Multi-agent site review panel. Dispatches senior-dev, frontend, and backend agents in parallel to review the current project from a functional perspective — not just bugs, but architecture, UX, performance, and improvement opportunities.

Unlike /site-audit (finds and fixes bugs), /site-review is a strategic review that asks: "Is this site as good as it could be? What would a senior team recommend?"

REQUIRES OPUS MODEL. Architectural judgment requires Opus-level reasoning.

Arguments

$ARGUMENTS = project directory path (default: current directory)
--quick = Skip backend agent, 2 agents only (~5 min)
--deep = Add architecture agent as 4th reviewer (~15 min)

Shared Review Context

Before dispatching agents, the orchestrator builds a shared context file that all agents read.

Pre-Flight (Orchestrator)

Identify the project from pwd + git remote get-url origin
Read project CLAUDE.md if it exists
Read the project's handoff (from superpowers repo) if one exists
Identify tech stack from package.json / project structure
Check if a live URL exists (from ~/Projects/site-to-repo-map.json)
Write _review/CONTEXT.md:

PROJECT: [name]
REPO: [github url]
LIVE_URL: [url or N/A]
TECH_STACK: [React/Next.js/Vue/etc.]
CSS_FRAMEWORK: [Tailwind/CSS modules/etc.]
BACKEND: [Cloudflare Workers/Express/None/etc.]
LAST_COMMIT: [date + message]
HANDOFF_STATUS: [summary of in-progress/blocked items, or "clean"]

Phase 1: Parallel Agent Dispatch

Launch 3 agents simultaneously (2 if --quick, 4 if --deep). Each agent reads _review/CONTEXT.md first, then performs its domain-specific review. Each writes findings to its own file.

Agent 1: Senior Frontend Reviewer

Agent type: general-purpose Skills: senior-frontend, react-best-practices, frontend-design, impeccable-design, design-critique Output: _review/frontend_review.md

Dispatch prompt:

You are a senior frontend engineer reviewing a production website. Read _review/CONTEXT.md for project context. Read ~/.claude/skills/senior-frontend/SKILL.md, ~/.claude/skills/react-best-practices/SKILL.md, ~/.claude/skills/frontend-design/SKILL.md, ~/.claude/skills/impeccable-design/SKILL.md, and ~/.claude/skills/design-critique/SKILL.md.

Review the ENTIRE frontend codebase. For each area, note both problems and improvement opportunities:

1. Component Architecture

Component composition: are components too large? Too granular? Right abstractions?

State management: prop drilling? Global state overuse? Missing context providers?

Code reuse: duplicated patterns that should be shared components?

File organization: intuitive structure? Easy to find things?

2. User Experience

All user states handled? (empty, loading, error, success, partial)

Navigation flow: can users get lost? Dead ends?

Mobile responsiveness: tested at 375px, 768px, 1280px?

Accessibility: ARIA labels, keyboard nav, color contrast?

Performance feel: perceived speed, loading indicators, skeleton screens?

3. Visual Design Quality

Run the AI Slop Detection checklist from impeccable-design — count fingerprints (0-10 scale)

Run Nielsen's Heuristics scoring from design-critique — score each of 10 heuristics 0-4 (/40 total)

Run cognitive load assessment from design-critique — 8-item checklist

Select 2-3 relevant personas from design-critique and walk through the primary user flow

Typography consistency and hierarchy (consult impeccable-design/reference/typography.md)

Spacing and alignment consistency (consult impeccable-design/reference/spatial-design.md)

Color usage — intentional or random? (consult impeccable-design/reference/color-and-contrast.md)

Does the design feel professional or amateur?

What one visual change would have the biggest impact?

4. Performance

Bundle size concerns (large dependencies, barrel imports)

Render performance (unnecessary re-renders, missing memoization)

Image optimization (formats, lazy loading, sizing)

Code splitting and lazy loading opportunities

5. Frontend Best Practices

TypeScript usage: strict enough? any abuse?

Error boundaries present?

