Unlighthouse Site Audit
Run local Lighthouse audits across many site URLs and aggregate performance, accessibility, best-practices, and SEO results.
The skill uses a local Unlighthouse CLI, a stated url_safety preflight, and subprocess timeout management, with no API key required, indicating relatively limited data exposure and controlled execution. However, the skill file does not explicitly define request scope, user confirmation, redirect/external effects, report cleanup, or recovery behavior, so 8 points are deducted.
Routing, device selection, route caps, output files, and cross-skill delegation are described consistently. However, the supplied material does not show the scripts or document diagnostics for invalid URLs, missing dependencies, timeout exceptions, or Unlighthouse failures; static evidence is therefore limited and subject to the static score cap.
The target scenarios are clear: site-wide Lighthouse audits with mobile/desktop selection, route limits, and optional persisted output. Boundaries, complete input semantics, Chinese-language support, and npm dependency reachability from mainland-China networks are not documented, so 5 points are deducted.
The skill includes a name, version, compatibility requirement, prerequisites, routing table, output fields, and delegation guidance. It lacks examples specific to this skill, troubleshooting, explicit limitations, maintenance ownership, an update path, and a changelog, so 6 points are deducted. Repository-level MIT licensing and version/maintenance metadata provide partial support.
The documented multi-page audit, aggregate metrics, per-route JSON, and HTML report are potentially directly usable. However, no skill-specific verified output or key-path execution evidence is supplied; report completeness, operational cost, and comparative benefit remain largely asserted, so the score is reduced under the static calibration cap.
The skill specifies auditable output filenames, fields, and script entry points, while repository CI and tests provide indirect evidence. The supplied tests do not directly cover the Unlighthouse skill's key path, and no reproducible skill-specific run or independent corroboration is provided, so points are deducted.
- The audit issues requests to multiple URLs on the target site; confirm scope, load, and authorization before running, especially for third-party or untrusted sites.
- Verify that extensions/unlighthouse/install.sh, scripts/unlighthouse_run.py, Node 18+, and the unlighthouse npm package are available in the target environment; the supplied material lacks skill-specific tests or failure examples.
- Reports use a temporary directory by default and can be persisted to a chosen directory; handling, access control, and cleanup for sensitive page data should be clarified.
- The core dependency is obtained through npm, but the documentation provides no mainland-China mirror, offline-install, or fallback procedure.
What does this skill do, and when should you use it?
This is one skill within the claude-seo repository, using the MIT-licensed Unlighthouse CLI to run Lighthouse across multiple URLs. It fits teams that have outgrown PageSpeed Insights' free quota, need local or restricted-environment measurements, or want a quick site-wide regression check after deployment. The default route cap is 200, with JSON and HTML reports produced alongside parsed aggregate results. It is a multi-page Lighthouse wrapper, not the full claude-seo SEO audit collection.
After extensions/unlighthouse/install.sh pre-warms the dependencies, it invokes Unlighthouse through scripts/unlighthouse_run.py; performs a URL-safety pre-flight and subprocess timeout management; reads ci-result.json from the Unlighthouse output directory; and returns parsed aggregate data. The output includes performance, accessibility, best-practices, and SEO scores, with performance reported as the median across audited routes, plus per-route details and JSON/HTML reports.
- An SEO or engineering team needs multi-page lab audits after PageSpeed Insights' free quota is insufficient.
- A development team wants a fast, site-wide performance regression check after a deployment.
- A CI pipeline or restricted environment requires local Core Web Vitals lab measurement.
- A site owner wants to audit only 50 routes and persist the reports in a chosen directory.
What are this skill's strengths and limitations?
- Runs locally without an API key or PageSpeed Insights quota usage.
- Audits multiple routes consistently and provides both aggregate scores and route-level details.
- Supports mobile, desktop, route-cap, and output-directory options.
- Useful for CI, restricted environments, and post-deployment regression checks.
- The SKILL.md does not specify an exact unlighthouse npm version or installation parameters.
- Its results are Lighthouse lab data; the skill does not provide single-URL CrUX field data.
- The default cap is 200 routes, so larger sites require an explicit cap decision and may take longer to run.
- It depends on Node.js, Unlighthouse, shell access, and local filesystem access; auditing target URLs generally requires network access.
How do you install this skill?
From the claude-seo repository, run: bash extensions/unlighthouse/install.sh. Node.js 18+ must be on PATH, and the unlighthouse npm package is required. No API key is needed.
How do you use this skill?
Run /seo unlighthouse <url> for a mobile audit of up to 200 routes, with JSON and HTML reports written to a temporary directory. Use /seo unlighthouse <url> --device desktop for the desktop form factor, or /seo unlighthouse <url> --max-routes 50 --output-dir ./reports to set a route cap and persist the reports. For single-URL field data, use seo-google psi or seo-google crux; for LCP subpart decomposition on slow pages, use the scripts/lcp_subparts.py workflow.
How does this skill compare with similar options?
Compared with calling PageSpeed Insights separately for each URL, this skill runs Lighthouse locally across multiple routes without consuming PageSpeed Insights' free API quota. Compared with the single-URL seo-google psi / seo-google crux workflows, it emphasizes site-wide route coverage and aggregation rather than field data for one page.