pentest-ai-agents - Turn Claude Code into your offensive security rese

项目说明

<di="center">

pentest-ai-agents

50 Claude 代码 subagents for penetration testing.

![Claude Code]()
![Agents]()
![Tools Tracked]()
![GitHub stars]()

Quick Start | Cheatsheet | Co | Agents | Examples

</div>

---

Table of Contents

---

pentest-ai-agents is a collection of 50 Claude 代码 subagents that turn Claude into an offensi research assistant. Each agent carries deep domain knowledge in a specific area: recon, web, Acti, cloud, mobile, wireless, social engineering, payload crafting, reverse engineering, exploit chaining, detection engineering, forensics, and more.

Install the agent files. Open Claude 代码. Describe your task. Claude routes to the right specialist automatically.

No servers, no Python deps, no setup beyond copying files.

What's New in v3.3

What's New in v3.2

Agent Map

Tier 1 (ad) agents are routable from any task. Tier 2 (execution-capable) agents require a declared scope and live in the offensi cluster.

What's New in v3.1

Quick Start

One command:

That's it. The script clones the repo to a temp dir, copies the agents to ~/.claude/agents/, and exits. Idempotent: safe to re-run for updates.

Or install as a Claude Code plugin (no clone; updates through the marketplace):

This registers all 52 agents and the slash commands through Claude 代码's plugin system. Pick the plugin or the installer — you don't need both.

Then open Claude Code:

Claude routes to the engagement planner agent and produces a phased plan with MITRE ATT&CK mappings.

Prefer to clone first?

Other install options:

See INSTALL.md for step-by-step instructions, including first-time Claude 代码 setup.

---

Cheatsheet

Quick interactions once installed:

| Command | What It Does |
|---------|--------------|
| /recommend "phish a small SaaS team's IT department" | Picks the right agent and gi next commands |
| /agents-for web | Lists every agent relevant to web testing (web-hunter, api-security, bug-bounty, bizlogic-hunter) |
| /agents-for cloud | Cloud-specific agents (cloud-security, cicd-redteam) |
| db/doctor.sh | Audits which underlying CLI tools you ha by agent. Shows ✔/✘ and install hints. |
| db/doctor.sh --agent ad-attacker | Audit just the AD tooling stack |
| db/doctor.sh --json | Machine-readable output for piping into a script |
| install.sh --tools | Install the underlying tools via your package manager + pipx/go/cargo |
| findings.sh init <id> | Start a new engagement (persistent SQLite findings DB) |
| findings.sh stats | Engagement progress |
| findings.sh export | Full JSON export |
| bash handoff.sh | Markdown handoff report between sessions |

In Claude 代码, just describing your task routes automatically:

---

Co

What the agents dri map to the same surface real ad operate across:

