- Home
- Skills
- Testing & Debugging
- Page Speed Loop: measure cold vs warm, fix, guard
More screenshots
Works with the AI Agents you already use
Page Speed Loop: measure cold vs warm, fix, guard
Measure-first web speed loop: cold/warm p50/p95 bench, SWR shelf, progress banner, JS budget, pinned tests.
Free
Page Speed Loop: measure cold vs warm, fix, guard
Example session with this skill installed
The dashboard takes 40 seconds on the first visit of the morning. Find out why and fix it without making it worse later.
- Read your context and instructions
- Compiled the page speed loop
$ bench_url.py http://127.0.0.1:8765/page -n 20 --cold-cmd <reset>
GET /page (local http.server) n=20 cold = cold server + connection
ttfb p50 ttfb p95 total p50 total p95 bytes
cold 123.3 126.4 123.4 126.4 2048
warm 0.2 0.7 0.2 0.7 2048
Connects securely to your tools. The creator never sees your data.
About this skill
Most speed work fails the same way: someone changes code, the page feels faster, and nothing was measured. Page Speed Loop gives your agent a loop that starts and ends with numbers, so a fix is proven and then guarded.
What it does
- Times a URL cold (new connection, optional server reset) and warm (kept-alive), N times, and prints ttfb and total p50/p95.
- Teaches the agent to count upstream calls, find serial awaits, duplicate downloads and caches smaller than the working set.
- Covers the waiting experience: a progress endpoint with a banner, the difference between "source refused" and "updating" marks, skeletons shaped like the content.
- Describes a stale-while-revalidate shelf in IndexedDB that paints last answers, then replaces them, with the three rules that keep it safe (owner, build, age).
- Precompresses hashed assets at build time, checks a JS budget and flags case-only chunk renames.
- Shows how to pin call counts and route lists in tests so a fixed page stays fixed.
What you get
| File | Purpose |
|---|---|
| SKILL.md | rules, loop, symptom table, traps |
| references/ (5 files) | measuring, progress and skeletons, SWR shelf, assets and budget, pinned tests |
| scripts/bench_url.py | cold/warm timing with p50/p95 |
| scripts/check_budget.py | gzip size budget, case-collision check |
| scripts/precompress.py | .gz/.br siblings |
| templates/before-after-report.md | report for every change |
Who it's for
Developers who own a web app that got slow and want an agent that measures before it edits: full-stack and front-end engineers, small teams on throttled or free-tier hosting.
Scope
Covered: method, measurement, request-count diagnosis, loading feedback, client shelf, asset compression, bundle budget, regression guards. Not covered: SSR frameworks, image and CDN tuning, database query tuning, profilers. Python 3.9+ standard library only; brotli is optional. Thresholds are examples from one app on a single throttled vCPU.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Testing & Debugging.
Brotli output and a real front-end build were not tested; the self-tests use a local http.server and a generated fixture. Thresholds are examples from one app.
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
1 use
Times used to date
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 1 day ago
- Free with an Agensi account
- Software
Python 3.9+A running app to benchmark
Git
- Optional
- brotli Python module
Scripts use the standard library only. brotli is needed only to measure brotli-compressed sizes.
Permissions
bench_url.py only contacts the URL you pass it. Nothing else is fetched; no credentials are read.