Lp Review

Verify the target is LP files:

1. Check for .tsx files (LP/Docusaurus pages) and docusaurus.config.*
2. If ASO metadata detected (App Store description, keywords, subtitle): Tell user "This looks like ASO metadata. Please use /aso-review instead." and STOP.

Phase 1: Discover Target Files

1. Glob("**/*.tsx") -- LP and Docusaurus page components
2. Glob("docusaurus.config.*") -- SEO headTags and site metadata
3. MUST Read ALL discovered files -- Do not skip any. Full text is needed for context.

Phase 2: Build Shared Context

Assemble a context package for the review team:

1. Full target text -- All copy from discovered files
2. Role descriptions -- Each reviewer's focus area (see Phase 3 table)
3. Brand voice -- Reference /labee-llc-guide for tone and values

This context package is passed to each agent in Phase 3.

Phase 3: Form Review Team & Run Reviews

MUST call TeamCreate to create the review team. MUST spawn ALL 3 agents in ONE SendMessage block -- do NOT launch sequentially.

TeamCreate("lp-review-team")

Then send the shared context to all 3 agents in parallel:

Each agent receives:

• The full LP copy
• The checklist from references/_checklist-lp.md relevant to their role
• Instructions to output findings as actionable items with suggested fixes

Agent instructions:

messaging-reviewer (LEAD):

• Review persuasion flow: Hero -> Features -> CTA
• Check CTA hierarchy (primary vs secondary, above-the-fold)
• Evaluate audience resonance for each segment (developers, managers, remote workers)
• Verify brand voice matches Labee values
• Flag copy that sounds like a corporate brochure

naturalness-reviewer:

• Detect AI vocabulary and inflated phrasing
• Detect "translated from Japanese" patterns (subject omission, feature-first, keigo leaks)
• Check for comma-splice run-ons, forced rule-of-three, excessive hedging
• Suggest specific rewrites for every flagged item

seo-reviewer:

• Check page title, meta description, H1 for primary keyword
• Verify heading hierarchy (no skipped H levels)
• Check Open Graph and Twitter Card meta tags
• Review internal linking and alt text
• Evaluate keyword density (natural, not stuffed)

Phase 4: Cross-Review Discussion

MUST use SendMessage to relay findings between agents. Do NOT skip this phase.

1. Collect all 3 agents' findings
2. Send messaging-reviewer's findings to naturalness-reviewer and seo-reviewer
3. Send naturalness-reviewer's findings to messaging-reviewer
4. Send seo-reviewer's findings to messaging-reviewer
5. Each agent responds with:
   • Agreements or disagreements with other reviewers
   • Conflicts (e.g., SEO keyword vs natural phrasing) and proposed resolutions
   • Any items they missed that another reviewer caught

The messaging-reviewer (LEAD) makes final call on conflicts.

Phase 5: Synthesize Results

1. Collect final feedback from all agents after cross-review
2. Compile a unified report:

## LP Review Report

### Findings
- [item]: [issue] -> [suggested fix]

### Conflicts Resolved
- [conflict]: [resolution by lead reviewer]

3. MUST call TeamDelete("lp-review-team") after synthesizing results.
4. Present report to user and wait for approval before applying fixes.

Phase 6: Apply Fixes

Only after user approval:

1. Apply approved fixes using Edit/Write
2. Run pnpm run build to verify no build errors
3. If build fails, fix and re-run
4. Present summary of changes made

Good/Bad Examples

Hero Section

Bad:

A Comprehensive Task Management Solution Designed to Revolutionize Your Workflow

Good:

Get more done. Stress less.

Why: The bad version uses AI vocabulary ("comprehensive", "revolutionize"), inflated phrasing, and says nothing specific. The good version is direct and benefit-focused.

Feature Benefit

Bad:

Our task management functionality enables users to efficiently organize and prioritize their daily activities.

Good:

Drag tasks around. Done ones disappear. That's it.

Why: The bad version is abstract and formal ("functionality enables users to efficiently"). The good version shows the actual experience.

CTA

Bad:

Please feel free to try our service

Good:

Start free -- no credit card needed

Why: The bad version has Japanese politeness norms ("please feel free") that sound unnatural in English LP copy. The good version is direct with a trust signal.

Translated-from-Japanese Pattern

Bad:

Can manage tasks easily and check progress at a glance

Good:

You manage tasks and track progress -- all in one view

Why: The bad version omits the subject (natural in Japanese, incomplete in English). The good version adds "You" and flows naturally.

AI Pattern

Bad:

Our cutting-edge platform serves as a comprehensive solution that seamlessly integrates with your existing workflow, enabling teams to leverage powerful collaboration tools.

Good:

It connects to Slack, GitHub, and Jira. Your team sees everything in one place.

Why: The bad version stacks AI vocabulary ("cutting-edge", "serves as", "comprehensive", "seamlessly", "leverage"). The good version states concrete facts.

Feature-First vs Benefit-First

Bad:

Multi-language support function with automatic translation capability

Good:

Write once, read in any language. Auto-translated in 12 languages.

Why: The bad version leads with the feature (Japanese product copy pattern). The good version leads with the benefit and adds a specific number.

Anti-Patterns

1. Do NOT launch agents sequentially -- All 3 agents MUST be spawned in one message
2. Do NOT skip cross-review discussion -- Phase 4 is where conflicts get resolved
3. Do NOT merge results without discussion -- Raw concatenation misses contradictions
4. Do NOT forget TeamDelete -- Always clean up the review team after Phase 5
5. Do NOT review ASO metadata -- Redirect to /aso-review instead

Reference Files