How do I install the Os Web Performance skill?

Run npx skillmds@latest add iterate/os-web-performance in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Os Web Performance safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Os Web Performance?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Os Web Performance free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Os Web Performance?

iterate (@iterate) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Os Web Performance by iterate — Agent Skill

Name: Os Web Performance
Author: iterate

Os Web Performance

Diagnose and improve apps/os loading speed, Core Web Vitals, SSR boundaries, asset caching, bundle waterfalls, hydration, and browser-to-itx startup latency. Use when os.iterate.com feels slow, a route shows generic loading UI, Web Vitals regress, repeat visits revalidate assets, or a performance PR needs production-shaped proof.

iterate bb418aa 4 files · 16.1 KB Updated Sep 17, 2026

File contents

OS web performance

Make OS feel immediate without hiding latency, errors, or state divergence.

Targets

Treat these as p75 aspirations for supported desktop and mobile routes:

Metric	Target
TTFB	<= 400 ms
FCP	<= 1.0 s
LCP	<= 1.8 s
INP	<= 100 ms
CLS	<= 0.03

Fingerprint repeat loads should transfer zero asset bytes. A direct route hit should return meaningful stable chrome or content in its HTML whenever the data and component are server-safe.

Workflow

Read references/playbook.md. For TanStack route, loader, bundle, or hydration work, also read references/tanstack-start.md. Classify the delay before editing: document/server, asset cache, JS graph, hydration, itx read, live subscription, or optional heavy runtime.
Capture cold and warm direct navigations plus a client-side navigation. Use an isolated Playwriter (or Playwright) browser session, the product fixture, and production or a preview—not Lighthouse against an unrepresentative dev server as the only evidence.
Follow the repository's PostHog instructions before field queries. Load the CLI agent help, list/install a matching skill, then query field p75s by route, device, release, and cold/warm cohort.
Inspect the production build manifest and browser network initiators. Count direct preloads, Brotli bytes, requests before FCP/LCP, cache status, and duplicated/heavy entry modules.
Reproduce with a red user-facing test. For SSR work, assert against the document response body, then assert the hydrated UI is interactive and free of page/hydration errors.
Make the narrowest architectural change. Move ClientOnly around the actual browser resource, lazy-load closed UI, aggregate chatty reads, or defer WASM; do not convert a measured problem into an unexplained fallback.
Prove the focused test, relevant unit tests, build, typecheck, lint, format, and full test suite. Then deploy a preview and repeat the same browser trace.
Audit Cloudflare traces and logs for the exact preview window. Zero new unexplained warnings/errors and coherent state are release criteria.

Evidence to report

environment, route, UTC window, commit, browser/device, cache state;
cold/warm TTFB, FCP, LCP, INP, CLS, transfer bytes, and request count;
initial HTML content and hydration result;
bundle/preload and browser-to-itx waterfall changes;
Playwright checks, preview URL, trace/log audit, and any remaining budget gap.

Read references/tooling.md before adding a third-party performance skill, MCP server, scanner, or CI dependency.

---name: os-web-performancedescription: Diagnose and improve apps/os loading speed, Core Web Vitals, SSR boundaries, asset caching, bundle waterfalls, hydration, and browser-to-itx startup latency. Use when os.iterate.com feels slow, a route shows generic loading UI, Web Vitals regress, repeat visits revalidate assets, or a performance PR needs production-shaped proof.---# OS web performanceMake OS feel immediate without hiding latency, errors, or state divergence.## TargetsTreat these as p75 aspirations for supported desktop and mobile routes:| Metric | Target || ------ | --------: || TTFB | <= 400 ms || FCP | <= 1.0 s || LCP | <= 1.8 s || INP | <= 100 ms || CLS | <= 0.03 |Fingerprint repeat loads should transfer zero asset bytes. A direct route hitshould return meaningful stable chrome or content in its HTML whenever the dataand component are server-safe.## Workflow1. Read [references/playbook.md](references/playbook.md). For TanStack route, loader, bundle, or hydration work, also read [references/tanstack-start.md](references/tanstack-start.md). Classify the delay before editing: document/server, asset cache, JS graph, hydration, itx read, live subscription, or optional heavy runtime.2. Capture cold and warm direct navigations plus a client-side navigation. Use an isolated Playwriter (or Playwright) browser session, the product fixture, and production or a preview—not Lighthouse against an unrepresentative dev server as the only evidence.3. Follow the repository's PostHog instructions before field queries. Load the CLI agent help, list/install a matching skill, then query field p75s by route, device, release, and cold/warm cohort.4. Inspect the production build manifest and browser network initiators. Count direct preloads, Brotli bytes, requests before FCP/LCP, cache status, and duplicated/heavy entry modules.5. Reproduce with a red user-facing test. For SSR work, assert against the document response body, then assert the hydrated UI is interactive and free of page/hydration errors.6. Make the narrowest architectural change. Move `ClientOnly` around the actual browser resource, lazy-load closed UI, aggregate chatty reads, or defer WASM; do not convert a measured problem into an unexplained fallback.7. Prove the focused test, relevant unit tests, build, typecheck, lint, format, and full test suite. Then deploy a preview and repeat the same browser trace.8. Audit Cloudflare traces and logs for the exact preview window. Zero new unexplained warnings/errors and coherent state are release criteria.## Evidence to report- environment, route, UTC window, commit, browser/device, cache state;- cold/warm TTFB, FCP, LCP, INP, CLS, transfer bytes, and request count;- initial HTML content and hydration result;- bundle/preload and browser-to-itx waterfall changes;- Playwright checks, preview URL, trace/log audit, and any remaining budget gap.Read [references/tooling.md](references/tooling.md) before adding a third-partyperformance skill, MCP server, scanner, or CI dependency.

Frequently asked questions