Pentest‑AI: One‑Command, Fully Automated Penetration Testing in 4 Minutes

Project Overview

Pentest‑AI is an AI‑driven autonomous penetration‑testing framework. It runs recon, logs into the target application, chains discovered vulnerabilities into multi‑step attack paths, validates each finding with a non‑destructive proof‑of‑concept, generates Sigma/SPL/KQL detection rules, and emits a client‑ready report in Markdown, HTML, PDF, SARIF, or JUnit format.

Key Features

10 autonomous agents covering recon, web‑app testing, Active Directory, cloud security, vulnerability chaining, PoC validation, detection‑rule generation, and report output.

Authentication session maintenance so downstream tools inherit cookies.

Credential references only (environment variables, 1Password path, Vault ARN); passwords never appear in shell history or LLM context.

Non‑destructive PoC validation for every finding.

YAML playbooks encode penetration‑testing methodology and are version‑controlled via Git.

Diff mode ( ptai retest <id>) compares old and new results, enabling fix‑retest‑confirm in a single step.

CI‑native support: GitHub Action template, SARIF output, severity gate.

LLM red‑team module with 80+ probes for OWASP LLM Top 10.

MCP server integrating 35+ tools (Claude Code, Cursor, Copilot) controllable via natural language.

Fully local execution under the MIT license; can run offline with Ollama.

Installation & Usage

Install with a single pip command: pip install ptai The first run automatically installs required tools (nmap, nuclei, ffuf, sqlmap, gobuster, etc.).

Basic scan: ptai start https://your-target.com Authenticated scan example (credential reference stored, not in plain text):

# Add authentication profile
ptai auth profile add staging-acme

# Start scan with auth
ptai start https://staging.acme.com --auth-profile staging-acme

Sample output (excerpt):

[+] engagement eng-e512f47b target=staging.acme.com scope=web
[auth]      ✓ Logged in as admin. Session captured, refresh in 14:32.
[recon]     ✓ 3 open ports, 7 subdomains, Apache/PHP fingerprint.
[web]       ✓ 21 findings behind auth. 3 SQLi, 4 XSS, missing CSP, CSRF gap.
[chain]     ✓ Attack path found in 2 hops:
             reflected XSS + cookie without Secure flag → admin session hijack
[validate]  ✓ 3 findings proven with non‑destructive PoCs.
[detect]    ✓ Generated Sigma, SPL, KQL rules for the blue team.
[report]    ✓ reports/eng-e512f47b.html · 12 pages · client‑ready

Total: 4m 18s. Cost: $0.73 in Claude tokens.

Technical Architecture

The workflow is a directed graph:

ptai start <target>
└─────────────────────────────────────────────────────────────┘
  ┌──────────────────┼──────────────────┐
  ▼                  ▼                  ▼
┌───────┐   ┌───────┐   ┌───────┐
│ recon │ → │ auth  │ → │ web   │
└───────┘   └───────┘   └───────┘
    │                       │
    │   ┌────────────────────────────────────┐
    ▼   ▼                                    ▼
┌───────┐                         ┌───────┐
│  ad   │   ┌──────────────────┐   │ cloud │
└───────┘   │ Findings DB     │   └───────┘
            │ (SQLite + scope guard + dedup) │
            └──────────────────────────────────┘
    ┌─────────────────┼─────────────────┐
    ▼                 ▼                 ▼
┌───────┐   ┌─────────┐   ┌──────────┐
│ chain │   │ validate│   │ detect   │
└───────┘   └─────────┘   └──────────┘
    │
    ▼
┌──────────┐
│ report   │  md·html·pdf·SARIF·JUnit
└──────────┘

Supported Scenarios

AppSec teams: CI/CD integration via GitHub Actions, automatic fail‑on‑high, SARIF upload.

Pentest engineers: encode AD penetration methodology as YAML playbooks, share and version‑control.

Bug bounty hunters: run a single command overnight and obtain a PoC‑validated vulnerability list ready for HackerOne.

Red teams: standardize custom AD attack paths as reusable playbooks.

AI feature testing: LLM red‑team mode detects prompt injection, jailbreak, and OWASP LLM Top 10 issues with built‑in probes.

Toolchain Scale

200+ tool wrappers (nmap, masscan, nuclei, ffuf, sqlmap, gobuster, wapiti, nikto, dalfox, xsstrike, enum4linux, bloodhound‑python, impacket suite, trufflehog, gitleaks, kube‑hunter, trivy, …).

4000+ Nuclei templates covering web, network, cloud, and CVE‑specific checks.

35+ MCP tools integrating AI assistants such as Claude Code and Cursor.

300+ LLM models (Anthropic, OpenAI, Ollama, Azure, OpenRouter, DeepSeek, Groq, Mistral, Together AI, Bedrock, Vertex AI, Cohere, …).

Comparison with Other Tools

Feature                | Pentest‑AI | Sn1per | Nuclei | Burp Pro | PentestGPT
-----------------------|------------|--------|--------|----------|-----------
Autonomous Agent Loop | ✅         | ❌     | ❌     | ❌       | ❌
Authenticated Scan     | ✅         | Partial| Raw HTTP| ✅      | ❌
Vulnerability Chaining| ✅         | Partial| ❌     | ❌       | ❌
PoC Validation        | ✅         | Partial| ❌     | ❌       | ❌
Diff / Retest         | ✅         | ❌     | ❌     | ❌       | ❌
CI‑native (SARIF+Gate)| ✅         | ❌     | Partial| ❌       | ❌
LLM Red‑Team          | ✅         | ❌     | ❌     | ❌       | ❌
YAML Playbook         | ✅         | ❌     | Template| ❌    | ❌
MCP Server            | ✅         | ❌     | ❌     | ❌       | ❌
License               | MIT        | GPL    | MIT    | Commercial| MIT

Usage Notes

⚠️ Disclaimer: Pentest‑AI is intended for authorized testing only. Scope files reject out‑of‑scope hosts. PoCs are non‑destructive and rate‑limiting enables stealth mode. Obtain written permission before testing any non‑owned target.

Conclusion

Pentest‑AI eliminates repetitive glue work for penetration‑testing engineers. A single command can produce a complete, PoC‑validated report with detection rules and attack‑path analysis in about four minutes.

Applicable audiences include AppSec engineers, penetration‑testing consultants, bug‑bounty hunters, red‑team members, and product‑security testers who need AI‑enhanced automation.

Source code: https://github.com/0xSteph/pentest-ai