- Home
- Skills
- Frontend & Web Apps
- Web Performance Audit
Works with the AI Agents you already use
Web Performance Audit
A performance claim without a number is an opinion.
$6
Web Performance Audit
Example session with this skill installed
Our dashboard feels slow. Audit performance for 3 page classes: /dashboard (app shell), /search (list), /product/[slug] (detail). Next.js 14 + Postgres, staging is available, and we have a 30-day RUM export. The checkout path matters most.
- Read your context and instructions
- Compiled the web performance audit
- Generated the image
Summary
Modes: C (RUM p75) + B (lab) + A (static). Baseline p75 field: LCP 4.1 s / INP 380 ms / CLS 0.18 / TTFB 1.6 s on /product - all Poor or Needs Improvement. Coverage: 74 of 82 applicable checks answered.
Top wins (estimated)
| # | Finding | Evidence | Est. win | Effort |
|---|---|---|---|---|
| 1 | Product hero image lazy-loaded and missing from initial HTML - it IS the LCP | web/app/product/[slug]/page.tsx:41 | LCP -1.8 s | S |
| 2 | N+1 on /search: 1 + 24 queries per page (reviews per product) | api/search/route.ts:78 | TTFB -700 ms | M |
| 3 | 640 KB uncompressed JS in the dashboard initial bundle; 1 charting lib loads eagerly | web/app/dashboard/layout.tsx:12 | INP -150 ms, LCP -600 ms | M |
| 4 | No cache headers on product API; every view re-renders server-side | api/product/route.ts:19 | TTFB -500 ms (repeat views) | S |
| 5 | Consent banner shifts layout (no reserved slot) | web/components/Consent.tsx:8 | CLS -0.12 | S |
Budgets (budgets.json)
product: LCP p75 <= 2.5 s, TTFB <= 800 ms, JS <= 200 KB compressed, requests <= 50 - wired to CI with a 10% regression gate
Next steps
- Eager-load the LCP image with fetchpriority=high, verify with the re-measure protocol in the report
- Batch reviews with a single join; add
pg_stat_statementsto catch the next N+1
Full report: perf/REPORT.md - raw runs: perf/measurements.json
page.tsx
PNG · 1536×1024
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
About this skill
Measure -> attribute -> fix -> re-measure. A performance claim without a number is an opinion. This skill audits web apps for real user-perceived speed, prioritizing by user impact on the money path, not by Lighthouse score.
What you get
perf/REPORT.md- baselines with measurement conditions, attributed findings (P0/P1/P2) with estimated wins, and a re-measurement protocolbudgets.json- CI-ready transfer budgets (HTML/JS/CSS/images/fonts/requests) per page classmeasurements.json+coverage.csv+evidence/- raw runs, per-check statuses and traces/HARs/EXPLAIN output- A 20-line chat summary: p75 status per metric, top 5 wins with estimated gains, report path
How it works
- Baseline - field first (CrUX/RUM p75 by page class and device), then lab (median of 3-5 runs, cold and warm cache); no baseline means no improvement claims
- Attribution - each failing metric gets a culprit: the LCP element and its delivery chain (TTFB -> discoverability -> transfer -> render), the INP interaction and its long tasks, the CLS shift sources, the TTFB breakdown via Server-Timing
- Static sweep - frontend, backend/database, caching and CDN checks, each marked PASS / FAIL / N/A / UNKNOWN with evidence
- Stack pack - framework-specific causes and fixes for Next.js, Node, Django, Laravel, FastAPI and WordPress
- Quantify & prioritize - every finding carries a number (ms / KB / queries), an estimated win, an effort, and a verification step; ranked by impact on the critical path
Guardrails
Non-destructive: no load, stress or spike testing - one sequential client, lab runs capped at 5 per URL, passive probes at 3 per URL and max 2 req/s. No production writes; EXPLAIN ANALYZE runs on staging or a replica only (plain EXPLAIN or slow-query logs on production). Paid APIs and credits need your approval. Measurement conditions are reported with every number; estimates are never presented as measurements.
Requirements
- Repo access to the web project (static mode needs nothing else)
curlfor the optional passive probe script; Lighthouse for lab mode- Optional: a staging URL, and a CrUX/RUM export for field truth
Known limitations
- Static mode (repo only) yields structural causes and estimates, no timings
- Field truth needs CrUX/RUM data; lab numbers are proxies, never treated as user experience
- Third-party removals are proposed as deferral/facade/self-hosting, never as deletions - analytics and consent tools have product and legal consequences
- The audit proposes patches and budgets but does not change your app unless asked
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Frontend & Web Apps.
No load/stress testing. No production writes. Max 5 lab runs per URL. Paid API keys (CrUX/WebPageTest) require approval. Static analysis is estimation-only without timing data.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed today
- Software
- lighthouse
Node.jsnextsource-map-explorervite-bundle-visualizernotcurlRepo access to the web project
- Optional
- Staging URL for lab measurementsCrUXorRUM export for field dataLighthouse for lab runs
Permissions
File Scopes
Reads the repo and build output, writes the report to ./perf/. Network access is only the passive probe script: exactly 3 sequential GET/HEAD requests per URL, max 2 req/s, plus optional local Lighthouse runs. EXPLAIN ANALYZE executes only on staging or a replica; production gets plain EXPLAIN or read-only slow-query logs. No load testing, no writes, ever.