \| Prevents task

People who liked this also viewed

6 picks

Type