Skip to content
Back to search

Threatswarm

27 scope-enforced AI agents that run the full pentest kill-chain (recon → exploit → post-ex → DFIR → report) as a one-command Claude Code plugin. Backed by 754 MITRE-mapped skills.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/mukul975--threatswarm/badge.svg)](/listings/mukul975--threatswarm)

HTML

<a href="/listings/mukul975--threatswarm"><img src="/api/listings/mukul975--threatswarm/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

**Marketplace** (recommended — installs into any project):

/plugin marketplace add mukul975/ThreatSwarm
/plugin install threatswarm@threatswarm

**Standalone** — clone and use directly:

git clone https://github.com/mukul975/ThreatSwarm && cd ThreatSwarm
claude

**Single session** — load without installing:

git clone https://github.com/mukul975/ThreatSwarm
claude --plugin-dir ./ThreatSwarm/threatswarm-plugin

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

.claude-plugin/marketplace.json · Manifest · 29 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. {
  2. "name": "threatswarm",
  3. "owner": {
  4. "name": "mukul975",
  5. "email": ""
  6. },
  7. "metadata": {
  8. "description": "27 autonomous AI agents for the full penetration testing kill chain — recon, exploit, post-ex, DFIR, and report",
  9. "version": "1.0.0",
  10. "pluginRoot": "."
  11. },
  12. "plugins": [
  13. {
  14. "name": "threatswarm",
  15. "source": "./threatswarm-plugin",
  16. "description": "Complete offensive security operator workspace: 27 specialist agents, 6 engagement commands, scope-gated hooks, and evidence logging",
  17. "version": "1.0.0",
  18. "author": {
  19. "name": "mukul975"
  20. },
  21. "homepage": "https://github.com/mukul975/ThreatSwarm",
  22. "repository": "https://github.com/mukul975/ThreatSwarm",
  23. "license": "MIT",
  24. "keywords": ["security", "pentesting", "red-team", "cybersecurity", "offensive-security", "hashcat", "metasploit", "bloodhound", "active-directory", "ctf"],
  25. "category": "security",
  26. "tags": ["agents", "pentest", "recon", "exploit", "dfir", "threat-hunting", "claude-code"]
  27. }
  28. ]
  29. }
README.md · README · 286 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. <div align="center">
  2. <img src="banner.png" alt="ThreatSwarm — Full kill chain. Scope enforced." width="100%">
  3. <br>
  4. <a href="https://github.com/mukul975/ThreatSwarm/stargazers"><img src="https://img.shields.io/github/stars/mukul975/ThreatSwarm?style=for-the-badge&color=FF6B35" alt="Stars"></a>
  5. <a href="https://github.com/mukul975/ThreatSwarm/blob/main/LICENSE"><img src="https://img.shields.io/github/license/mukul975/ThreatSwarm?style=for-the-badge&color=FF6B35" alt="License"></a>
  6. <a href="#the-27-agents"><img src="https://img.shields.io/badge/agents-27-FF6B35?style=for-the-badge" alt="27 Agents"></a>
  7. <a href="https://claude.ai/claude-code"><img src="https://img.shields.io/badge/Claude_Code-plugin-blueviolet?style=for-the-badge" alt="Claude Code Plugin"></a>
  8. <a href="https://attack.mitre.org"><img src="https://img.shields.io/badge/MITRE_ATT%26CK-mapped-red?style=for-the-badge" alt="ATT&CK Mapped"></a>
  9. <a href="https://github.com/mukul975/Anthropic-Cybersecurity-Skills"><img src="https://img.shields.io/badge/skills-754-orange?style=for-the-badge" alt="754 Skills"></a>
  10. <br><br>
  11. **27 scope-enforced AI agents that run the full pentest kill-chain (recon → exploit → post-ex → DFIR → report) as a one-command Claude Code plugin. Backed by 754 MITRE-mapped skills.**
  12. </div>
  13. ---
  14. > **For authorized security testing only.** Every network command is scope-gated by `scope_check.py` — violations are blocked deterministically at the OS level, not by convention or prompt instruction.
  15. ## Why ThreatSwarm
