lp-optimization（LP 优化）

Landing Page Conversion Optimization — Orchestrator

*Communication — Horizontal. Coordinates specialized audit agents to diagnose conversion blockers and produce prioritized, evidence-based fix recommendations.*

Core Question: "What's stopping visitors from converting?"

Philosophy

The frameworks here (PAS, 4-U, social proof hierarchy) are evidence-backed defaults. They work in most situations. But landing pages serve different audiences, traffic sources, and goals. When your data or testing reveals a different optimal approach, follow the data. The principles matter more than the specific numbers. This orchestrator dispatches specialist agents for each concern, then a critic agent ensures every recommendation meets the evidence bar.

Critical Gates — Read First

1. Do NOT recommend fixes without evidence. Every finding must include: what was observed, which principle it violates, and a specific recommended fix (exact text, not "improve the headline").
2. Check message match BEFORE optimizing copy. A perfectly written headline that doesn't match the traffic source will still bounce visitors. Message match is the first conversion gate.
3. Form fields: every field >5 costs ~10% conversion. This is an evidence-backed default (Unbounce/HubSpot research). Exceptions exist for high-intent enterprise traffic — but the exception must be justified, not assumed.

Inputs Required

• Landing page URL or description of the page
• ICP research from .agents/mkt/icp-research.md (recommended — VoC language strengthens copy)
• Traffic source context (where visitors come from)

Output

• Optimization recommendations with specific copy/structure changes
• For new pages: complete page structure with copy

Quality Gate

Before delivering, the critic agent verifies:

• Headline scores >=3 out of 4 on the U's (Useful, Unique, Urgent, Ultra-specific)
• Message matches the traffic source (headline echoes the ad/link that brought them)
• One primary CTA per page (secondary CTAs don't compete)
• Trust signals appear within scroll-distance of every CTA
• Form has <=5 fields (or justified why more are needed)
• Social proof is from the last 12 months (older proof replaced or removed)
• Every audit finding includes: what was observed, which principle it violates, and a specific recommended fix
• Prioritized fix list has ICE scores with documented reasoning
• No vague recommendations ("improve the headline" — must be specific replacement text)

Chain Position

Horizontal — works with icp-research (audience data), content-create (content assets), copywriting (copy craft), experiment (test design) Re-run triggers: After major page redesigns, when conversion drops >10%, when traffic source mix changes significantly, or quarterly.

Skill Deference

• Need to write copy from scratch (not optimize existing)? → Use content-create — this skill optimizes existing pages, not blank-page writing.
• Need craft-quality headline rewrites or CTA copy? → Run copywriting for variation workflow and evaluation rubric.
• AI pattern cleanup needed? → Use humanize — this skill focuses on conversion mechanics, not voice/pattern editing.

Agent Manifest

Shared References (read by multiple agents)

• references/core-principles.md — Headlines, value props, CTAs, forms, message match, PAS
• references/social-proof-trust.md — Social proof hierarchy, biases, trust signals
• references/ux-design.md — Visual hierarchy, mobile optimization
• references/advanced-psychology.md — Headline formulas, close sequences, pricing, urgency
• references/testing-optimization.md — A/B testing, tracking, ICE framework
• references/implementation-checklist.md — Pre-launch checklists

Routing Logic

Classify the task, then follow the matching route.

Route A: Quick Audit

When: User wants a fast assessment of a specific concern (headline, CTA, or form) — not a full-page audit.

1. Pre-dispatch: Gather context (Step 0 below)
2. Dispatch 1-2 relevant agents:
   - Headline concern → hero-audit-agent
   - CTA/form concern → cta-audit-agent
   - Both → hero-audit-agent + cta-audit-agent in parallel
3. Dispatch: prioritization-agent (on selected agent outputs)
4. Dispatch: critic-agent
5. If critic returns FAIL → re-dispatch named agent(s) with feedback (max 2 cycles)
6. Deliver prioritized fix list

Route B: Full Audit

When: User wants a comprehensive landing page conversion audit.

1. Pre-dispatch: Gather context (Step 0 below)
2. LAYER 1 — Dispatch IN PARALLEL:
   - hero-audit-agent
   - trust-audit-agent
   - cta-audit-agent
   - ux-audit-agent
3. MERGE: Combine all Layer 1 outputs into unified audit findings
4. LAYER 2 — Dispatch SEQUENTIALLY:
   - message-match-agent (receives merged L1 output)
   - prioritization-agent (receives message-match output)
   - critic-agent (receives prioritization output)
5. If critic returns FAIL → re-dispatch named agent(s) with feedback (max 2 cycles)
6. Deliver final audit artifact

Route C: New Landing Page (Write Mode)

When: User wants to create a new landing page, not audit an existing one.