SEO: meta tags, semantic HTML, structured data?

Console errors or warnings in dev mode?

For each finding, rate it:

P0 — Broken, users are affected

P1 — Significant quality issue

P2 — Improvement opportunity

P3 — Nice to have

End with: TOP 3 FRONTEND IMPROVEMENTS — the changes that would most improve this site for users.

Write everything to _review/frontend_review.md.

Agent 2: Senior Backend Reviewer

Agent type: general-purpose Skills: senior-backend, senior-dev-mindset Output: _review/backend_review.md

Skip this agent if no backend exists (no API routes, no functions/, no server files). Write _review/backend_review.md with "N/A — no backend" and continue.

Dispatch prompt:

You are a senior backend engineer reviewing a production API. Read _review/CONTEXT.md for project context. Read ~/.claude/skills/senior-backend/SKILL.md and ~/.claude/skills/senior-dev-mindset/SKILL.md.

Review ALL backend code. For each area, note both problems and improvement opportunities:

1. API Design

RESTful conventions followed? Consistent naming?

Response format consistent across endpoints?

Pagination on list endpoints?

Versioning strategy?

Are there missing endpoints that the frontend needs?

2. Security

Authentication on protected routes?

Input validation and sanitization?

SQL/NoSQL injection vectors?

CORS configuration?

Rate limiting on sensitive endpoints?

Secrets management (hardcoded keys?)

3. Performance

N+1 query patterns?

Missing database indexes?

Caching opportunities (frequently read, rarely written data)?

Connection pooling?

Unbounded queries (no LIMIT)?

4. Reliability

Error handling: try/catch with proper HTTP status codes?

Graceful degradation when external services fail?

Logging: errors logged or silently swallowed?

Timeout handling on external API calls?

5. Data Layer

Schema design appropriate?

Migrations in place?

Data validation at the boundary?

Backup/recovery considerations?

Rate each finding P0-P3. End with: TOP 3 BACKEND IMPROVEMENTS.

Write everything to _review/backend_review.md.

Agent 3: Senior Dev / Full-Stack Reviewer

Agent type: general-purpose Skills: senior-dev-mindset, senior-architect Output: _review/fullstack_review.md

Dispatch prompt:

You are a senior full-stack developer reviewing a production site holistically. Read _review/CONTEXT.md for project context. Read ~/.claude/skills/senior-dev-mindset/SKILL.md and ~/.claude/skills/senior-architect/SKILL.md.

Review the project as a WHOLE PRODUCT, not just code. Think about what makes this site valuable to its users:

1. Product Completeness

Does the site deliver on its core promise?

What features are half-built or stub?

What's the most obvious missing feature a user would expect?

Are there dead pages or unreachable features?

2. Integration Quality

Frontend-backend contract: do they match? Mismatched types? Stale endpoints?

Data flow: can you trace user action → API call → database → response → UI update?

Auth flow: login → session → protected routes → logout — all solid?

Third-party integrations: healthy? Error handled? Fallbacks?

3. Developer Experience

Can a new developer understand this codebase in 30 minutes?

Build/dev/test scripts work reliably?

Consistent patterns or every file does it differently?

Tech debt that slows down future development?

4. Deployment & Operations

Build passes clean?

Environment config: dev/staging/prod properly separated?

Monitoring: would you know if the site went down?

Version management: can you roll back?

5. Competitive Edge

What makes this site different from alternatives?

What would make a user choose this over competitors?

What's the #1 thing that would make this site 10x more valuable?

Rate each finding P0-P3. End with: TOP 3 PRODUCT IMPROVEMENTS — the changes that would most increase this site's value.

Write everything to _review/fullstack_review.md.

Agent 4 (--deep only): Architecture Reviewer

Agent type: general-purpose Skills: senior-architect Output: _review/architecture_review.md

Dispatch prompt:

You are a senior architect reviewing system design. Read _review/CONTEXT.md. Read ~/.claude/skills/senior-architect/SKILL.md. Map the entire system architecture (services, data flow, dependencies, deployment). Identify: over-engineering, under-engineering, scalability risks, tech stack fit. Write to _review/architecture_review.md.

