Agent Skills › hexclave/hexclave › first-screen-audit

first-screen-audit

GitHub

通过分析转化用户数据与竞品研究，审计落地页首屏（标题、副标题、CTA）的表现，识别转化弱点，并基于证据提出具体的文案和布局优化建议，以提升转化率。

apps/growth-agent/agent/skills/first-screen-audit/SKILL.md hexclave/hexclave

Trigger Scenarios

需要优化落地页首屏转化率分析首屏文案与目标用户匹配度

Install

npx skills add hexclave/hexclave --skill first-screen-audit -g -y

More Options

Non-standard path

npx skills add https://github.com/hexclave/hexclave/tree/dev/apps/growth-agent/agent/skills/first-screen-audit -g -y

Use without installing

npx skills use hexclave/hexclave@first-screen-audit

指定 Agent (Claude Code)

npx skills add hexclave/hexclave --skill first-screen-audit -a claude-code -g -y

安装 repo 全部 skill

npx skills add hexclave/hexclave --all -g -y

预览 repo 内 skill

npx skills add hexclave/hexclave --list

SKILL.md

Frontmatter

{
    "name": "first-screen-audit",
    "description": "Execute the \"First-screen audit\" growth analysis topic — audit the landing page's first screen against what converting users actually respond to, and propose concrete copy\/layout changes."
}

First-screen audit

Goal: make the first screen of the project's landing page convert better by grounding its headline, subheadline, and primary CTA in who actually signs up and activates.

Data to pull

get-context-bundle — product context, website URL, and any website-research findings (positioning, audience, competitor headlines) already saved in this run.
get-metrics — signup baseline, so recommendations reference the current conversion reality.
sql-query — where the data exists, answer:
- Which referrer/UTM sources produce users who activate (fire product events after signup) vs. bounce?
- What share of signups happens within one session of first visit (a proxy for how hard the first screen has to work)?
- Send the queries above as ONE batched sql-query call (it takes up to 10); they are independent, so splitting them across calls only adds model round trips.
If website-research findings describe the current hero (headline, CTA, social proof), use them; do not re-crawl.

What to produce

2-4 save-finding calls, kind above-the-fold: each one names a specific weakness of the current first screen (e.g. headline describes features while activating users come from a jobs-to-be-done search term) with the evidence behind it.
One save-artifact, kind copy_proposal, titled "Above-the-fold proposal": a markdown document containing, for each proposed change — current state, proposed replacement (exact copy, not directions), and the evidence for it. Include at least: headline, subheadline, primary CTA label, and one social-proof/credibility element.
Optionally one create-action-item (type custom) if a change is high-confidence and self-contained enough to hand straight to the team; reference the artifact id in the payload.

Quality bar

Proposed copy is written out verbatim and speaks the ICP's language as evidenced by the data (sources, search terms, activating segments) — never generic marketing filler.
Every recommendation cites its evidence; if the project lacks traffic data, say so in a finding and ground proposals in the website/competitor research instead, labeling them as such.
Do not propose A/B testing infrastructure or redesigns; stay within copy and first-screen structure.