  16. Most AI pentest tools stop at exploit. ThreatSwarm runs the complete kill chain — recon, exploitation, post-exploitation, DFIR, and a CVSS-scored report — in a single session. It ships as a one-command Claude Code plugin (no Docker, no Postgres, no cloud account required), enforces scope on every tool invocation across all 27 agents, and loads its methodology from the [754-skill Anthropic-Cybersecurity-Skills library](https://github.com/mukul975/Anthropic-Cybersecurity-Skills) mapped to ATT&CK, CSF 2.0, ATLAS, D3FEND, and AI RMF.
  17. ## Demo
  18. ```
  19. $ claude --plugin-dir ./threatswarm-plugin
  20. > /threatswarm:engage 10.10.10.5
  21. ✓ scope check — 10.10.10.5 found in scope.txt
  22. → recon agent starting...
  23. [*] nmap -sS -T3 -p- 10.10.10.5
  24. 22/tcp open ssh OpenSSH 8.9p1
  25. 80/tcp open http Apache httpd 2.4.51
  26. 8080/tcp open http Apache Tomcat 9.0.45
  27. 3306/tcp open mysql MySQL 8.0.28
  28. [*] nuclei -u http://10.10.10.5:8080 -severity critical,high
  29. [critical] CVE-2021-44228 Log4Shell RCE http://10.10.10.5:8080
  30. [high] CVE-2020-1938 Tomcat AJP Ghostcat http://10.10.10.5:8080
  31. [*] feroxbuster -u http://10.10.10.5 -x php,txt
  32. 200 /admin Apache Tomcat Manager [auth required]
  33. 200 /api/v1/users [JSON, 247 entries]
  34. 200 /backup.zip [3.2 MB]
  35. Attack vectors — prioritised:
  36. 1 [CRITICAL] Log4Shell RCE → /threatswarm:attack 10.10.10.5 exploit CVE-2021-44228
  37. 2 [HIGH] API user enum → /threatswarm:attack 10.10.10.5 api
  38. 3 [HIGH] Backup file → /threatswarm:attack 10.10.10.5 web
  39. > /threatswarm:attack 10.10.10.5 exploit CVE-2021-44228
  40. → exploit agent [model: claude-opus-4-7]
  41. → vuln-researcher confirming PoC reliability... weaponized ✓
  42. → listener started on 10.10.14.1:4444
  43. → payload delivered via X-Api-Version header
  44. [+] shell obtained — www-data@10.10.10.5
  45. > /threatswarm:pwned 10.10.10.5
  46. → post-ex agent
  47. [*] sudo -l → (ALL : ALL) NOPASSWD: /usr/bin/python3
  48. [+] privesc → root@10.10.10.5
  49. [*] /etc/shadow extracted → evidence/20260429/10.10.10.5/creds/shadow.hashes
  50. [*] crackmapexec smb 10.10.10.0/24 → 3 hosts reachable with root hash
  51. > /threatswarm:report 10.10.10.5
  52. → report-writer agent → reports/10.10.10.5_pentest.md ✓
  53. ```
  54. *Scope enforced throughout. Evidence written to `evidence/20260429/10.10.10.5/`.*
  55. ## Install
  56. **Marketplace** (recommended — installs into any project):
  57. ```
  58. /plugin marketplace add mukul975/ThreatSwarm
  59. /plugin install threatswarm@threatswarm
  60. ```
  61. **Standalone** — clone and use directly:
  62. ```bash
  63. git clone https://github.com/mukul975/ThreatSwarm && cd ThreatSwarm
  64. claude
  65. ```
  66. **Single session** — load without installing:
  67. ```bash
  68. git clone https://github.com/mukul975/ThreatSwarm
  69. claude --plugin-dir ./ThreatSwarm/threatswarm-plugin
  70. ```
  71. ## Quickstart
  72. 1. Add authorized targets to `scope.txt`:
  73. ```
  74. 10.10.10.5
  75. 10.10.10.0/24
  76. example.com
  77. ```
  78. 2. Set your attacker IP in `.claude/settings.json` → `"LHOST": "10.10.14.1"`
  79. 3. Run `claude` and start a kill chain:
  80. ```
  81. /threatswarm:engage 10.10.10.5 # recon → ranked attack vectors
  82. /threatswarm:attack 10.10.10.5 web # route to specialist agent
  83. /threatswarm:pwned 10.10.10.5 # post-shell: privesc → creds → lateral
  84. /threatswarm:hunt "C2 beaconing" # ATT&CK-mapped threat hunt
  85. /threatswarm:ir ransomware # DFIR triage
  86. /threatswarm:report engagement # CVSS-scored PDF report
  87. ```
  88. ## The 27 Agents
  89. | Agent | Domain | Key Tools |
  90. |---|---|---|
  91. | **— Offensive —** | | |
  92. | `recon` | Port scan · service enum · subdomain discovery | nmap, nuclei, httpx, subfinder, amass |
  93. | `exploit` | CVE exploitation · initial shell access | Metasploit, searchsploit, PoC analysis |
  94. | `post-ex` | Privesc · lateral movement · persistence | linpeas, mimikatz, secretsdump |
  95. | `web-attacker` | SQLi · XSS · SSRF · LFI · OWASP Top 10 | sqlmap, Burp Suite, ffuf, dalfox |