Phase 2: Synthesis (Orchestrator)

After all agents complete, the orchestrator reads ALL review files and produces a unified report.

Read _review/frontend_review.md, _review/backend_review.md, _review/fullstack_review.md (and _review/architecture_review.md if --deep)
Deduplicate findings — same issue flagged by multiple agents gets ONE entry with "[Flagged by: Frontend + Backend]"
Merge all P0-P3 items into a single ranked list
Identify cross-cutting themes — patterns that appear across multiple reviews
Build the final recommendation list

Phase 3: Present

SITE REVIEW — [Project Name]
==============================
Generated: [date]
Live site: [URL or N/A]
Tech stack: [stack]
Reviewers: Senior Frontend + Senior Backend + Senior Dev [+ Architect]

DESIGN SCORES
-------------
Heuristics: [N]/40 — [Rating]  |  AI Slop: [N]/10 fingerprints  |  Cognitive Load: [Low/Moderate/High]
Weakest heuristic: [name] ([score])  |  Personas tested: [names]

EXECUTIVE SUMMARY
-----------------
[3-4 sentences: overall health, biggest strengths, biggest gaps]

REVIEW PANEL FINDINGS
---------------------

### P0 — Critical (fix immediately)
| # | Issue | Domain | Impact | Effort |
|---|-------|--------|--------|--------|
| 1 | ... | Frontend/Backend/Product | ... | ... |

### P1 — Significant (fix this sprint)
| # | Issue | Domain | Impact | Effort |
|---|-------|--------|--------|--------|

### P2 — Improvement Opportunities
| # | Suggestion | Domain | Impact | Effort |
|---|------------|--------|--------|--------|

### P3 — Nice to Have
[Bulleted list]

CROSS-CUTTING THEMES
---------------------
[Patterns that multiple reviewers flagged — these are systemic, not isolated]

TOP 5 IMPROVEMENTS (RANKED)
----------------------------
The review panel's consensus on what would most improve this site:

#1. [Title]
    Why: [reasoning from multiple reviewers]
    Impact: [what users would notice]
    Effort: [session estimate]

#2. ...
#3. ...
#4. ...
#5. ...

WHAT'S WORKING WELL
--------------------
[Strengths the reviewers noted — what to preserve and build on]

SUGGESTED SESSION PLAN
----------------------
If you start working now, here's the optimal order:
1. [First task — highest value or unblocks others]
2. [Second task]
3. [Third task]

Phase 4: Clean Up

# Keep _review/ for reference — don't delete
echo "Review files saved in _review/"

Rules

All agents run in parallel. The whole point is multiple perspectives simultaneously.
Not just bugs — improvements. Each agent must suggest what would make the site BETTER, not just what's broken.
READ-ONLY by default. This command reviews and recommends. It does NOT make changes. The user decides what to act on.
Rate everything P0-P3. Consistent severity across all reviewers.
Deduplicate in synthesis. Same issue from 3 reviewers = 1 entry, not 3.
End with a session plan. The user wants to know what to do next.
Preserve _review/ directory. The individual review files are valuable context for the agent that does the actual work later.
Skip backend agent if no backend exists. Don't waste tokens reviewing what isn't there.

Unlike /site-audit (finds and fixes bugs), /site-review is a strategic review that asks: "Is this site as good as it could be? What would a senior team recommend?"

REQUIRES OPUS MODEL. Architectural judgment requires Opus-level reasoning.

Arguments

$ARGUMENTS = project directory path (default: current directory)
--quick = Skip backend agent, 2 agents only (~5 min)
--deep = Add architecture agent as 4th reviewer (~15 min)

Shared Review Context

Before dispatching agents, the orchestrator builds a shared context file that all agents read.

