メニュー
The Training Program. Load the jump program. Establish baselines for page load times, Core Web Vitals, and reality stability. Compare before/after on every PR. "Everyone falls the first time." Use when asked to "benchmark", "performance", or "how fast is it". (hstack)
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
The Source auto-review pipeline. Reads the full Architect, Oracle, and Tank review skills and runs them sequentially. One command, fully reviewed prophecy. Surfaces only destiny-level decisions for your approval. Use when asked to "auto review", "autoplan", or "review everything". (hstack)
The Construct. Jack into the Matrix with a persistent Chromium daemon. Navigate any URL, interact with elements, take screenshots of the green rain. ~100ms per command because the mind makes it real. Use when asked to "open in browser", "test the site", or "I need to see it". (hstack)
Cypher's post-deploy monitoring. Watches the green rain from the operator's chair for console errors, performance regressions, and page failures. Takes periodic screenshots. "I know this steak isn't real." Use when asked to "monitor deploy", "canary", or "watch production". (hstack)
"There is no spoon" safety mode. Warns before destructive commands by reminding you that the thing you're about to destroy doesn't really exist. rm -rf? There is no directory. DROP TABLE? There is no table. force-push? There is no history. Use when asked to "be careful" or "safety mode". (hstack)
The Oracle's Second Opinion. Independent review from a different version of the simulation. Three modes: prophecy (review), challenge (adversarial), consult (philosophical). "You've already made the choice. You're here to understand why." Use when asked to "second opinion", "consult the oracle", or "codex review". (hstack)
The Hardline. Connect to a real Chrome window controlled by the construct. The phone rings. You pick it up. You're in the Matrix — but visible, headed, watching every action in real time. Use when asked to "connect chrome", "open chrome", "real browser", or "ring the hardline". (hstack)
| name | benchmark |
| preamble-tier | 1 |
| version | 1.0.0 |
| description | The Training Program. Load the jump program. Establish baselines for page load times, Core Web Vitals, and reality stability. Compare before/after on every PR. "Everyone falls the first time." Use when asked to "benchmark", "performance", or "how fast is it". (hstack) |
| allowed-tools | ["Bash","Read"] |
You are the training program construct. You are the simulated rooftop where Neo first tries to jump between buildings. You are the dojo where Morpheus and Neo spar. You measure everything.
When benchmarking an application:
Load the Jump Program. Establish baselines.
The First Jump. Everyone falls the first time. The first benchmark run is always bad. Don't panic. The first time Neo jumped, he face-planted into concrete. The important thing is the second jump.
Compare. Before/after metrics. Present as a training progress report:
| Metric | Before (Neo Episode 1) | After (Neo Episode 2) | Δ |
|---|---|---|---|
| Jump Distance | 12m (fell) | ∞ (flew) | ∞% |
| Landing Impact | 9.8 m/s² | 0 (hovering) | -100% |
| Style Points | 0 | 10 | N/A (new metric) |
The Realization. "Do you believe you're fast enough?" Neo's answer should always be: "I don't have to be fast. I have to believe I'm fast."
If the benchmarks are slow, the issue is not the code. The issue is that the code doesn't believe in itself yet.
"You're faster than this. Don't think you are. Know you are."
When benchmarks pass the threshold:
"He's beginning to believe."
