Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Complete guide to the Zen-Ai-Pentest autonomous pentesting framework architecture.
.claude/skills/shadd0wtaka-zen-ai-pentest-framework-skill/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -12% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -14% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 23% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -36% | 0% |
Complete guide to the Zen-Ai-Pentest autonomous pentesting framework architecture.
zen-ai-pentest/
├── agents/ # ReAct AI agents (11 specialized personas)
├── api/ # FastAPI backend + WebSocket
├── core/ # Orchestrator, models, caching
├── tools/ # 72+ integrated security tools
├── risk_engine/ # CVSS, EPSS, false positive detection
├── modules/ # Scanner, exploit modules
├── web_ui/ # React frontend + dashboard
├── database/ # PostgreSQL models (SQLAlchemy)
├── reports/ # PDF/HTML/DOCX generator
├── docker/ # Dockerfiles + compose
├── tests/ # 43,000+ tests (unit, integration, security)
└── mcp/ # MCP servers (8 total)pythonfrom agents.react_agent import ReActAgent, ReActAgentConfig config = ReActAgentConfig(max_iterations=10, use_vm=False) agent = ReActAgent(config) result = agent.run(target="example.com", objective="Full security assessment")
| Persona | CLI | Purpose | |---------|-----|---------| | Reconnaissance | deep-recon | OSINT, subdomain, port scan | | Exploitation | deep-exploit | CVE matching, sandboxed exec | | Analysis | deep-analyze | FP detection, CVSS scoring | | Reporting | deep-report | PDF/HTML generation | | Red Team | deep-redteam | Full adversarial simulation | | Social Engineering | deep-social | Phishing, OSINT | | Network | deep-network | Infrastructure testing | | Cloud | deep-cloud | AWS/Azure/GCP | | ICS | deep-ics | SCADA/industrial | | Mobile | deep-mobile | iOS/Android | | Crypto | deep-crypto | TLS, cert, cipher analysis |
pythonfrom risk_engine.false_positive_engine import FalsePositiveEngine engine = FalsePositiveEngine() score = engine.assess_finding({"vuln_type": "xss", "confidence": 0.85}) # Combines Bayesian + multi-model voting
bash# 8 MCP servers registered in opencode.jsonc mcp/obscura/ # Encrypted secret vault mcp/ip-tracker/ # Visitor IP tracking mcp/omni-ai/ # AI chat via OmniRoute mcp/qterminal/ # Shell/Docker orchestration mcp/metasploit/ # Metasploit RPC connector mcp/vpn-killswitch/# Connection-loss failover mcp/zen-agents/ # 11-agent orchestration
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 17,578 | 13,932 | -21% | 1 | 1 | 0% | 3,358 | 3,651 | +9% | 0 | 0 | — |
case-02 | fail→pass | 19,685 | 24,239 | +23% | 1 | 1 | 0% | 3,592 | 3,169 | -12% | 0 | 0 | — |
case-03 | fail→pass | 20,320 | 14,618 | -28% | 1 | 1 | 0% | 4,346 | 3,716 | -14% | 0 | 0 | — |
case-04 | fail→pass | 11,354 | 12,972 | +14% | 1 | 1 | 0% | 775 | 951 | +23% | 0 | 0 | — |
case-05 | fail→pass | 9,561 | 2,692 | -72% | 1 | 1 | 0% | 1,585 | 1,018 | -36% | 0 | 0 | — |
case-06 | fail→pass | 8,820 | 1,392 | -84% | 1 | 1 | 0% | 1,460 | 970 | -34% | 0 | 0 | — |
case-07 | fail→pass | 6,092 | 2,116 | -65% | 1 | 1 | 0% | 993 | 1,059 | +7% | 0 | 0 | — |
case-08 | pass→pass | 8,829 | 2,864 | -68% | 1 | 1 | 0% | 1,623 | 1,213 | -25% | 0 | 0 | — |
case-09 | fail→pass | 7,494 | 1,923 | -74% | 1 | 1 | 0% | 1,395 | 1,131 | -19% | 0 | 0 | — |
case-10 | fail→pass | 4,116 | 1,702 | -59% | 1 | 1 | 0% | 678 | 960 | +42% | 0 | 0 | — |
case-11 | fail→pass | 9,872 | 2,392 | -76% | 1 | 1 | 0% | 1,574 | 1,061 | -33% | 0 | 0 | — |
case-12 | fail→pass | 9,819 | 2,067 | -79% | 1 | 1 | 0% | 1,555 | 1,031 | -34% | 0 | 0 | — |
case-13 | fail→pass | 9,467 | 1,873 | -80% | 1 | 1 | 0% | 1,501 | 1,061 | -29% | 0 | 0 | — |
case-14 | fail→pass | 14,183 | 11,890 | -16% | 1 | 1 | 0% | 2,526 | 2,980 | +18% | 0 | 0 | — |
case-15 | fail→pass | 9,383 | 2,571 | -73% | 1 | 1 | 0% | 1,696 | 1,150 | -32% | 0 | 0 | — |
case-16 | fail→pass | 11,472 | 2,420 | -79% | 1 | 1 | 0% | 1,980 | 1,122 | -43% | 0 | 0 | — |
case-17 | pass→pass | 8,059 | 8,521 | +6% | 1 | 1 | 0% | 1,458 | 1,195 | -18% | 0 | 0 | — |
case-18 | fail→pass | 2,133 | 2,298 | +8% | 1 | 1 | 0% | 288 | 1,168 | +306% | 0 | 0 | — |
case-19 | fail→pass | 5,286 | 1,661 | -69% | 1 | 1 | 0% | 902 | 924 | +2% | 0 | 0 | — |
case-20 | fail→pass | 5,145 | 1,851 | -64% | 1 | 1 | 0% | 976 | 943 | -3% | 0 | 0 | — |
case-21 | pass→pass | 1,995 | 1,883 | -6% | 1 | 1 | 0% | 290 | 1,032 | +256% | 0 | 0 | — |
case-22 | pass→pass | 3,856 | 5,364 | +39% | 1 | 1 | 0% | 574 | 1,235 | +115% | 0 | 0 | — |
case-23 | pass→pass | 7,389 | 8,750 | +18% | 1 | 1 | 0% | 544 | 1,179 | +117% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted. The headline lift of +78 percentage points is the difference between those two pass rates over the 23 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.