# Web perf > Use when diagnosing slow loads or layout shift from browser traces and network evidence. Diagnose page-load and interaction performance using browser traces, network evidence, and source code. Use for site-speed, Lighthouse, Core Web Vitals, or layout-shift investigations; a general accessibility audit is separate. ## Facts - Type: Agent skill - Discipline: Engineering - Author: Matt Silverlock (https://github.com/elithrar) - Upstream: https://github.com/elithrar/dotfiles/blob/main/.agents/skills/web-perf/SKILL.md - Source file: https://raw.githubusercontent.com/elithrar/dotfiles/main/.agents/skills/web-perf/SKILL.md (2884 bytes, sha256:3e7ee7394181a1d9bec0d554b70f30c205f2acc93a95077950eff81497c88bf5) - Licence: MIT - Upstream copy last changed: 2026-09-17 - Status: listed, not yet ruled on by the curator ## Use it I want to use the Web perf skill by Matt Silverlock. This is what it does, in the author's own words: Diagnose page-load and interaction performance using browser traces, network evidence, and source code. Use for site-speed, Lighthouse, Core Web Vitals, or layout-shift investigations; a general accessibility audit is separate. Download the whole folder at https://github.com/elithrar/dotfiles/tree/main/.agents/skills/web-perf, SKILL.md and every file beside it, into your skills directory: ~/.claude/skills/web-perf/ for Claude Code, ~/.agents/skills/web-perf/ for Codex, ~/.cursor/skills/web-perf/ for Cursor. Get it from the author's repo, not from startupskills.dev. Then tell me the path you installed it to and how to invoke it. Nothing here is vendored: install from the author's repo, not from this site. A verbatim copy of the author's file follows for reading, under the licence reproduced after it; the current file is in the author's repo. ## Links - Page: https://startupskills.dev/elithrar/web-perf - Author: https://startupskills.dev/elithrar - Index: https://startupskills.dev/llms.txt ## The file itself --- name: web-perf description: Diagnose page-load and interaction performance using browser traces, network evidence, and source code. Use for site-speed, Lighthouse, Core Web Vitals, or layout-shift investigations; a general accessibility audit is separate. --- # Web performance investigation Identify the target page, reported symptom, and relevant device conditions. For audit requests, report findings. For optimization or fix requests, implement the supported changes and validate their effect within the requested scope. ## Choose the evidence path Inspect the available browser and profiling tools before calling them. Prefer Chrome DevTools traces when available; read [references/devtools.md](references/devtools.md) for trace and network investigation. Use the installed tool schemas rather than assuming example parameters match the current version. If that MCP server is unavailable, use existing trace or Lighthouse artifacts, another available profiler, or relevant source code. Complete the analysis those sources support and state what could not be measured. Missing profiling tools do not justify stopping source inspection or forcing an MCP installation. ## Diagnose Record the page/build, viewport, device or emulation, network/CPU throttling, cache state, and interactions relevant to the result. Reloading alone does not establish a cold cache. Keep mobile and desktop observations distinct. Separate field Core Web Vitals from laboratory measurements. A load-only trace does not establish INP; TBT is a diagnostic lab metric, not a substitute INP measurement. Use the current metric definitions when assigning ratings, and label missing measurements as unavailable. Follow the bottleneck from trace insight to request, element, or source path. Inspect the relevant LCP phases, layout-shift participants, long tasks, late resource discovery, or cache headers. Confirm ownership and use before recommending removal, especially for third-party scripts or resources used only in later interactions. Zero estimated savings is not evidence of a material problem, but one trace cannot prove a resource is never useful. ## Validate and report Prioritize findings by observed user impact and confidence. Distinguish measured values, tool estimates, and unmeasured hypotheses; avoid promising additive savings from overlapping insights. For a fix, compare the affected behavior under equivalent conditions. Repeat measurements when variance or the size of the claimed improvement warrants it. Broaden testing only for an unresolved regression risk or a project gate, then finish. Report the measured symptom, its supported cause, the concrete fix or recommendation, and the relevant verification limit. Add a metric table when several measurements help comparison. Do not expand a performance request into a mandatory accessibility or whole-codebase audit. ## Licence of the file above The file above is https://raw.githubusercontent.com/elithrar/dotfiles/main/.agents/skills/web-perf/SKILL.md, by Matt Silverlock, reproduced without changes. - Source: https://github.com/elithrar/dotfiles/blob/main/.agents/skills/web-perf/SKILL.md - Copyright: Copyright (c) 2014 Matt Silverlock - Licence: MIT, https://raw.githubusercontent.com/elithrar/dotfiles/main/LICENSE The repo's licence file, verbatim from https://raw.githubusercontent.com/elithrar/dotfiles/main/LICENSE: The MIT License (MIT) Copyright (c) 2014 Matt Silverlock Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.