speed-guard-plus

Built on affaan-m/ECC by @affaan-m (246,421 stars, MIT). All credit for the original idea to them. This fork improves and repackages it; upstream license preserved in UPSTREAM_LICENSE.

A Claude Code skill that checks page, API, and build speed, saves a fair baseline, and finds slowdowns after code changes.

🚦 Why

Small code changes can make an app slow.

Speed tests are hard to trust when the setup changes between runs. Cache state, screen size, test data, and other apps can change the result.

speed-guard-plus gives Claude Code a clear test plan. It is for developers who want to check a pull request, fix a slow app, prepare a launch, or compare two stacks.

It is a single-file skill with no added packages. It plugs into your current Claude Code flow.

📦 Install

mkdir -p ~/.claude/skills/benchmark && curl -fsSL https://raw.githubusercontent.com/speed-guard-plus/speed-guard-plus/main/skill/SKILL.md -o ~/.claude/skills/benchmark/SKILL.md

Restart Claude Code after the install.

🧪 Usage

Ask Claude Code:

Use the benchmark skill to check if this pull request made the home page slower.

The skill will guide a fair before-and-after test. It warms up first, runs the same test at least three times, and reports the middle result and full range.

Expected output:

| Metric | Before | After | Change | Result |
|---|---:|---:|---:|---|
| LCP | 1.8 s | 2.2 s | +400 ms | WARN |
| Image size | 420 KB | 730 KB | +310 KB | WARN |

The home page is slower.
LCP rose by 400 ms, or 22%.
It still meets the 2.5 s goal.
The larger images are the main cause found.
Result: WARN

The four modes are:

1. Page speed
2. API speed
3. Build speed
4. Before and after

Results use five clear labels: BETTER, SAME, WARN, FAIL, and N/A.

🔧 What we changed vs upstream

• Rewritten from Japanese into clearer, simpler English while preserving the four benchmark modes and default targets.
• Added explicit safety and privacy rules for live sites, load tests, credentials, user data, and side-effecting API calls.
• Added repeatability guidance: document the setup, warm up, run tests at least three times, report median/range, retain raw data, and mark missing results N/A.
• Expanded page, API, and build testing with cold/warm cache states, responsive viewports, real INP actions, error/rate-limit reporting, and consistent concurrency.
• Made regression judgments noise-aware and threshold-based, though the provided improved file is truncated before completing the final example and omits the original storage/CI integration sections.

📄 License

This project is released under the MIT License.

The upstream MIT license and credit are preserved in UPSTREAM_LICENSE.