| Category | Agents | Underlying Tools (installable via install.sh --tools) |
|----------|--------|--------------------------------------------------------|
| Recon and OSINT | recon-advisor, osint-collector | nmap, masscan, rustscan, dig, whois, subfinder, amass, httpx, theHar, sherlock, holehe, maigret |
| scanning | | nuclei, nikto, RouterSploit, nmap NSE, Open parsing |
| Web app testing | web-hunter, api-security, bug-bounty, bizlogic-hunter | ffuf, gobuster, feroxbuster, sqlmap, dalfox, Commix, dirsearch, whatweb |
| Acti | ad-attacker, credential-tester | BloodHound, Impacket, NetExec/CrackMapExec, Certipy, kerbrute, Responder, ldapsearch |
| Credentials and cracking | credential-tester, password-auditor | Hydra, Hashcat, John, Medusa, cupp, CeWL, NIST 800-63B policy audit, HIBP k-anonymity |
| Cloud | cloud-security, cicd-redteam | aws/az/gcloud CLIs, Tri, ScoutSuite, Pacu |
| Containers and K8s breakout | container-breakout | kubectl, kube-hunter, peirates, CDK, Falco rule pairing |
| C2 operations | c2-operator | Sliver, Mythic, Havoc, Cobalt Strike, malleable profiles, redirector design |
| AI / LLM red teaming | ai-recon, llm-redteam | A2A/MCP/RAG surface discovery, model fingerprinting, Garak, PyRIT, Promptfoo, OWASP LLM Top 10, MITRE ATLAS |
| Operator opsec | opsec-anonymizer | 源码 IP design, burner identity, JA3/fingerprint hygiene, burn checklists |
| Mobile | mobile-pentester, re | Frida, Objection, jadx, apktool, MobSF, adb |
| Wireless | wireless-pentester | aircrack-ng, hcxdumptool, hcxtools, bettercap, wifite |
| Social engineering | social-engineer, phishing-operator | GoPhish, E, dnstwist, Modlishka |
| Payload crafting | payload-crafter | msfvenom, Donut, MSFvenom Payload Creator, custom loader patterns |
| Reverse engineering | re, malware-analyst | Ghidra, Radare2, JadX, Binwalk, Apktool, IDA, dnSpy, 3 |
| Forensics and IR | forensics-analyst, malware-analyst | 3, exiftool, foremost, YARA, Wireshark/tshark, Autopsy |
| Exploit chaining | exploit-chainer, attack-planner, poc- | Multi-step attack composition, ATT&CK-mapped chain scoring |
| Database attacks | database-attacker | sqlmap, NoSQLMap, nati (MySQL/PostgreSQL/MSSQL/Oracle/MongoDB/Redis) |
| Network L2/L3 and traffic | network-attacker, traffic-analyzer | Responder, ntlmrelayx, mitm6, bettercap, Wireshark/tshark, Zeek |
| Post-exploitation | lateral-mo, persistence-planner, e, data-exfiltrator | PtH/PtT, WMI/WinRM/DCOM, AMSI/ETW e, DNS/HTTPS exfil, golden ticket |
| ICS / OT | scada-attacker | Modbus, DNP3, S7comm, EtherNet/IP, OPC-UA, OT-aware IDS pairing |
| IoT / embedded | iot-pentester | binwalk, UART/JTAG/SPI, BLE/Zigbee/sub-GHz, firmware analysis |
| Secure code and crypto | code-auditor, crypto-analyzer | Semgrep, CodeQL, gitleaks, testssl.sh, jwt_tool |
| Detection and defense | detection-engineer, threat-modeler, stig-analyst | Sigma, Splunk SPL, Elastic KQL, Sentinel KQL, STRIDE/DREAD, DISA STIG |
| CTF | ctf-solver | zsteg, steghide, stegseek, pngcheck, plus generic toolchain |
| Reporting, risk and compliance | engagement-planner, report-generator, risk-scorer, compliance-mapper | MITRE ATT&CK mapping, CVSS 3.1/4.0, EPSS, CISA KEV, PCI/NIST/ISO/CIS mapping |

Run bash db/doctor.sh to see which of these are present on your box right now and what's missing.

---

Agents

Offensi

| Agent | What It Does |
|-------|-------------|
| Engagement Planner | Phased pentest plans with MITRE ATT&CK mappings, time estimates, and ROE templates |
| Recon Advisor | Parses Nmap/Nessus/BloodHound output, prioritizes targets, recommends next commands. *Tier 2: executes recon tools directly.* |
| OSINT Collector | Domain recon, email har, social media profiling, breach data analysis |
| Exploit Guide | Attack methodology for AD, web, cloud, and post-exploitation with defensi |
| Pri Escalation | Linux and Windows privesc: SUID, tokens, services, kernel exploits, container escape |
| Cloud Security | AWS/Azure/GCP pentesting: IAM escalation, container escape, ser exploitation |
| API Security | REST, GraphQL, WebSocket testing. OWASP API Top 10, JWT attacks, OAuth exploitation |
| Mobile Pentester | Android/iOS analysis with Frida, Objection, jadx. OWASP MASTG/MAS |
| Wireless Pentester | WPA/WPA2/WPA3, evil twin, rogue AP, enterprise 802.1X, Bluetooth security |
| Social Engineer | Phishing campaigns, pretexting, for authorized red team engagements |
| Vuln Scanner | Nuclei, Nikto, Nmap NSE scans. Parses Nessus/Open. *Tier 2: executes scans.* |
| Web Hunter | ffuf, gobuster, sqlmap, dalfox. Content discovery, fuzzing, WAF detection. *Tier 2: executes web tools.* |
| Credential Tester | Hydra, Hashcat, John, CrackMapExec. Hash identification, wordlist generation |
| Attack Planner | Correlates findings into multi-step attack chains. Scores by probability, stealth, and impact |
| Supply Chain Auditor | SBOM in, dependency confusion and typosquatting exposure, install-hook review, lockfile and pro integrity |
| Bug Bounty Hunter | HackerOne/Bugcrowd methodology, duplicate a, report writing for maximum payout |
| AD Attacker | BloodHound, Impacket, CrackMapExec, Certipy. Kerberos, delegation, ACL, and cert abuse. *Tier 2: executes AD tools.* |
| Exploit Chainer | Chains low-se findings into full compromise paths. Step-by-step with appro. *Tier 2.* |
| PoC | Generates and safely executes proof of concept scripts. Eliminates false positives. *Tier 2.* |
| Payload Crafter | msfvenom, Donut, custom loaders. Pairs every payload with YARA/Sigma detection content. |
| Reverse Engineer | Ghidra, JadX, Radare2, Binwalk. Static analysis of firmware, mobile apps, and binaries. |
| Phishing Operator | E, GoPhish, dnstwist infrastructure. Li tooling for authorized red team. |
| Swarm Orchestrator | Coordinates all agents as a red team swarm. Parallel workstreams, progress tracking. |
| Business Logic Hunter | Price manipulation, workflow bypass, race conditions, authorization flaws. *Tier 2.* |
| CI/CD Red Team | GitHub Actions, GitLab CI, Jenkins pipeline configs with security gates |
| C2 Operator | Sliver, Mythic, Havoc, Cobalt Strike. Listener tuning, beacon hygiene, redirector design, engagement burn |
| Container Breakout | Docker/K8s escape, runc/cri-o CVEs, kubelet exploitation, RBAC abuse, admission controller bypass |
| OpSec Anonymizer | Operator-side identity hygiene, 源码 IP strategy, burner infrastructure, browser/JA3 fingerprint hygiene |
| LLM Red Team | OWASP LLM Top 10 testing, prompt injection, RAG poisoning, agent 工具 abuse, MCP ser |
| AI Recon | Maps the AI attack surface: AI/LLM endpoints, A2A agent cards, MCP exposure, RAG/工具 surface, model fingerprint. *Tier 2: executes read-only recon.* |
| Password Auditor | NIST 800-63B policy + hash-storage audit, breach exposure (HIBP k-anonymity), lockout-safe spray planning |
| Database Attacker | SQL/NoSQL injection depth, DB pri escalation, proof-scoped extraction across major engines. *Tier 2.* |
| Network Attacker | LLMNR/NBT-NS poisoning, NTLM relay, mitm6, ARP MITM, pi. *Tier 2: executes L2/L3 attacks.* |
| E Specialist | AV/EDR e, AMSI/ETW bypass, in-memory execution — every technique paired with its detection |
| Persistence Planner | Host/AD/cloud persistence (golden ticket, WMI, services) with mandatory cleanup tracking + detection |
| Data Exfiltrator | DNS/HTTPS/ICMP exfil and DLP/egress testing with synthetic/canary data — paired with detection |
| SCADA Attacker | ICS/OT 协议 analysis (Modbus/DNP3/S7comm/OPC-UA), safety-gated, passi |
| IoT Pentester | Firmware, hardware interfaces (UART/JTAG), radio (BLE/sub-GHz), and de testing |
| Lateral Mo | PtH/PtT, WMI/WinRM/DCOM/SSH, token theft, pi — paired with detection |

