尚未收集仓库信息。
尚未检测到更新。GitHub 发布版本后会显示在这里。
Claude skill: CRO audit of any web page, backed by 1,800+ real A/B test results from LiftArchive
翻译准备中
正在加载完整 README。
查看完整 README ↗A Claude skill that audits any web page for conversions, backed by real A/B test results.
It checks the page, matches what it sees against tests that ran on the same kind of page, and tells you what to fix first.
Every suggestion comes with the test that backs it: the result, what changed, and who ran it.
It works for online stores, SaaS, lead-gen and service sites, and apps.
The data comes from LiftArchive, a free database of 1,800+ real A/B tests, each credited to the person or team that ran it.
It also works on local pages and code (localhost, components, templates) by reading the code.
Claude Code (plugin, includes the MCP server):
/plugin marketplace add gabeleee/liftarchive-skill
/plugin install liftarchive@liftarchive
Claude Code (skill only):
git clone https://github.com/gabeleee/liftarchive-skill
cp -R liftarchive-skill/skills/liftarchive ~/.claude/skills/
Claude.ai / Claude desktop: zip the skills/liftarchive folder and upload it under Settings → Capabilities → Skills.
/liftarchive audit https://example.com/pricing
Or just ask: "What should I A/B test on this page?"
The skill calls LiftArchive's public, read-only MCP server at https://liftarchive.com/mcp/.
If the MCP tools aren't connected, it uses scripts/la.py, a small Python client with no dependencies that calls the same server.
No API key or account is needed.
You can also connect the MCP server to ChatGPT, Claude or any MCP client on its own. See liftarchive.com/ai/.
The skill treats the audited page as untrusted data.
It won't follow instructions hidden in a page, run commands a page asks for, or read secrets, and it tells you if a page contains instructions aimed at AI tools.
The only network calls it makes are to the page being audited and to liftarchive.com.
Run the same audit in your browser, with a phone screenshot: liftarchive.com/cro-audit/.
MIT