celeroncoder/flame-graph-skills

Flame Audit is an agent skill that helps an AI coding assistant find the parts of a web page that actually slow it down, using the same thinking as a CPU flame graph: the widest piece of work is the most important thing to fix, and stack depth shows what caused what. It captures a real performance trace and network activity, then turns that raw data into a short ranked report instead of a generic checklist of best practices. When used, it measures a cold page load or a specific interaction before suggesting anything, then ranks the three to five biggest bottlenecks across main-thread work, the largest contentful paint, JavaScript bundle size, rendering, and network activity. It keeps a clear line between what it actually measured and what it is only guessing based on reading the codebase. After changes are made, it re-runs the same trace so any claimed improvement is backed by a new measurement rather than assumed. Every audit follows the same format: a plain-language verdict naming the dominant bottleneck, a ranked list of the widest measured problem areas with their cost and confidence, Core Web Vitals scores, a look at which network transfers are slowing things down, a prioritized fix plan tied to the evidence, and a before-and-after comparison once fixes are applied. The skill prefers to gather its data through Chrome DevTools, and falls back to Lighthouse, a profile the user supplies directly, or a clearly labeled static review if neither is available. It installs either through a skills registry command or by copying the skill's files into an agent's skills directory, and its guiding rules are to measure before recommending, fix the biggest cost first, avoid claiming savings the data does not support, and stop once a page is already performing well. It is released under the MIT license.