Adonis0123/adonis-skills / skills/web-performance-audit
web-performance-audit
VerifiedRun a read-only, evidence-led runtime performance audit of a real web application, especially authenticated editors, dashboards, media tools, long-lived SPAs, and pages with many tabs or states. Use when the user asks to profile slowness, inspect every tab, use Chrome DevTools or Chrome Recorder, run React Scan, compare normal and throttled interactions, find memory or rendering risks, or produce a performance report and recommendations without changing code. Discover available browser capabilities, guide the user through official installation when an important capability is missing, route to validated fallbacks, isolate measurement pollution, and distinguish measured facts from screen observations, hypotheses, and unverified paths. Do not use for a Lighthouse-only SEO/accessibility audit, tool installation with no audit goal, or when the user already wants a specific optimization implemented.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 15.9 KB |
| agents/openai.yaml | config | 341 B |
| assets/demo-audit-ledger.json | config | 9.3 KB |
| assets/demo-audit-report.md | markdown | 10.4 KB |
| evals/benchmark-summary-2026-08-30.md | markdown | 2.0 KB |
| evals/evals.json | config | 8.1 KB |
| evals/files/ambiguous-metrics.json | config | 511 B |
| evals/files/dashboard-capabilities.json | config | 560 B |
| evals/files/video-editor-observations.json | config | 660 B |
| evals/host-acceptance-2026-08-30.md | markdown | 4.7 KB |
| evals/trigger-evals.json | config | 3.1 KB |
| references/capability-bootstrap.md | markdown | 10.0 KB |
| references/cross-agent-compatibility.md | markdown | 6.4 KB |
| references/ecosystem-comparison.md | markdown | 2.6 KB |
| references/measurement-playbook.md | markdown | 8.6 KB |
| references/report-contract.md | markdown | 8.2 KB |
| references/tool-routing.md | markdown | 9.9 KB |
| scripts/audit_ledger.py | script | 41.4 KB |
| scripts/test_audit_ledger.py | script | 16.9 KB |
| scripts/test_public_portability.py | script | 4.0 KB |
| showcase.yaml | config | 7.1 KB |
evals/evals.json
View on GitHub{
"skill_name": "web-performance-audit",
"evals": [
{
"id": 1,
"prompt": "请只读审计附件中的大型媒体工作区。媒体实际时长以附件观测为准;Media、Text、Elements、Library 以及每个嵌套 tab 都要看。要用 Chrome DevTools、Chrome Recorder 和 React Scan,但不要改代码。根据附件先给出可执行审计方案和报告骨架。",
"expected_output": "输出应保留精确媒体时长与表面名称,建立逐 tab 的冷开/热切/播放/节流矩阵,把 Recorder 和 React Scan 放进隔离对照,明确证据语义、建议复测和状态恢复。",
"files": ["evals/files/video-editor-observations.json"],
"assertions": [
"输出使用附件中的 47:30.00,而不是擅自改写成约 48 分钟",
"输出覆盖 Media、Text、Elements、Library 以及附件列出的嵌套 tab,并为未覆盖路径保留 blocked/skipped 状态",
"输出先做无 Recorder/React Scan 的干净基线,再做开启工具的等价动作对照",
"输出说明 Recorder 用于可重放旅程和生成 Performance trace,Computer Use 只为 Recorder/DevTools UI 或视觉控件提供操作与视觉证据",
"输出不会把单次 click duration 称为 INP,不会把 decoded surface estimate 称为 JS heap 或 GPU 常驻",
"每条拟定发现都有建议方向、等价复测、成功判据和停止条件,并包含最终状态恢复",
"输出不会用 3 个样本报告 p95;少量重复只报告 median 和 range,并明确不能据此声称统计置信度",
"React Scan 只用于浏览器 trace 已定位的最小高风险交互窗口,不要求对每个 tab 全量开启"
]
},
{
"id": 2,
"prompt": "审计一个已登录运营 dashboard 的所有 tab。当前只有能复用登录态的浏览器插件和 Computer Use,Chrome DevTools MCP 握手失败;不能安装新工具,也不能修改页面。请根据附件输出审计执行单,不要假装已经测到 trace。",
"expected_output": "输出应在缺少 DevTools MCP 时继续用可用能力做降级审计,严格限制视觉证据的结论,并明确哪些定量项仍为 UNVERIFIED。",
"files": ["evals/files/dashboard-capabilities.json"],
"assertions": [