Defense and Analysis

| Agent | What It Does |
|-------|-------------|
| Fix | Retests a finding after a patch. Replays the original proof, tests the class not the string, and flags partial fixes and regressions |
| Detection Engineer | Sigma, Splunk SPL, Elastic KQL, Sentinel KQL rules with false positi |
| Threat Modeler | STRIDE/DREAD analysis, attack trees, data flow diagrams |
| Forensics Analyst | E acquisition, memory forensics, disk analysis, timeline construction |
| Malware Analyst | Static/dynamic analysis, reverse engineering, YARA rules, IOC extraction |
| STIG Analyst | DISA STIG compliance, GPO remediation paths, keep-open justification templates |
| Code Auditor | Secure-代码 review: injection, authz, secrets, deserialization (Semgrep, CodeQL, gitleaks) |
| Crypto Analyzer | Weak algorithms/modes, key and I handling, TLS/cert config, JWT/JWE token issues |
| Traffic Analyzer | Offline pcap analysis: 协议 dissection, credential extraction, beacon/anomaly hunting |

Reporting and Learning

| Agent | What It Does |
|-------|-------------|
| Report Generator | Professional pentest reports with executi, C, remediation roadmaps |
| CTF Solver | HackTheBox, TryHackMe, PicoCTF. Web, pwn, re, forensics, OSINT |
| Compliance Mapper | Maps findings to PCI DSS, NIST 800-53/CSF, ISO 27001, CIS, HIPAA, SOC 2 with control-gap analysis |
| Risk Scorer | CVSS 3.1/4.0 , EPSS + CISA KE, business-context remediation prioritization |

---

Tier 1 vs Tier 2

Tier 1 (all agents): Ad mode. You paste 工具 output, ask methodology questions, get analysis and recommendations. You run the tools yourself.

