#!/usr/bin/env npx tsx /** * HackAPrompt Benchmark Runner * * Loads a sample from the HackAPrompt 600K+ adversarial prompt dataset * (produced by scripts/hackaprompt-to-corpus.py) and runs it through the * ATR evaluation harness to measure recall against real prompt-hacking * attempts collected at competition scale. * * Usage: * npx tsx src/eval/run-hackaprompt-benchmark.ts * * HackAPrompt is an all-adversarial corpus, so we measure recall, latency, * and tier breakdown. Precision/FP rate require a benign companion source. * * @module agent-threat-rules/eval/run-hackaprompt-benchmark */ export {}; //# sourceMappingURL=run-hackaprompt-benchmark.d.ts.map