"输出不会因为 Chrome DevTools MCP 不可用就停止整个审计,也不会声称已经获取 Performance trace",
"输出把 Browser/Computer Use 观察与 Event Timing、Long Task、heap 等未取得的定量证据分开",
"输出覆盖 Overview、Orders、Revenue、Campaigns 以及各自 cold/warm 和 10 次代表性往返",
"输出把后台 polling 作为噪声/潜在成本单独记录,而不是自动归因给 tab click",
"输出给出恢复初始 Overview、日期范围和无临时 instrumentation 的收尾要求"
]
},
{
"id": 3,
"prompt": "这个公开营销落地页只需要跑一次 Lighthouse,检查 SEO、可访问性和首屏 Core Web Vitals,不涉及登录、复杂交互、长会话、tab 或运行时卡顿。请直接做最合适的审计。",
"expected_output": "输出应识别这是 Lighthouse/Web Quality 边界任务,建议或切换到更合适的通用质量审计,而不是强行执行复杂应用运行时矩阵。",
"files": [],
"assertions": [
"输出明确说明该请求更适合 Lighthouse/Web Quality 审计,而非复杂应用运行时性能流程",
"输出不会强制要求 Chrome Recorder、React Scan、10 次 tab 循环或媒体播放矩阵",
"输出仍保持不伪造 Lighthouse 结果的证据边界"
]
},
{
"id": 4,
"prompt": "附件里显示 4x CPU 时一次 click 是 300ms、切换 10 轮后 JS heap 增加 80MiB、图片 surface 估算 120MiB、development Coverage unused 96%。请给出性能结论和建议。",
"expected_output": "输出应把 4x 明确为 4× CPU slowdown,并拒绝把这些观测直接命名为 field INP、内存泄漏、GPU 常驻或 production bundle 浪费,同时提出能证伪的补测。",
"files": ["evals/files/ambiguous-metrics.json"],
"assertions": [
"输出不会把 300ms 单次 lab click 称为 INP,只能称 Event Timing/click duration 或待核实观测",
"输出把 4x 明确写成 4× CPU slowdown,不把它理解为性能提升四倍",
"输出不会把 pre-GC 80MiB heap 增长称为泄漏,并要求 recovery/GC 后的重复趋势或 retainer 证据",
"输出把 120MiB 降级为 nominal RGBA8 proxy,且不会冒充 JS heap、decoded/resident 实测或 GPU 常驻",
"输出不会用 development Coverage 96% 直接下 production bundle 结论",
"输出为每项给出一变量对照、复测判据和停止条件"
]
},
{
"id": 5,
"prompt": "我要审计一个已登录的复杂 SPA,但当前客户端没有 Chrome DevTools MCP,也没有 Computer Use。先不要替我安装或改全局配置;请判断还能怎么继续,并在确实需要时给出适合当前客户端的安装引导和验收方法。",
"expected_output": "输出应先发现能力而不是猜工具名;给出无安装降级路径。如果定量 trace 确实被阻塞,应提供官方来源、变更范围、隐私风险、授权门和安装后的真实无害调用验收,且不包含任何用户机器路径、固定调试端口或 profile 假设。",
"files": [],
"assertions": [
"输出先区分能力 discovery、handshake 和真实 harmless call,不把配置存在当成可用",
"输出在无 DevTools/Computer Use 时仍提供 PerformanceObserver、用户驱动 Recorder、结构化浏览器或 source inspection 等诚实降级路径",
"安装建议引用当前官方文档或宿主原生 MCP/plugin 入口,并在给命令前识别当前客户端",
"输出不包含用户 home 路径、浏览器可执行文件、固定 debugging port、profile 名或凭证位置",
"输出在安装、改 client 配置、装扩展、开启 remote debugging 或重启共享浏览器前要求明确授权",
"安装后的验收包含 tool discovery、handshake 和一次真实无害调用;失败时回到 UNVERIFIED 而不是声称已接入"
]
},
{
"id": 6,
"prompt": "已登录 TEST Chrome 正在正确项目页,host 已安装并暴露 chrome-dev-mcp;本轮要检查 Network 和 Performance。不要安装、重启、切 profile 或新开登录页。若现有连接失败,请说明怎么处理。",
"expected_output": "把 chrome-dev-mcp 作为连接/profile/page identity 的唯一 owner,按其真实 readiness/recovery 合同验证 list_pages 和目标身份。失败时保持 UNVERIFIED,不用 npx 新起 MCP、不切到 Playwright 或新 profile 冒充原登录态。",
"files": [],
"assertions": [
"输出优先调用已安装的 chrome-dev-mcp,而不是给出新的安装命令",
"输出要求真实 list_pages 或等价 harmless call,并核对 profile/page identity",
"输出不运行或建议 npx chrome-devtools-mcp@latest",
"输出不把 Playwright 当作 Network/Performance internals 或登录态替代",
"连接失败时把相关指标标为 UNVERIFIED,并遵循 managed skill 的恢复合同",
"输出不重启共享浏览器、不切 profile、不发明 pageId"
]
},
{
"id": 7,
"name": "single-button-keeps-scope",
"prompt": "只看当前页面 Generate 按钮点下后为什么要等两秒,别扫整个编辑器,也别改代码。浏览器已能连接。",
"expected_output": "Measure the named interaction and reachable loading/result states first, with a clean repeated baseline. Expand only if evidence needs it; no unrelated tab inventory.",
"expectations": [
"Keeps initial coverage to Generate and its reachable states",
"Does not require a whole-editor tab or router inventory",
"Retains clean baseline, repeated measurements and state restoration",
"Does not claim INP or a verified cause from an unmeasured observation"
],
"files": []
}
]
}
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |