@amandamalavedev/precheck
v0.1.0
Published
Check your website before you launch — test speed, accessibility, and security headers, generate essential SEO files, and get guidance on what to fix. Works with your terminal or AI coding assistant.
Maintainers
Readme
Website Precheck
Check your website before you launch — all in one toolkit.
Test speed, accessibility, and security headers, generate essential SEO files, and get guidance on what to fix. Works with your terminal or AI coding assistant. Free and open, from Pacific AI Labs.
| Use it as… | For whom |
|---|---|
| A command — precheck <check> <url> | Anyone. No AI, no account. |
| Claude skills — drop into ~/.claude/skills/ | Claude Code users |
| An MCP server — any agent calls the checks live | Any MCP-capable AI (incl. ChatGPT) |
| A Custom GPT or Gemini Gem | ChatGPT and Gemini users |
The same checks underneath (lib/); five ways to run them. You don't need to use Claude, or any AI
at all.
Run the checks from the command line (works for everyone)
Install it straight from this repo (Node 18+):
npm install -g github:amandamalavedev/website-precheck…or run any check without installing: npx github:amandamalavedev/website-precheck headers https://example.com.
It isn't on the npm registry yet. Don't run
npx precheck— that name isn't ours, and if someone else ever publishes a package calledprecheck, that command would run their code instead. When this is published it will be under the scoped name@amandamalavedev/precheck.
precheck report https://example.com --out ./report # all six checks → one shareable HTML/JSON/MD report
precheck headers https://example.com # security headers: present vs missing
precheck cookies https://example.com # every cookie the site sets, flagged for missing Secure/SameSite
precheck privacy https://example.com # third-party trackers, and whether a privacy policy is linked
precheck lighthouse https://example.com # performance (median of N runs) + LCP breakdown + what to fix
precheck a11y https://example.com # accessibility (WCAG 2.1 AA) via axe-core
precheck schema https://example.com # SEO: structured data (JSON-LD) found + how complete it is
precheck media https://example.com # local <video> files over a size threshold
precheck sitemap --dir ./build --base https://example.com # generate sitemap.xml + robots.txt + llms.txtThe six checks in report: Security (headers), Governance (cookies), Privacy, Speed (Lighthouse),
Accessibility, SEO (structured data).
headers,cookies,privacy,schema,mediaandsitemapare pure Node — no browser, no external install.lighthouseusesnpx lighthouse@12under the hood (needs Chrome).- The IP/SSRF classifier is a vendored copy of
ipaddr.js(MIT) bundled with the scripts — nothing extra to install.
These are tools for sites you own or are authorized to test. They harden against an agent being tricked into a local-command or internal-network attack (no shell; private/metadata addresses refused in every canonical IPv4/IPv6 form; writes confined to the target directory), but that is defense in depth, not a guarantee — see SECURITY.md for the threat model and the residual risks (DNS rebinding, Lighthouse browser-driven redirects) that it does not fully mitigate.
a11yneeds Chrome +puppeteer-core(npm install --no-save puppeteer-core); it bypasses the page's CSP for the scan so it works even on well-hardened sites.
No model, no account, no telemetry — it just runs the check and prints what to fix.
Use them as Claude skills
The command tells you what's wrong; the Claude skill teaches Claude how to fix it — the full method, not just a score. Install all five (or keep only the ones you want):
git clone https://github.com/amandamalavedev/website-precheck.git
cd website-precheck
for s in site-hardening-and-speed site-admin-panel seo-geo-schema analytics-and-search-console accessibility-launch-readiness; do
cp -r "$s" ~/.claude/skills/
doneTo update later: git pull in the clone and run the same loop again — it copies the new version of
every file over the old one.
| Skill | What it does |
|---|---|
| site-hardening-and-speed | Outsider-view security audit, Lighthouse performance, US/local privacy + cookie compliance. |
| site-admin-panel | The gated back office: edit content, upload & replace pictures (WebP + EXIF-stripped), view analytics, manage users. |
| seo-geo-schema | Meta/Open Graph, JSON-LD schema, sitemap, robots, llms.txt. (Public sites only.) |
| analytics-and-search-console | Cookieless analytics (no consent banner) + Google Search Console. (Public sites only.) |
| accessibility-launch-readiness | WCAG 2.1 AA pass + the pre-launch QA checklist. |
Then just ask Claude to harden a site, run Lighthouse, add schema, or do a launch review — the matching skill loads itself.
Use them from any agent (MCP), ChatGPT or Gemini
- MCP server (
mcp/) — exposes the checks as tools any MCP-capable agent can call live. One command adds it to Claude Code; see mcp/README.md. - Custom GPT kit (
gpt/) — the instructions to paste into ChatGPT's builder so a Custom GPT uses the same method. - Gemini Gem kit (
gemini/) — the same, for Gemini's Gem manager. - The GPT and the Gem give the method only; they don't run the real checks unless you connect them to the command or the MCP server (each kit explains how).
Two ideas run through all of it
- Two site profiles. Every build starts by asking: Public (indexed, wants SEO and analytics)
or Private (gated,
noindex, no public analytics)? Some skills only apply to public sites — running them on a private one is actively wrong. - Verify, don't guess. Pasted audit reports are often stale or hallucinated. These skills check the real code and the live site, and never call something done until it's verified against production.
License
MIT — see LICENSE. Use them, fork them, build on them.
More free skills coming from Pacific AI Labs — guardrails for AI-written code.
Not legal advice. The privacy/cookie and accessibility guidance is solid engineering; for real legal exposure a lawyer reviews the privacy notice, and automated accessibility checks catch only part of the issues — do the manual pass too.
