Every team has a deployment pipeline. Almost none have an optimization gate that verifies what actually lands on the wire before a search engine, an LLM crawler, or a mobile user touches it.
The /audit skill is the final gate in the Wanessa Labs development loop. It tests live production URLs against Google PageSpeed Insights benchmarks, Core Web Vitals runtime signals, Google Search Console parity, and agentic readiness—scoring every property on a rigorous 100-point scale.
Why Optimization Must Be The Final Step
Optimization cannot happen in the abstract while writing components. A feature can have clean React code and passing unit tests while simultaneously failing five critical production realities:
- The Font CDN Tax: Referencing external font CDNs (
fonts.googleapis.com) blocks the critical rendering path for 500–750ms on mobile 4G connections. - Non-Composited Animations: Animating properties like
box-shadow,width, ormargintriggers continuous paint and layout recalculations on the CPU, causing visible jank and degrading Cumulative Layout Shift (CLS). - Forced Synchronous Reflows: Interrogating geometry properties (
offsetWidth,getBoundingClientRect()) immediately after modifying DOM nodes forces synchronous layout recalculations during hydration. - The SPA Soft-404 Trap: Standard single-page applications configure wildcard rewrites (
/* /index.html 200) so direct navigation works. This causes broken URLs to return HTTP 200 with an empty shell, polluting search engine indexes with duplicate phantom content. - Redirect Friction: Missing trailing slashes on static internal links trigger unnecessary 301/308 redirect roundtrips on Cloudflare Pages edge servers.
The /audit skill was designed to catch every one of these regressions before declaring a build complete.
The Three Audit Pillars
The skill executes through an automated evaluation harness (audit-agentic.mjs) that inspects both static build artifacts and live network responses.
1. Core Web Vitals & Runtime Performance
- Self-Hosted Font Architecture: Zero external font CDNs. All typography is bundled locally via
@fontsourcewithfont-display: swapand preloaded.woff2subsets. - GPU-Composited Motion: Strict enforcement of composite-only animations. Pulsing status dots and indicators use
transform: scale()andopacityexclusively, running on the GPU without triggering browser paint invalidation. - Zero-Reflow Hydration: Dynamic layout measurements are refactored to CSS containers,
ResizeObserver, or deferred until idle viarequestAnimationFrame. - Image Hygiene: Explicit
widthandheightdimensions declared on all<img>elements to lock layout boxes before fetch, alongside full.webp/.avifconversion.
2. Search Engine & GSC Parity
- Authentic 404 Status Codes: Soft-404 wildcard traps are replaced with prerendered static route shells for all valid client paths, paired with an authentic
404.htmlthat returns a true HTTP 404 header. - Clean Trailing Slashes: Internal links, fallback navigations, and XML sitemaps strictly enforce trailing slash parity (
/route/), eliminating 301/308 redirect penalties. - Canonical Domain Consolidation: Validation that apex and
wwwhostnames consolidate cleanly without diluting domain equity.
3. Agentic & Robot Discovery
- Structured AI Manifests: Validates the presence of
/llms.txtand comprehensive/llms-full.txtfast paths with clean markdown link health. - Alternate Link Discovery: Injects
<link rel="alternate" type="text/markdown" href="/llms.txt">directly into<head>. - Token Thrift & Semantic Landmarks: Audits the text-to-code efficiency, pruning bloated inline SVGs, script boilerplate, and enforcing semantic landmarks (
<main>,<article>).
Proof of Work: 23-Project Full Fleet Audit
To validate the skill at scale, /audit was run across all 23 Cloudflare Pages projects deployed under Wanessa Labs.
| Fleet Metric | Before Fleet Audit | After Fleet Remediation |
|---|---|---|
| Projects Scoring ≥ 90/100 | 18 / 23 | 23 / 23 (100%) |
| Soft-404 Traps Detected | 6 projects | 0 projects |
| Non-Composited Animations | 8 instances | 0 instances |
| Average Parallel Fleet Audit Time | — | 2.4 seconds |
The audit engine runs asynchronously across the entire 23-domain matrix in under 2.5 seconds, providing immediate feedback on whether any commit broke performance or crawlability.
The Discipline
A project is not complete when it compiles. A project is complete when it serves clean semantics, renders with 60fps composited hardware acceleration, responds with authentic HTTP status codes, and provides fast paths for both humans and AI agents. The /audit skill turns that discipline into an automated standard.