hackprobelisted
Install: claude install-skill wuguizh0u/hackxplus-2.0
# hackprobe — AI-Assisted Black-Box Security Audit
## CONFIRMATION (MANDATORY)
Before ANY action, ask: **"You are about to run an active security audit against `<URL>`. Are you sure you want to proceed?"** Only proceed after explicit confirmation.
## STEP 0 — Directory Init + Tool Discovery (MANDATORY)
```bash
source infra/directory_setup.md
init_hackprobe_dir "<URL>"
source scripts/discover_tools.sh # 导出 $HACKPROBE_NMAP/$HACKPROBE_SQLMAP/$HACKPROBE_NUCLEI 等
cd "$WORK_ROOT"
```
`discover_tools.sh` 自动发现 `D:/1渗透tools/` 下 40+ 工具,导出为 `$HACKPROBE_*` 环境变量。找不到的工具标记 skip,不阻塞流程。后续所有 Wave 优先用 `$HACKPROBE_*` 调工具,不存在再 curl 手工。
## PREREQUISITE — audit-rules (MANDATORY)
**Before Wave 1 starts**, invoke `Skill("audit-rules")` to load the audit discipline cheat-sheet. This refreshes:
- 🗑 Junk-finding blacklist: CORS w/o credentials, Self-XSS, Sourcemap, version-only, TLS warnings
- 🔬 No PoC = does not exist. Phenomenon ≠ vulnerability.
- ⏱ 20-minute rule: No exploitable finding in 20 min on one endpoint → switch attack surface
- 🎯 Attack surface decision tree: login→IDOR, API→unauth access, upload→bypass, payment→race condition
All audit-rules constraints override any conflicting instructions in downstream wave files.
## Architecture — Wave-Driven Pipeline
```
Wave 0 → Pre-flight 威胁建模(10-15s,不阻塞 Wave 1)
├── curl 首页特征 + httpx 技术栈 + DNS/CDN/WAF
└─��� → _shared/target_profile.json(类型+策略)
↓
Wave 1 → 侦察与攻击面发现
├── 1a: 4 Agent 并行(OSINT/子域/端口/域名)+ A5 Host碰撞
├──