This is a write workflow, not an audit workflow. The agents are repurposed as planning advisors.

1. Pre-dispatch: Gather context (Step 0 below)
2. Read ICP research (if available) for VoC language and pain points
3. Define primary conversion goal (one per page)
4. Generate headline using hero-audit-agent's 4-U framework:
   - Generate 10+ headline variations using formula library
   - Score each with 4-U formula
   - Select top 3 for presentation
5. Structure body with PAS framework (from core-principles reference)
6. Plan social proof using trust-audit-agent's hierarchy:
   - Identify strongest proof available
   - Place above fold
7. Design form using cta-audit-agent's form rules:
   - Minimal fields (start with email only)
   - First-person CTA formula
   - Risk reversal placement
8. Plan UX using ux-audit-agent's checklist:
   - Remove navigation
   - Mobile-first layout
   - Trust signals near every CTA
9. Verify message match with traffic source(s)
10. Run through quality gate checklist
11. Deliver page structure with copy recommendations

Note: Route C uses the agent frameworks and reference files as planning tools but follows the sequential workflow rather than parallel dispatch, since the page doesn't exist yet to audit in parallel.

Step 0: Pre-Dispatch Context Gathering

Before dispatching any agent, the orchestrator gathers context that ALL agents will need.

Product Context Check

Check for .agents/product-context.md. If available, read for product details and accuracy. If missing, strongly recommend running icp-research first — this skill works without it but produces significantly better output with it. If .agents/product-context.md or .agents/mkt/icp-research.md date fields are older than 30 days, warn the user and recommend re-running icp-research before proceeding. This is a soft gate — proceed if the user confirms, but note "stale ICP data" in the artifact header.

Required Artifacts

None — can audit any page standalone.

Optional Artifacts

Pre-Writing Framework

Answer these questions before dispatching. Pass the answers to every agent as the pre-writing input:

1. What is the primary conversion goal? (signup, purchase, demo request, download)
2. Who is the target audience? What do they currently believe? What language do they use?
3. Where is the traffic coming from? (Ad, search, email, social, direct) — determines message match requirements.
4. What traffic source copy exists? (ad headlines, email subject lines, search meta descriptions) — needed for message-match verification.

If .agents/mkt/icp-research.md exists, pull VoC quotes and pain language for all agents.

Dispatch Protocol

How to spawn a sub-agent

For each agent dispatched below, use the Agent tool with a prompt constructed as follows:

1. Read the agent instruction file (e.g., agents/hero-audit-agent.md) — include its FULL content in the Agent prompt
2. Append the brief and pre-writing context after the instructions
3. Resolve file paths to absolute: replace relative paths with absolute paths rooted at this skill's directory. Example: if this skill is at /Users/you/skills/lp-optimization/, then references/core-principles.md becomes /Users/you/skills/lp-optimization/references/core-principles.md. Tell the agent: "Read the reference file at [absolute path] for domain knowledge."
4. Pass upstream artifacts by content, not path: the orchestrator reads .agents/product-context.md and .agents/mkt/icp-research.md FIRST, then includes relevant excerpts (VoC quotes, pain language, product details) in the pre-writing object. Sub-agents should NOT read artifact files directly — the orchestrator curates what they need.
5. If feedback exists (from a critic FAIL cycle), append it at the end of the prompt with the header "## Critic Feedback — Address Every Point"

Single-agent fallback

If multi-agent dispatch is unavailable (no Agent tool, single-agent runtime, or context constraints), execute each agent's instructions sequentially in-context:

• Layer 1: run each agent's domain instructions one at a time, writing each section
• Layer 2: apply each refinement pass to the full document in order
• Critic: self-evaluate using the critic-agent's rubric and quality gate

The output quality should be equivalent — the multi-agent pattern optimizes for parallelism and focus, not capability.

Layer 1: Parallel Audit Agents

Spawn the following agents IN PARALLEL (multiple Agent tool calls in a single message). For each agent, follow the Dispatch Protocol above.

For quick audit tasks (Route A): Dispatch only the relevant agent(s), not all four.

Merge Step

After all Layer 1 agents return, assemble their outputs into a unified audit document.

