gsp-project-review

Phase 6 of the GSP project diamond. QA validates that the actual codebase implementation matches the design intent — checking real source files for token usage, screen coverage, component quality, and accessibility compliance.

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

QA validate the codebase implementation against design intent.

Input: BUILD-LOG.md + actual codebase files + git diff + design chunks + brand system
Output: {project}/review/ (acceptance-report.md + issues.md + INDEX.md) + exports/INDEX.md update
Agent: gsp-project-reviewer

@${CLAUDE_SKILL_DIR}/../../templates/phases/review.md </execution_context> ## 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, codebase_type.

Prior code accessibility audit: Check if {PROJECT_PATH}/review/accessibility-audit.md exists from a prior /gsp-accessibility-audit --code run. If yes, load it — the reviewer will reference these findings instead of performing inline a11y checks.

Load all artifacts

Build log: Read {PROJECT_PATH}/build/BUILD-LOG.md — what the builder says was implemented (files created, files modified, component map).

Actual codebase: Read the files listed in BUILD-LOG.md. Use Grep/Glob to find and verify actual source files.

Git diff: Run git diff (and git diff --cached) to see what actually changed in the codebase. Cross-reference against BUILD-LOG.md.

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

Brand patterns: Read {BRAND_PATH}/patterns/INDEX.md → load foundation + component chunks.
Also read {BRAND_PATH}/patterns/{brand-name}.yml (the brand's token/style source of truth).

Brand identity (selective): Read {BRAND_PATH}/identity/imagery-style.md (if exists) — needed for imagery audit.

Brief: Read {PROJECT_PATH}/brief/INDEX.md → load scope and adaptations.

Research: Read {PROJECT_PATH}/research/INDEX.md → load reference-specs.md (to verify specs were followed).

Critique: Read {PROJECT_PATH}/critique/INDEX.md → load prioritized-fixes and accessibility-fixes (to verify they were addressed).

Codebase context: Read .design/system/COMPONENTS.md (if exists) — what existed before build. Read {PROJECT_PATH}/codebase/MANIFEST.md (if exists) — what build claims it produced. Cross-reference both against BUILD-LOG.md and actual git diff.

Step 1.5: Scope check

If design_scope is tokens:

1. Run token-audit-only review: verify token naming, scale consistency, contrast ratios, and brand alignment. Skip screen coverage and component coverage checks.
2. Write {PROJECT_PATH}/review/acceptance-report.md (token-focused verdict) and {PROJECT_PATH}/review/issues.md (token issues only)
3. Write {PROJECT_PATH}/review/INDEX.md
4. Update {PROJECT_PATH}/exports/INDEX.md between <!-- BEGIN:review --> and <!-- END:review --> with populated table
5. Update {PROJECT_PATH}/STATE.md — set Phase 6 (Review) to complete or needs-revision
6. Route: display verdict or re-run /gsp-project-review
7. Stop here

Step 1.8: Load agent methodology

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

Step 2: Spawn reviewer

Spawn the gsp-project-reviewer agent with:

• Agent methodology (loaded in Step 1.8)
• BUILD-LOG.md contents
• Actual codebase file paths (from BUILD-LOG.md)
• git diff output
• Design chunks
• Brand system chunks + {brand-name}.yml
• Brief chunks
• Critique fixes (to verify resolution)
• .design/system/COMPONENTS.md (when exists — to verify existing components weren't broken)
• MANIFEST.md (when exists — to verify build claims match reality)
• codebase_type from config.json
• Review output template (from execution_context)
• Output path: {PROJECT_PATH}/review/
• Prior code accessibility findings (if {PROJECT_PATH}/review/accessibility-audit.md exists — tell reviewer to reference these instead of performing inline a11y checks; reviewer keeps inline checks as fallback if no prior audit exists)
• Clear instruction: "Review actual codebase files, not .design/build/ specs. Use Grep to search for hardcoded values. Use git diff to verify changes. Reference actual file paths in issues."

The agent writes chunks directly:

• review/acceptance-report.md
• review/issues.md
• review/INDEX.md

Step 3: Write exports

Update {PROJECT_PATH}/exports/INDEX.md:

<!-- BEGIN:review -->
| Section | File |
|---------|------|
| Acceptance Report | [acceptance-report.md](../review/acceptance-report.md) |
| Issues | [issues.md](../review/issues.md) |
<!-- END:review -->

Step 4: Assess results

Read review/acceptance-report.md for the verdict:

Pass: All screens implemented, tokens used correctly, accessibility compliant.
Conditional Pass: Minor issues found, but shippable.
Fail: Critical issues — must address before shipping.

Step 5: Update state

Update {PROJECT_PATH}/STATE.md:

• Set Phase 6 (Review) status to complete or needs-revision
• Record completion date
• If Pass or Conditional Pass: Set Prettiness Level to 100%
• Update ## Screen Build Status table — set Review Status per screen based on acceptance-report.md findings

If Pass or Conditional Pass, update .design/CLAUDE.md — replace the existing ### {project-name} entry (written by gsp-project-brief when started) with the completed entry:

### {project-name} · complete · {DATE}
brand: {brand-name} · .design/projects/{project-name}/

QA loop — if Fail

If verdict is Fail:

1. Set Phase 6 (Review) status to needs-revision
2. Set Phase 5 (Build) status to needs-revision
3. Ensure review/issues.md is written with actionable issues

Finalize git tracking

1. If git.branch is set in config.json:
   • Run gh pr list --head {branch} --json url,number --limit 1 to find an open PR
   • If found, update git.pr in config.json and STATE.md ## Git table
   • If not found, note: "No PR found for branch {branch}."
   • If gh is not available, skip silently — leave PR field as "—"
2. Include PR link in the CHANGELOG.md entry if available (see format below)

Update manifest + changelog

1. Update {PROJECT_PATH}/codebase/MANIFEST.md:

   • Update Status to complete (Pass) or partial (Conditional Pass)
   • Populate Branch and PR lines in the manifest header from config.json git values
   • Fix component paths if renamed during implementation

2. Append to .design/CHANGELOG.md:

   • Add entry with project name, date, brand, scope summary
   • List added/modified components, patterns, file count
   • Link to manifest for detail
   • Use this format:

     ## [{project-name}] — {DATE}
     > Brand: {brand} | Scope: {one-line scope from BRIEF.md}
     
     **Added:** {component list, comma-separated}
     **Modified:** {component list, or "—"}
     **Patterns:** {patterns established, comma-separated, or "—"}
     **PR:** [{#number}]({url}) or "—"
     **Files:** {count} files touched → [manifest](./projects/{name}/codebase/MANIFEST.md)

Step 6: Phase transition output

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

If review identified brand-level issues (token values that don't work in context), note: "Some issues are brand-level — run /gsp-brand-refine to adjust tokens without re-running identity."