L’interface est localisée ; les présentations et documents des Skills sont traduits automatiquement par IA dans votre langue. Le code, les commandes, chemins et termes techniques restent en version originale, et le document de l’auteur fait foi.
Explorer les Skills
Design et création · SKILL.md

pitch-landing

pitch-landing

Use when asked to structure a landing page for positioning, plan a conversion-optimized page layout, or design a launch page. Examples: "landing page for product launch", "conversion-optimized layout for SaaS"

Independent GitHub snapshot · Automated classification · Original documentation governs

Documentation de la Skill

GitHub

name: pitch-landing description: | Use when asked to structure a landing page for positioning, plan a conversion-optimized page layout, or design a launch page. Examples: "landing page for product launch", "conversion-optimized layout for SaaS" allowed-tools: Read, Bash, Glob, Grep version: 0.6.6 author: tonone-ai hello@tonone.ai license: MIT

pitch-landing — Launch & Positioning Landing Page

Follow the output format defined in docs/output-kit.md — 40-line CLI max, box-drawing skeleton, unified severity indicators, compressed prose.

When to use

User needs a landing page structured around product positioning, launch messaging, or conversion for a specific audience.

Workflow

  1. Identify product type and positioning anchor from user request or brief
  2. Search landing page patterns:
    python3 -m pitch_agent.uiux search --domain landing --query "{product_type}" --limit 3
    
  3. Search product reasoning for audience + messaging context:
    python3 -m pitch_agent.uiux search --domain product --query "{product_type}" --limit 3
    
  4. Layer in positioning: CTA strategy, social proof placement, objection handling
  5. Output section order with conversion and messaging optimization

Output format

┌─ Launch Landing Page — {product_type} ──────────────────────────────┐
│ #  │ Section            │ Purpose                    │ CTA?          │
├────┼────────────────────┼────────────────────────────┼───────────────┤
│  1 │ {section_name}     │ {purpose}                  │ Primary CTA   │
│  2 │ {section_name}     │ {purpose}                  │ —             │
│  3 │ {section_name}     │ {purpose}                  │ Secondary CTA │
│  … │ …                  │ …                          │ …             │
└────┴────────────────────┴────────────────────────────┴───────────────┘

CTA strategy:          {cta_strategy}
Social proof:          {social_proof_placement}
Objection handling:    {objection_section}
Positioning anchor:    {positioning_anchor}

Anti-patterns

  • Never structure copy without a clear positioning anchor (who it's for + what makes it different)
  • Never add sections that don't serve conversion or objection handling
  • Never place social proof after the primary CTA — it should reinforce before the ask
  • Never launch without a single, unambiguous primary CTA per viewport

Delivery

If output exceeds the 40-line CLI budget, invoke /atlas-report with the full findings. The HTML report is the output. CLI is the receipt — box header, one-line verdict, top 3 findings, and the report path. Never dump analysis to CLI.

Skills enregistrées