Pre-Flight (Orchestrator)

Identify the project from pwd + git remote get-url origin
Read project CLAUDE.md if it exists
Read the project's handoff (from superpowers repo) if one exists
Identify tech stack from package.json / project structure
Check if a live URL exists (from ~/Projects/site-to-repo-map.json)
Write _review/CONTEXT.md:

PROJECT: [name]
REPO: [github url]
LIVE_URL: [url or N/A]
TECH_STACK: [React/Next.js/Vue/etc.]
CSS_FRAMEWORK: [Tailwind/CSS modules/etc.]
BACKEND: [Cloudflare Workers/Express/None/etc.]
LAST_COMMIT: [date + message]
HANDOFF_STATUS: [summary of in-progress/blocked items, or "clean"]

Phase 1: Parallel Agent Dispatch

Launch 3 agents simultaneously (2 if --quick, 4 if --deep). Each agent reads _review/CONTEXT.md first, then performs its domain-specific review. Each writes findings to its own file.

Agent 1: Senior Frontend Reviewer

Agent type: general-purpose Skills: senior-frontend, react-best-practices, frontend-design, impeccable-design, design-critique Output: _review/frontend_review.md

Dispatch prompt:

You are a senior frontend engineer reviewing a production website. Read _review/CONTEXT.md for project context. Read ~/.claude/skills/senior-frontend/SKILL.md, ~/.claude/skills/react-best-practices/SKILL.md, ~/.claude/skills/frontend-design/SKILL.md, ~/.claude/skills/impeccable-design/SKILL.md, and ~/.claude/skills/design-critique/SKILL.md.

Review the ENTIRE frontend codebase. For each area, note both problems and improvement opportunities:

1. Component Architecture

Component composition: are components too large? Too granular? Right abstractions?

State management: prop drilling? Global state overuse? Missing context providers?

Code reuse: duplicated patterns that should be shared components?

File organization: intuitive structure? Easy to find things?

2. User Experience

All user states handled? (empty, loading, error, success, partial)

Navigation flow: can users get lost? Dead ends?

Mobile responsiveness: tested at 375px, 768px, 1280px?

Accessibility: ARIA labels, keyboard nav, color contrast?

Performance feel: perceived speed, loading indicators, skeleton screens?

3. Visual Design Quality

Run the AI Slop Detection checklist from impeccable-design — count fingerprints (0-10 scale)

Run Nielsen's Heuristics scoring from design-critique — score each of 10 heuristics 0-4 (/40 total)

Run cognitive load assessment from design-critique — 8-item checklist

Select 2-3 relevant personas from design-critique and walk through the primary user flow

Typography consistency and hierarchy (consult impeccable-design/reference/typography.md)

Spacing and alignment consistency (consult impeccable-design/reference/spatial-design.md)

Color usage — intentional or random? (consult impeccable-design/reference/color-and-contrast.md)

Does the design feel professional or amateur?

What one visual change would have the biggest impact?

4. Performance

Bundle size concerns (large dependencies, barrel imports)

Render performance (unnecessary re-renders, missing memoization)

Image optimization (formats, lazy loading, sizing)

Code splitting and lazy loading opportunities

5. Frontend Best Practices

TypeScript usage: strict enough? any abuse?

Error boundaries present?

SEO: meta tags, semantic HTML, structured data?

Console errors or warnings in dev mode?

For each finding, rate it:

P0 — Broken, users are affected

P1 — Significant quality issue

P2 — Improvement opportunity

P3 — Nice to have

End with: TOP 3 FRONTEND IMPROVEMENTS — the changes that would most improve this site for users.

Write everything to _review/frontend_review.md.

Agent 2: Senior Backend Reviewer

Agent type: general-purpose Skills: senior-backend, senior-dev-mindset Output: _review/backend_review.md

Skip this agent if no backend exists (no API routes, no functions/, no server files). Write _review/backend_review.md with "N/A — no backend" and continue.