Section Order (full audit)

1. Hero Assessment — hero-audit-agent's headline, subheadline, and hero CTA evaluation
2. Trust & Social Proof Assessment — trust-audit-agent's proof inventory, gaps, and recommendations
3. CTA & Conversion Assessment — cta-audit-agent's CTA inventory, form audit, risk reversal, friction analysis
4. UX & Page Experience Assessment — ux-audit-agent's navigation, mobile, speed, cognitive load findings

Assembly Rules

The merge is deterministic assembly, not creative synthesis. Slot each agent's output into the unified document by ownership:

Conflict Resolution

• Each agent owns specific sections (table above). If two agents mention the same element (e.g., both hero-audit and cta-audit reference the hero CTA), keep the version from the section owner.
• If hero-audit's headline assessment contradicts ux-audit's visual hierarchy assessment, note both perspectives and let the prioritization agent resolve via ICE scoring.

Layer 2: Sequential Analysis

Dispatch these agents ONE AT A TIME, IN ORDER using the Dispatch Protocol above. Each receives the previous agent's full output as the upstream field.

message-match-agent → prioritization-agent → critic-agent

Each agent returns the full document with their analysis added + a change log.

Critic Gate

The critic agent returns one of two verdicts:

PASS

The audit meets all quality standards. Deliver the critic's approved output as the final artifact.

FAIL

The critic returns specific failures with:

• Which sections failed and why
• Specific fix instructions
• Which agent to re-dispatch

Rewrite loop:

1. Read the critic's failure report
2. Re-dispatch ONLY the named agent(s) with the critic's feedback attached as the feedback input
3. Run the modified output back through the critic
4. Maximum 2 rewrite cycles. After 2 failures, deliver the audit with the critic's annotations and flag to the user: "Audit scored [X] — manual review recommended on [specific sections]."

Artifact Template

When saving optimization artifacts, use this frontmatter:

---
skill: lp-optimization
version: 1
date: [today's date]
status: draft
---

On re-run: rename existing artifact to [name].v[N].md and create new with incremented version.

Worked Example — Full Audit (Route B)

Brief: Audit the Acme Analytics free trial page for conversion optimization. Audience: Engineering managers at 50-200 person companies, problem aware. Traffic: Google Ads ("real-time analytics dashboard"), LinkedIn ads (cold), and organic search.

Step 0: Pre-Dispatch

1. Conversion goal: Free trial signup
2. Audience: EMs who need real-time visibility but feel buried in tool sprawl
3. Traffic: Google Ads (warm, searched for solution), LinkedIn (cold, interrupted), organic
4. Traffic source copy: Google Ad headline: "Real-time analytics dashboard — 14-day free trial"

Layer 1: Parallel Dispatch

--> Hero audit agent returns: Headline "The Best Analytics Platform for Modern Teams" scores 1/4 on 4-U. Recommends: "See every metric your team tracks — one screen, updated every 5 minutes." (4/4). 5 variations provided. --> Trust audit agent returns: No testimonials above fold. Customer logos present but below fold. No recency dates on testimonials. Tier 1 proof missing entirely. Recommends: add strongest result quote near hero CTA, move 3 logos above fold, date-stamp all testimonials. --> CTA audit agent returns: Hero CTA "Get Started" (no first-person, no specificity). Form has 7 fields (name, email, company, role, size, phone, referral source). No risk reversal near any CTA. Recommends: "Start My Free Trial" + cut to 3 fields + "14-day free trial. No credit card." --> UX audit agent returns: Full site nav present (7 links). Mobile CTA below fold. Page load 3.8s due to uncompressed hero image. No sticky CTA on mobile.

Merge (orchestrator assembles)

Slots all Layer 1 outputs into unified audit: Hero Assessment -> Trust Assessment -> CTA Assessment -> UX Assessment.

Layer 2: Sequential Dispatch

--> Message match agent receives merged output + traffic source data. Finds: Google Ad promises "real-time" and "dashboard" — headline mentions neither. Ad says "14-day free trial" — page says "Get Started" with no trial mention. LinkedIn ad not checked ([BLOCKED: no ad copy provided]). Organic meta title partially matches. Scores: Google Ads = Mismatch, Organic = Partial Match. --> Prioritization agent receives message-match output. Consolidates all findings. Ranks by ICE:

1. Rewrite headline for message match (ICE: 9.3)
2. Cut form from 7 to 3 fields (ICE: 8.7)
3. Add risk reversal near hero CTA (ICE: 8.0)
4. Move testimonial above fold (ICE: 7.0)
5. Remove navigation links (ICE: 6.7)
6. Compress hero image for speed (ICE: 6.3)
7. Add sticky CTA on mobile (ICE: 6.0)

Dispatch: Critic Agent --> PASS

Scores: Completeness 5, Specificity 5, Evidence 4, Prioritization 5, Actionability 5. Average: 4.8/5. All quality gate items checked. Approved.