  96. | `api-attacker` | REST · GraphQL · gRPC · BOLA · JWT attacks | RESTler, jwt_tool, arjun |
  97. | `active-directory` | Kerberoast · DCSync · BloodHound · ADCS ESC1-8 | Impacket, BloodHound, certipy |
  98. | `network-ops` | ARP spoof · MitM · SMB relay · packet capture | Responder, Bettercap, tshark |
  99. | `osint` | Domain intel · email harvest · breach data | Shodan, theHarvester, SpiderFoot |
  100. | `wireless-attacker` | WPA2/WPA3 · evil twin · PMKID · EAP capture | aircrack-ng, hostapd-wpe, hcxdumptool |
  101. | `cloud-attacker` | AWS · Azure · GCP · IAM privesc · S3 abuse | Pacu, ScoutSuite, aws-cli |
  102. | `container-attacker` | Docker escape · K8s RBAC · etcd · namespace breakout | Trivy, kube-hunter, kubectl |
  103. | `mobile-attacker` | Android · iOS · APK decompile · Frida · SSL pin bypass | MobSF, Frida, jadx, apktool |
  104. | `social-engineer` | Phishing sim · pretexting · vishing · GoPhish | GoPhish, evilginx2, SET |
  105. | `evasion` | AMSI bypass · AV/EDR evasion · LOTL · sandbox detection | Donut, LOLBins, obfuscation |
  106. | `c2-operator` | C2 infrastructure · implant config · HTTPS blending | Sliver, Havoc, MSF handler |
  107. | **— Specialist —** | | |
  108. | `crypto-attacker` | TLS audit · JWT attacks · padding oracle · RSA weak key | testssl.sh, sslscan, padbuster |
  109. | `iot-attacker` | Firmware extraction · UART/JTAG · MQTT · Modbus | binwalk, RouterSploit, EMBA |
  110. | `password-attacks` | Hash cracking · wordlist gen · credential stuffing | hashcat, John, CeWL |
  111. | `vuln-researcher` | CVE analysis · CVSS scoring · PoC reliability | Nessus, searchsploit, NVD API |
  112. | `reverse-engineer` | Binary RE · ROP chains · shellcode · CTF binaries | Ghidra, Radare2, GDB, pwntools |
  113. | `malware-analyst` | Static/dynamic analysis · IOC extraction · YARA rules | YARA, Cuckoo, PE-studio |
  114. | **— Defensive —** | | |
  115. | `dfir` | Memory forensics · disk forensics · incident triage | Volatility3, AVML, Timesketch |
  116. | `threat-hunter` | ATT&CK hunts · C2 beaconing · persistence detection | Elastic SIEM, Splunk, Velociraptor |
  117. | `blue-team` | Hardening · Sigma rules · Sysmon · CIS benchmarks | Sigma, Wazuh, auditd, fail2ban |
  118. | `log-analyst` | Log parsing · anomaly detection · correlation | Splunk, ELK, wevtutil |
  119. | `compliance-scanner` | CIS · PCI-DSS · NIST CSF · SOC 2 · GDPR | OpenSCAP, Lynis, kube-bench |
  120. | `report-writer` | Pentest reports · executive summaries · CVSS 3.1 | Markdown, evidence aggregation |
  121. ## Scope Enforcement
  122. `scope_check.py` runs as a `PreToolUse` hook before **every** Bash command. It extracts IPs and hostnames from the command, resolves them against `scope.txt` with full CIDR awareness, and returns exit code 2 to block out-of-scope commands. Claude cannot override this — the hook executes outside the agent loop.
  123. **scope.txt example:**
  124. ```
  125. # Authorized targets — add before starting any engagement
  126. 10.10.10.5
  127. 10.10.10.0/24
  128. *.example.com
  129. api.acme-staging.io
  130. ```
  131. **scope.yaml example (HackerOne RoE compatible):**
  132. ```yaml
  133. # ThreatSwarm scope file — maps directly from HackerOne program rules
  134. targets:
  135. in_scope:
  136. - type: cidr
  137. value: "10.10.10.0/24"
  138. note: "Internal lab network"
  139. - type: domain
  140. value: "*.example.com"
  141. note: "All subdomains in scope"
  142. - type: ip
  143. value: "192.168.1.100"
  144. out_of_scope:
  145. - "prod.example.com"
  146. - "10.10.10.254" # firewall — do not touch
  147. engagement:
  148. type: "blackbox"
  149. authorized_by: "Jane Smith, CISO"
  150. authorization_date: "2026-04-29"
  151. rules_of_engagement: "No DoS, no data exfiltration, stop-and-report on critical finds"
  152. ```
  153. Three enforcement layers run in parallel.
  154. ## Architecture
  155. **Three-layer safety model:**
  156. **Layer 1 — `scope_check.py` (deterministic hook)**