Dispatch prompt:

You are a senior backend engineer reviewing a production API. Read _review/CONTEXT.md for project context. Read ~/.claude/skills/senior-backend/SKILL.md and ~/.claude/skills/senior-dev-mindset/SKILL.md.

Review ALL backend code. For each area, note both problems and improvement opportunities:

1. API Design

RESTful conventions followed? Consistent naming?

Response format consistent across endpoints?

Pagination on list endpoints?

Versioning strategy?

Are there missing endpoints that the frontend needs?

2. Security

Authentication on protected routes?

Input validation and sanitization?

SQL/NoSQL injection vectors?

CORS configuration?

Rate limiting on sensitive endpoints?

Secrets management (hardcoded keys?)

3. Performance

N+1 query patterns?

Missing database indexes?

Caching opportunities (frequently read, rarely written data)?

Connection pooling?

Unbounded queries (no LIMIT)?

4. Reliability

Error handling: try/catch with proper HTTP status codes?

Graceful degradation when external services fail?

Logging: errors logged or silently swallowed?

Timeout handling on external API calls?

5. Data Layer

Schema design appropriate?

Migrations in place?

Data validation at the boundary?

Backup/recovery considerations?

Rate each finding P0-P3. End with: TOP 3 BACKEND IMPROVEMENTS.

Write everything to _review/backend_review.md.

Agent 3: Senior Dev / Full-Stack Reviewer

Agent type: general-purpose Skills: senior-dev-mindset, senior-architect Output: _review/fullstack_review.md

Dispatch prompt:

You are a senior full-stack developer reviewing a production site holistically. Read _review/CONTEXT.md for project context. Read ~/.claude/skills/senior-dev-mindset/SKILL.md and ~/.claude/skills/senior-architect/SKILL.md.

Review the project as a WHOLE PRODUCT, not just code. Think about what makes this site valuable to its users:

1. Product Completeness

Does the site deliver on its core promise?

What features are half-built or stub?

What's the most obvious missing feature a user would expect?

Are there dead pages or unreachable features?

2. Integration Quality

Frontend-backend contract: do they match? Mismatched types? Stale endpoints?

Data flow: can you trace user action → API call → database → response → UI update?

Auth flow: login → session → protected routes → logout — all solid?

Third-party integrations: healthy? Error handled? Fallbacks?

3. Developer Experience

Can a new developer understand this codebase in 30 minutes?

Build/dev/test scripts work reliably?

Consistent patterns or every file does it differently?

Tech debt that slows down future development?

4. Deployment & Operations

Build passes clean?

Environment config: dev/staging/prod properly separated?

Monitoring: would you know if the site went down?

Version management: can you roll back?

5. Competitive Edge

What makes this site different from alternatives?

What would make a user choose this over competitors?

What's the #1 thing that would make this site 10x more valuable?

Rate each finding P0-P3. End with: TOP 3 PRODUCT IMPROVEMENTS — the changes that would most increase this site's value.

Write everything to _review/fullstack_review.md.

Agent 4 (--deep only): Architecture Reviewer

Agent type: general-purpose Skills: senior-architect Output: _review/architecture_review.md

Dispatch prompt:

You are a senior architect reviewing system design. Read _review/CONTEXT.md. Read ~/.claude/skills/senior-architect/SKILL.md. Map the entire system architecture (services, data flow, dependencies, deployment). Identify: over-engineering, under-engineering, scalability risks, tech stack fit. Write to _review/architecture_review.md.

Phase 2: Synthesis (Orchestrator)

After all agents complete, the orchestrator reads ALL review files and produces a unified report.

