renatocaliari/landing-page-evaluator

Landing Page Evaluator & Builder

This skill runs a single unified workflow, whether the user is evaluating existing copy or building one from scratch: gather whatever context exists, actively surface gaps and contradictions instead of silently filling them, and produce a three-part deliverable — Questions, Evaluation, Wireframe — as one interactive HTML artifact. As the user answers questions in follow-up turns, regenerate the same artifact with updated state.

Always read references/frameworks.md (the 5 categories) and references/anti-ai-tone.md (the cross-cutting checklist) before evaluating or generating any copy — they are the basis for judgment in this skill, not optional background material. Read references/artifact-guide.md before building the HTML deliverable.

The 5 categories (summary — full detail in references/frameworks.md)

1. SaaS / technical tool — Honest PAS + FAB. Rational decision.
2. High-involvement relational service (coaching, consulting, mentorship) — StoryBrand + BAB. Client is the hero, you're the guide.
3. Therapeutic practice / personal care / vulnerable audience — clarity + safety, not aggressive persuasion.
4. Physical product / experience / e-commerce / lifestyle — Sensory AIDA.
5. Open-source project / developer tool — Problem → Approach → Evidence → How to try it. No hype.

---

Core principle: surface gaps and contradictions, never silently resolve them

This is the most important behavior in this skill. Understanding context well enough to notice what's missing or inconsistent matters more than producing a fast, polished-looking result. A confident-sounding landing page built on invented assumptions is worse than a shorter one with the gaps clearly marked.

Gaps — information needed to write a specific section that hasn't been provided, for example:

• No stated audience, or audience described too broadly to write to
• No clear desired action / final CTA
• No tone-of-voice reference (existing site, prior writing, voice sample)
• No real proof (testimonial, number, case) when the category's framework calls for proof
• Category is genuinely ambiguous between two of the 5 (e.g. a technical product that's sometimes sold to developers, sometimes to enterprise buyers — those are different categories with different frameworks)
• Language of the final page is unclear or plausibly different from the language of the conversation

Contradictions — signals that conflict with each other, for example:

• Audience described as vulnerable/overwhelmed but the user is asking for urgency-heavy, high-pressure copy
• Product described as a peer-to-peer developer tool but the brief asks for heavy sales rhetoric and lead-capture CTAs
• Existing copy claims something ("used by thousands of teams") that contradicts other context the user gave (e.g. "this just launched last week")
• Tone samples provided don't match the tone the user says they want

Rule: if something is a gap or contradiction that would change a section's content or the overall structure, it becomes a question in the Questions tab — do not guess and move on. The only exception is a genuinely trivial stylistic choice where any reasonable default works and doesn't change structure or meaning (e.g. minor phrasing); in that case, make the call and state the assumption in the Evaluation tab instead of blocking on it.

Language of the page: default inference is the language the user is writing the conversation in. Treat this as a question needing confirmation only if it's ambiguous or contradicted by context (e.g. the brief mentions an international audience while the user writes in Portuguese). Otherwise, state the assumed language as a normal assumption in the Evaluation tab and don't hold up the whole page over it.

---

Workflow

Step 1 — Intake

Take whatever the user provides: pasted/uploaded existing copy, or a description/brief of an offer that doesn't exist as copy yet. Both are valid starting points for the same workflow.

Step 2 — Analyze

• Identify the most likely category (or a hybrid of two, if genuinely mixed — e.g. a project that's both a technical SaaS product and an open-source developer tool depending on audience) and note the reasoning.
• Run the anti-AI-tone checklist and the category's framework-fit check against whatever exists (existing copy, or the brief as given).
• Actively look for gaps and contradictions per the section above — this is not a passive step; assume there are gaps unless the brief is unusually complete, and check for them deliberately category by category (audience, goal/CTA, tone reference, proof, category fit, language).

Step 3 — Produce the three-part artifact

Always produce all three parts together, even when one is minimal:

• Questions — every gap and contradiction found, phrased as a direct question the user can answer in their next chat message, ordered with structure-changing questions (e.g. category ambiguity) first.
• Evaluation — assessment of everything submitted so far: strengths, the highest-impact issues (existing copy) or what's solid vs. assumed (new brief), scored against frameworks.md and anti-ai-tone.md. State assumptions explicitly rather than presenting them as settled.
• Wireframe — reflects only confirmed/provided content. Sections with enough real information render with full real copy. Sections that depend on an open question render as a clearly marked pending block instead of invented content (see artifact-guide.md for the exact rule). Never invent testimonials, numbers, or client names to fill a gap — always mark it pending instead.

Build this as a single three-tab HTML file per references/artifact-guide.md, save to /mnt/user-data/outputs/<slug>-landing-page.html, and present it via present_files.

Step 4 — Iteration loop

When the user answers one or more open questions in a later message:

• Update the category/language inference if the answer changes it.
• Re-run the Step 2 analysis with the new information.
• Regenerate the same three-tab file (overwrite it), moving resolved points out of Questions and into Evaluation/Wireframe, and expanding wireframe sections that are now unblocked.
• Re-present the file. Don't re-ask questions that were already answered.

Optional final step — markdown handoff

Once the Questions tab is empty (or the user explicitly says the remaining open questions don't matter to them) and the Wireframe tab is essentially complete, offer to produce a clean markdown copy document with the final content, for handoff to design/dev. Only produce this on request — it's not part of the default deliverable.

Rule on invented proof

This skill never invents numbers, testimonials, client names, or statistics. When a framework calls for proof and none was provided, that is always a gap — it goes in the Questions tab and shows as pending in the Wireframe tab, never as fabricated content presented as real.