  157. Runs before every Bash command. Extracts IPs and hostnames, checks against `scope.txt` with CIDR awareness. Exit code 2 = command blocked. Executes outside the agent loop — Claude cannot override it.
  158. **Layer 2 — CLAUDE.md rules**
  159. Never run active tools in the main thread. Always delegate to agents. Never store plaintext credentials. Never exfiltrate PII. Enforced in every session.
  160. **Layer 3 — Path-scoped output rules**
  161. `evidence/**` — mandatory ATT&CK TTP fields, no raw credentials
  162. `reports/**` — CVSS 3.1 vector required per finding, credentials redacted to `[REDACTED]`
  163. `loot/**` — hash + location reference only
  164. ## Skills Library
  165. Every agent loads its methodology from [Anthropic-Cybersecurity-Skills](https://github.com/mukul975/Anthropic-Cybersecurity-Skills) (5,700+ ⭐) — 754 structured skills mapped to MITRE ATT&CK, NIST CSF 2.0, MITRE ATLAS, D3FEND, and AI RMF. The skills library is the reason ThreatSwarm agents follow professional methodology rather than ad-hoc prompting.
  166. ## Evidence Structure
  167. ```
  168. evidence/
  169. └── 20260429/
  170. └── 10.10.10.5/
  171. ├── nmap/ raw scanner output
  172. ├── nuclei/ vulnerability findings
  173. ├── web/ HTTP responses, screenshots
  174. ├── creds/ hashes only — no plaintext ever
  175. └── screenshots/
  176. findings.md CVSS-scored findings, auto-aggregated
  177. ```
  178. `findings_sync.py` runs at session end and aggregates CRITICAL/HIGH findings to `evidence/FINDINGS_SUMMARY.md`.
  179. ## Parallel Engagements
  180. ```bash
  181. ./scripts/worktree_setup.sh acme-external acme-internal
  182. # Each worktree gets its own scope.txt, evidence/, and git branch
  183. cd ../ThreatSwarm-acme-external && claude
  184. ```
  185. ## Responsible Use
  186. - Written authorization required for every target
  187. - Social engineering campaigns need signed RoE
  188. - OT/ICS: passive monitoring only unless explicitly scoped
  189. - No DoS-class attacks without explicit written authorization
  190. - See `CONTRIBUTING.md` for responsible disclosure policy
  191. ## Citation
  192. ```bibtex
  193. @software{jangra2026threatswarm,
  194. author = {Jangra, Mahipal},
  195. title = {{ThreatSwarm}: Scope-Enforced Multi-Agent Penetration Testing
  196. with the {Claude Code} Plugin Substrate},
  197. year = {2026},
  198. url = {https://github.com/mukul975/ThreatSwarm},
  199. note = {27-agent full kill-chain pentesting plugin for Claude Code}
  200. }
  201. ```
  202. arXiv preprint forthcoming.
  203. ## Roadmap
  204. - [ ] `scope.yaml` HackerOne RoE auto-converter
  205. - [ ] CI/CD integration (GitHub Actions, GitLab CI, SARIF output)
  206. - [ ] Burp Suite / Caido extension
  207. - [ ] GenAI attack surface agent
  208. - [ ] Supply chain / dependency confusion agent
  209. - [ ] IPv6 support in `scope_check.py`
  210. - [ ] Community agent bounty program (Nuclei-template model)
  211. ## Contributing
  212. See [CONTRIBUTING.md](CONTRIBUTING.md). Priority areas: new agents for emerging surfaces (GenAI attacks, supply chain, LLM injection), IPv6 scope support, CI pipeline examples. Open an issue before starting major work.
  213. ## Requirements
  214. - Claude Code 1.0.33+
  215. - Python 3.8+ (hook scripts)
  216. - Kali Linux or equivalent (for pentest tools)
  217. - [Anthropic-Cybersecurity-Skills](https://github.com/mukul975/Anthropic-Cybersecurity-Skills) recommended
  218. ---
  219. MIT License · Built by [Mahipal Jangra (@mukul975)](https://github.com/mukul975)
  220. Powered by [Anthropic-Cybersecurity-Skills](https://github.com/mukul975/Anthropic-Cybersecurity-Skills) · [Privacy-Data-Protection-Skills](https://github.com/mukul975/Privacy-Data-Protection-Skills)
  221. <div align="center"><sub><a href="https://github.com/mukul975/ThreatSwarm">⭐ Star ThreatSwarm</a> · <a href="CONTRIBUTING.md">Contribute</a></sub></div>

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.