Read _review/frontend_review.md, _review/backend_review.md, _review/fullstack_review.md (and _review/architecture_review.md if --deep)
Deduplicate findings — same issue flagged by multiple agents gets ONE entry with "[Flagged by: Frontend + Backend]"
Merge all P0-P3 items into a single ranked list
Identify cross-cutting themes — patterns that appear across multiple reviews
Build the final recommendation list

Phase 3: Present

SITE REVIEW — [Project Name]
==============================
Generated: [date]
Live site: [URL or N/A]
Tech stack: [stack]
Reviewers: Senior Frontend + Senior Backend + Senior Dev [+ Architect]

DESIGN SCORES
-------------
Heuristics: [N]/40 — [Rating]  |  AI Slop: [N]/10 fingerprints  |  Cognitive Load: [Low/Moderate/High]
Weakest heuristic: [name] ([score])  |  Personas tested: [names]

EXECUTIVE SUMMARY
-----------------
[3-4 sentences: overall health, biggest strengths, biggest gaps]

REVIEW PANEL FINDINGS
---------------------

### P0 — Critical (fix immediately)
| # | Issue | Domain | Impact | Effort |
|---|-------|--------|--------|--------|
| 1 | ... | Frontend/Backend/Product | ... | ... |

### P1 — Significant (fix this sprint)
| # | Issue | Domain | Impact | Effort |
|---|-------|--------|--------|--------|

### P2 — Improvement Opportunities
| # | Suggestion | Domain | Impact | Effort |
|---|------------|--------|--------|--------|

### P3 — Nice to Have
[Bulleted list]

CROSS-CUTTING THEMES
---------------------
[Patterns that multiple reviewers flagged — these are systemic, not isolated]

TOP 5 IMPROVEMENTS (RANKED)
----------------------------
The review panel's consensus on what would most improve this site:

#1. [Title]
    Why: [reasoning from multiple reviewers]
    Impact: [what users would notice]
    Effort: [session estimate]

#2. ...
#3. ...
#4. ...
#5. ...

WHAT'S WORKING WELL
--------------------
[Strengths the reviewers noted — what to preserve and build on]

SUGGESTED SESSION PLAN
----------------------
If you start working now, here's the optimal order:
1. [First task — highest value or unblocks others]
2. [Second task]
3. [Third task]

Phase 4: Clean Up

# Keep _review/ for reference — don't delete
echo "Review files saved in _review/"

Rules

All agents run in parallel. The whole point is multiple perspectives simultaneously.
Not just bugs — improvements. Each agent must suggest what would make the site BETTER, not just what's broken.
READ-ONLY by default. This command reviews and recommends. It does NOT make changes. The user decides what to act on.
Rate everything P0-P3. Consistent severity across all reviewers.
Deduplicate in synthesis. Same issue from 3 reviewers = 1 entry, not 3.
End with a session plan. The user wants to know what to do next.
Preserve _review/ directory. The individual review files are valuable context for the agent that does the actual work later.
Skip backend agent if no backend exists. Don't waste tokens reviewing what isn't there.

Adoption

nhouseholder/site-review

$ install --global

Security Scan Results

SKILL.md

Arguments

Shared Review Context

Pre-Flight (Orchestrator)

Phase 1: Parallel Agent Dispatch

Agent 1: Senior Frontend Reviewer

Agent 2: Senior Backend Reviewer

Agent 3: Senior Dev / Full-Stack Reviewer

Agent 4 (--deep only): Architecture Reviewer

Phase 2: Synthesis (Orchestrator)

Phase 3: Present

Phase 4: Clean Up

Rules

Related Skills

nhouseholder/compactor

nhouseholder/webapp-testing

nhouseholder/using-ultraplan

nhouseholder/ui-ux-pro-max

nhouseholder/site-review

$ install --global

Security Scan Results

SKILL.md

Arguments

Shared Review Context

Pre-Flight (Orchestrator)

Phase 1: Parallel Agent Dispatch

Agent 1: Senior Frontend Reviewer

Agent 2: Senior Backend Reviewer

Agent 3: Senior Dev / Full-Stack Reviewer

Agent 4 (--deep only): Architecture Reviewer

Phase 2: Synthesis (Orchestrator)

Phase 3: Present

Phase 4: Clean Up

Rules

Related Skills

nhouseholder/compactor

nhouseholder/webapp-testing

nhouseholder/using-ultraplan

nhouseholder/ui-ux-pro-max