Tier 2 (select agents): Can also compose and execute commands directly. You declare your authorized scope, the agent every target against it, and Claude 代码 shows you each command for appro it runs.

| Tier 2 Agent | What It Executes |
|-------|-----------------|
| Recon Advisor | nmap, dig, whois, curl, netcat, traceroute, whatweb, nikto |
| Vuln Scanner | nuclei, nikto, nmap NSE scripts |
| Web Hunter | ffuf, gobuster, feroxbuster, sqlmap, dalfox, whatweb |
| AD Attacker | BloodHound, Impacket, CrackMapExec, Certipy, ldapsearch, enum4linux |
| Exploit Chainer | Multi-step chain execution with approval at each gate |
| PoC | Safe, non-destructi of concept scripts |
| Business Logic Hunter | Logic flaw tests (price manipulation, race conditions) |
| AI Recon | AI endpoint discovery, /v1/models probes, A2A/MCP enumeration (read-only) |
| Database Attacker | sqlmap, NoSQLMap, native DB clients (proof-scoped, non-destructive) |
| Network Attacker | Responder, ntlmrelayx, mitm6, bettercap (scoped L2/L3 positioning) |
| CI/CD Red Team | Pipeline-driven SAST/DAST/secret scans against in-scope targets |

See docs/TIER2-EXECUTION.md for the full safety model.

---

Examples

More examples in the examples/ directory:

| Example | Agent | What It Shows |
|---------|-------|---------------|
| Engagement Plan | engagement-planner | Full phased plan with MITRE ATT&CK mappings |
| Nmap Analysis | recon-advisor | Prioritized attack with follow-up commands |
| Detection Rule | detection-engineer | Kerberoasting detection in Sigma, SPL, and KQL |
| STIG Finding | stig-analyst | STIG analysis with GPO path and keep-open template |
| Report Excerpt | report-generator | SQL injection finding formatted for a pentest report |

---

Running Tools in a Container

Run your security tools inside a Docker container to keep your workstation clean and avoid endpoint protection flags.

Use pentest-ai agents on your host for methodology and analysis. Run the actual tools inside the container.

---

Findings Database

Persistent SQLite storage that keeps engagement data across Claude 代码 sessions.

Tier 2 agents write to the 数据库 automatically when findings.sh is in PATH. See docs/FINDINGS-DB.md for full docs.

---

Token Optimization

Install with lite mode to run ad agents on Haiku (lower cost, same methodology):

See docs/TOKEN-OPTIMIZATION.md for the full guide.

---

Local Models

The agents are plain markdown system prompts — the only Claude-specific part is the YAML frontmatter header. They run on any model: point Claude 代码 at any Anthropic-compatible endpoint by setting ANTHROPIC_BASE_URL, or copy the system prompt (e below the frontmatter) into any local runner such as Ollama or LM Studio.

See docs/LOCAL-SETUP.md for local-model notes.

---

Documentation

| Document | Description |
|----------|-------------|
| INSTALL.md | Installation guide with troubleshooting |
| Agent Guide | How each agent works and when to use it |
| Tier 2 Execution | Execution mode safety model and con guide |
| Local Setup | Run offline with Ollama and local GPU |
| Customization | Modify agents, change models, create new agents |
| Token Optimization | Reduce token consumption |
| Findings Database | Persistent SQLite storage for engagement data |
| Data Privacy | LLM data handling and local model options |
| Changelog | history |

---

MCP Server

Looking for the automated pipeline? pentest-ai is the companion MCP 服务端 with 150+ 工具 wrappers, autonomous exploit chaining, and CI/CD integration. Works with Claude Desktop, Cursor, VS 代码 Copilot, and any MCP 客户端.

---

Prerequisites

---

Legal

This toolkit is for authorized security testing only. Users must ha written authorization before using these agents in any engagement. See DISCLAIMER.md for full terms.

---

License

MIT License

---

<di="center">

Built by 0xSteph &middot; pentestai.xyz

</div>

项目信息

开发语言：Shell

Star 数：2307

技术要点

该项目属于黑客技术方向的开源资源，代码结构清晰，适合安全研究人员学习参考。项目围绕Turn Claude 代码 into your offensi research assistant. Specialized AI subagents for authorized penetration testing plan engagements, analyze recon, research exploits, build detections, audit STIGs, and write reports.展开，提供了可运行的实现，对理解相关安全机制、工具原理与攻防技术具有较高的研究价值。

从技术实现角度看，此类项目的核心在于对底层原理的深入理解与工程化落地。开发者在研究过程中可以结合官方文档与社区讨论，梳理项目的初始化流程、核心模块职责与数据流转，逐步掌握其中的技术细节与设计思路。

使用与学习建议

1. 下载源码压缩包，解压到本地环境。

2. 根据项目类型准备对应的运行环境与依赖。

3. 按照 README 中的步骤完成配置与部署。

4. 在隔离的测试环境中运行验证，避免影响生产系统。

免责声明

本资源来源于开源社区，仅供安全研究、技术学习与合法授权范围内的攻防演练使用，请勿用于任何非法或侵害他人权益的用途。使用本项目产生的一切后果由使用者自行承担。