gsp-project-critique

---

name: gsp-project-critique description: Critique designs + accessibility audit (creative phase — benefits from capable models) — use when: review the designs, critique this, check accessibility, what's wrong with, give feedback on user-invocable: true context: fork allowed-tools:

• Read
• Write
• Bash
• Agent

---

Phase 4 of the GSP project diamond. Runs two agents in parallel: the Design Critique Partner (Nielsen's 10 heuristics) and the Accessibility Auditor (WCAG 2.2 AA).

Works with the dual-diamond architecture: reads brand context from .design/branding/{brand}/ via brand.ref, reads/writes project assets in .design/projects/{project}/.

Critique design quality and audit accessibility compliance.

Input: All prior project chunks + brand identity Output: {project}/critique/ (critique + accessibility chunks + INDEX.md) + exports/INDEX.md update Agents: gsp-project-critic + gsp-accessibility-auditor Output mode: Honor ${CLAUDE_SKILL_DIR}/../../policies/output-modes.md per preferences.project_size (default compact). Chunk file counts and consolidation rules live in the policy.

## Step 0: Resolve project and brand

If .design/projects/ does not exist: output "No GSP project found. Run /gsp-start to begin." and stop.

Resolve project from .design/projects/ (one → use it, multiple → ask). Set PROJECT_PATH.

Read {PROJECT_PATH}/brand.ref → set BRAND_PATH.

Step 1: Load context

Read {PROJECT_PATH}/config.json to get implementation_target, design_scope, and accessibility_level.

Load all prior chunks (chunk-first with fallbacks)

BRIEF: {PROJECT_PATH}/BRIEF.md

Identity: Read {BRAND_PATH}/identity/INDEX.md → load all chunks.

Patterns: Read {BRAND_PATH}/patterns/INDEX.md → load all chunks.

Design: Read {PROJECT_PATH}/design/INDEX.md → load all chunks.

Brief: Read {PROJECT_PATH}/brief/INDEX.md → load all chunks (if exists).

Research: Read {PROJECT_PATH}/research/INDEX.md → load recommendations.md (if exists).

Step 1.5: Scope check

If design_scope is tokens:

1. Check if {PROJECT_PATH}/critique/accessibility-token-audit.md exists (from prior /gsp-accessibility --tokens). If yes, reference it and skip inline token checks. If no, suggest running /gsp-accessibility --tokens for detailed token contrast analysis.
2. Review system chunks only — token foundations, naming, scale consistency
3. Run accessibility audit on color contrast and token values only (unless prior token audit exists)
4. Write results to {PROJECT_PATH}/critique/accessibility-audit.md and accessibility-fixes.md
5. Write {PROJECT_PATH}/critique/INDEX.md
6. Update STATE.md — set Phase 4 to complete
7. Route: "Run /gsp-project-build."
8. Stop here

Otherwise: If design chunks don't exist and scope is not tokens, tell the user to complete the design phase first.

Step 1.8: Load critique references

Read these reference files and hold their content for inlining into agent prompts in Step 2:

• ${CLAUDE_SKILL_DIR}/../gsp-accessibility/wcag-checklist.md
• ${CLAUDE_SKILL_DIR}/../gsp-color/references/color-composition.md
• ${CLAUDE_SKILL_DIR}/../gsp-typography/domains/scale.md — type-scale verification rules
• ${CLAUDE_SKILL_DIR}/../gsp-visuals/domains/imagery.md — imagery vocabulary for critique
• ${CLAUDE_SKILL_DIR}/../gsp-accessibility/methodology/gsp-accessibility-auditor.md
• ${CLAUDE_SKILL_DIR}/../../templates/phases/critique.md — critique output template

Note: Nielsen's heuristics, visual taste, and anti-patterns are distilled into the gsp-project-critic agent prompt. anti-patterns.md is a critic-owned consolidated checklist; canonical sources are gsp-typography, gsp-color, gsp-visuals — update those when fixing drift, not the consolidated checklist alone.

Step 1.9: Load agent methodology

Read ${CLAUDE_SKILL_DIR}/methodology/gsp-project-critic.md. Include the full content as Agent methodology in the gsp-project-critic agent prompt below.

Step 2: Spawn critics (parallel)

Inline all project content — agents should not need to read project files. Reference files for supplementary evaluation (visual-taste, anti-patterns) are on disk — the critic reads them as needed.