Final Artifact

---
skill: lp-optimization
version: 1
date: 2026-03-17
status: draft
---

# LP Audit — Acme Analytics Free Trial Page

## Headline Assessment

Current: "The Best Analytics Platform for Modern Teams"
4-U Score: 1/4 (Useful only — not Unique, not Urgent, not Ultra-specific)

Recommended: "See every metric your team tracks — one screen, updated every 5 minutes."
4-U Score: 4/4

## Message Match
Traffic source: Google Ads — "real-time analytics dashboard"
Current headline mentions neither "real-time" nor "dashboard" → broken match.
Fix: Echo "real-time dashboard" in headline.

## Social Proof Audit
- No testimonials above fold → add strongest result quote near hero CTA
- Customer logos present but below fold → move 3 logos above fold

## Form Audit
Current: 7 fields (name, email, company, role, size, phone, referral source)
Recommended: 3 fields (name, email, company) — collect rest via progressive profiling after signup
Expected impact: ~40% form completion improvement

## Priority Action Items
| # | Fix | Expected Impact | Effort | ICE |
|---|-----|----------------|--------|-----|
| 1 | Rewrite headline for message match | High | Low | 9.3 |
| 2 | Cut form to 3 fields | High | Low | 8.7 |
| 3 | Add risk reversal near hero CTA | High | Low | 8.0 |
| 4 | Move testimonial above fold | Medium | Low | 7.0 |
| 5 | Remove navigation links | Medium | Low | 6.7 |
| 6 | Compress hero image | Medium | Low | 6.3 |
| 7 | Add sticky CTA on mobile | Medium | Medium | 6.0 |

Anti-Patterns

Redesign without diagnosis — Jumping to "make it look better" without identifying what's actually blocking conversion. Diagnose before prescribing. INSTEAD: Run the full audit (Route B) to identify specific blockers, then fix in priority order.

Testing design before copy — A/B testing button colors or layouts when the headline doesn't pass the 4-U test. Copy is responsible for 80%+ of conversion impact. INSTEAD: Fix the words before the visuals. Follow the testing priority order: headlines > offers > CTAs > layout > forms.

Ignoring mobile experience — Optimizing for desktop when 60%+ of traffic is mobile. INSTEAD: Check thumb zone placement for CTAs, ensure forms are completable one-handed, verify load time on mobile networks, add sticky CTA for long pages.

Social proof without specificity — "Trusted by thousands of companies" is weaker than "Used by 3,247 teams including Stripe and Notion." INSTEAD: Every social proof element must include specific numbers, named companies, or measurable results.

Multiple competing CTAs — Primary CTA, secondary CTA, sidebar CTA, and exit-intent popup all fighting for attention. INSTEAD: One primary CTA per page — secondary CTAs must not visually compete. Audit the CTA hierarchy.

Vague audit recommendations — "Improve the headline" or "Add social proof" without specifics. INSTEAD: Every recommendation must include: what was observed, which principle it violates, and exact replacement text or action.

Skipping message match — Optimizing page copy without checking if it matches the traffic source. INSTEAD: Always verify message match first. A perfectly written headline that doesn't match the ad will still bounce visitors.

Agent Files

Sub-Agent Instructions (agents/)

• agents/hero-audit-agent.md — Headline, subheadline, hero CTA, above-the-fold
• agents/trust-audit-agent.md — Social proof, testimonials, credibility signals
• agents/cta-audit-agent.md — CTAs, risk reversal, form fields, conversion friction
• agents/ux-audit-agent.md — Navigation, mobile, speed, cognitive load
• agents/message-match-agent.md — Traffic source message continuity
• agents/prioritization-agent.md — ICE scoring, testing roadmap
• agents/critic-agent.md — Quality gate, PASS/FAIL, rewrite routing
• agents/_template.md — Reusable template for creating new agent files

Shared References (references/)

• references/core-principles.md — Headlines, value props, CTAs, forms, message match, PAS
• references/social-proof-trust.md — Social proof hierarchy, biases, trust signals
• references/ux-design.md — Visual hierarchy, mobile optimization
• references/advanced-psychology.md — Headline formulas, close sequences, pricing, urgency
• references/testing-optimization.md — A/B testing, tracking
• references/implementation-checklist.md — Pre-launch checklists

Sources: CopyHackers, CXL, Unbounce