Agent 1: gsp-project-critic — Pass in the agent prompt:

• Agent methodology (loaded in Step 1.9)
• Content of all design chunks (loaded in Step 1)
• Content of all identity chunks (loaded in Step 1)
• Content of all patterns chunks (loaded in Step 1)
• Content of STYLE.md from {BRAND_PATH}/patterns/ (if exists) — the critic checks designs against STYLE.md constraints, patterns, effects vocabulary, and bold bets
• Content of brief chunks (loaded in Step 1)
• Content of research recommendations.md (loaded in Step 1)
• Content of BRIEF.md
• Content of color composition reference (loaded in Step 1.8)
• Content of typography scale reference (loaded in Step 1.8)
• Content of imagery vocabulary reference (loaded in Step 1.8)
• Content of critique output template (loaded in Step 1.8)
• references_path: ${CLAUDE_SKILL_DIR}/ — for supplementary Read access to visual-taste.md, anti-patterns.md
• Output path: {PROJECT_PATH}/critique/

Agent 2: gsp-accessibility-auditor — Check if {PROJECT_PATH}/critique/accessibility-audit.md already exists from a prior /gsp-accessibility run. If yes, skip spawning the accessibility auditor — reuse the existing output. If no, pass in the agent prompt:

• Content of all design chunks (loaded in Step 1)
• Content of identity color-system.md and typography.md (loaded in Step 1)
• Content of patterns tokens chunks (loaded in Step 1)
• Content of WCAG checklist reference (loaded in Step 1.8)
• Agent methodology (loaded in Step 1.8)
• accessibility_level from config (defaults to "WCAG 2.2 AA")
• Output path: {PROJECT_PATH}/critique/

Model assignment: Spawn gsp-accessibility-auditor with model: sonnet. The accessibility audit is checklist-based (WCAG compliance) and works well on Sonnet. This splits rate-limit pressure — the critic runs on the user's current model while the auditor runs on Sonnet.

Step 3: Write critique INDEX.md

After both agents complete, write {PROJECT_PATH}/critique/INDEX.md:

# Critique
> Phase: critique | Project: {name} | Generated: {DATE}

## Critique

| Chunk | File | ~Lines |
|-------|------|--------|
| Critique | [critique.md](./critique.md) | ~{N} |
| Prioritized Fixes | [prioritized-fixes.md](./prioritized-fixes.md) | ~{N} |
| Alternative Directions | [alternative-directions.md](./alternative-directions.md) | ~{N} |
| Strengths | [strengths.md](./strengths.md) | ~{N} |

## Accessibility

| Chunk | File | ~Lines |
|-------|------|--------|
| Accessibility Audit | [accessibility-audit.md](./accessibility-audit.md) | ~{N} |
| Accessibility Fixes | [accessibility-fixes.md](./accessibility-fixes.md) | ~{N} |

Update {PROJECT_PATH}/exports/INDEX.md:

<!-- BEGIN:critique -->
| Section | File |
|---------|------|
| Critique | [critique.md](../critique/critique.md) |
| Prioritized Fixes | [prioritized-fixes.md](../critique/prioritized-fixes.md) |
| Alternative Directions | [alternative-directions.md](../critique/alternative-directions.md) |
| Strengths | [strengths.md](../critique/strengths.md) |
| Accessibility Audit | [accessibility-audit.md](../critique/accessibility-audit.md) |
| Accessibility Fixes | [accessibility-fixes.md](../critique/accessibility-fixes.md) |
<!-- END:critique -->

Step 4: Assess results

Read critique/critique.md for the heuristics score (X/50) and brand contract score (X/25, present when STYLE.md was used). Read critique/prioritized-fixes.md for critical issues. Determine verdict:

Pass: Nielsen ≥ 40/50 AND brand contract ≥ 20/25 (when present) AND no critical fixes. Design is solid, proceed to build. Conditional Pass: Nielsen 30-39/50 OR brand contract 15-19/25 OR critical fixes are minor. Shippable with notes, proceed to build. Fail: Nielsen < 30/50 OR brand contract < 15/25 OR any brand contract dimension at 1 (constraint violation) OR critical fixes affect layout/navigation/IA. Design needs revision before building.

Note: If no STYLE.md was used, the brand contract score is absent — apply Nielsen thresholds only.

Step 5: Update state

Update {PROJECT_PATH}/STATE.md:

• Set Phase 4 (Critique) status to complete or needs-revision
• Record review loop count and completion date

Critique→Design loop — if Fail

If verdict is Fail:

1. Set Phase 4 (Critique) status to needs-revision
2. Set Phase 3 (Design) status to needs-revision
3. Ensure critique/prioritized-fixes.md and critique/accessibility-fixes.md contain actionable issues

Step 6: Phase transition output

Invoke /gsp-phase-transition with phase critique and output directory {PROJECT_PATH}/critique/.

If critique identified brand-level issues (palette contrast, typography weight, spacing scale), note: "Some issues are brand-level — run /gsp-brand-refine to adjust tokens without re-running identity."