Claude-BugHunter: 83 AI Skills for Bug Bounty and Red Team Work

By Prahlad Menon 5 min read

Security researchers just got a serious upgrade. Claude-BugHunter is a skill bundle that transforms Claude Code from a coding assistant into a bug-hunting partner — with 83 skills, 15 slash commands, and 681 disclosed vulnerability patterns baked in.

Built by Sachin Sharma, this is designed for authorized penetration testing, bug bounty hunting, and red team engagements.

Repo: github.com/elementalsouls/Claude-BugHunter

What’s In the Box

The bundle covers the external attack surface — anything reachable from the internet:

CategorySkillsCoverage
Web App Vulns58XSS, SQLi, SSRF, IDOR, LFI, SSTI, XXE, CSRF, CORS, open-redirect, SharePoint, ASP.NET/NTLM
Enterprise Platforms10M365/Entra, Okta, vCenter, SSL VPNs, cloud IAM, APK red-team
Reporting & Validation6Triage, evidence hygiene, VRT-aware severity, H1/Bugcrowd formats
Recon & OSINT5Subdomain enum, identity mapping, cert transparency, JS analysis
Methodology4bb-methodology, redteam-mindset, local toolkit setup

The 681 vulnerability patterns come from disclosed HackerOne reports — 433 are individually cited and auditable.

How It Works

The magic is automatic skill loading. Describe what you’re testing in plain English, and the relevant skills activate:

You: Testing acme.com — an in-scope HackerOne target. Run recon and rank the surface.

Claude: ⟳ loading skills: web2-recon, offensive-osint, bb-methodology …
        → subdomain enum (subfinder + crt.sh) … 47 hosts
        → live hosts (httpx) … 12 · tech fingerprint … 6 distinct stacks
        → ranked surface:
            api.acme.com (GraphQL, introspection ON) ← start here
            auth.acme.com (OAuth, SSO) ← hunt-oauth
        
        Next: want me to probe the GraphQL introspection + OAuth redirect_uri?

No memorizing command names. Just describe the target and the work.

Installation

/plugin marketplace add elementalsouls/Claude-BugHunter
/plugin install claude-bughunter@elementalsouls

All skills load namespaced under claude-bughunter: and auto-update.

Option B: Copy Install

git clone https://github.com/elementalsouls/Claude-BugHunter.git
cd Claude-BugHunter
bash scripts/install.sh  # or pwsh ./scripts/install.ps1 on Windows

This copies skills to ~/.claude/skills/.

Multi-Harness Support

The skills work across multiple AI coding tools:

bash scripts/install.sh --all --burp-mcp
HarnessSkill Path
Claude Code~/.claude/skills/
OpenCode~/.claude/skills/ or ~/.agents/skills/
OpenAI Codex CLI~/.agents/skills/
Hermes Agent~/.hermes/skills/
Google AntiGravity~/.gemini/config/skills/

The knowledge layer ports to all five — slash commands stay Claude-Code-only.

The 6-Phase Workflow

Claude-BugHunter enforces a structured methodology:

  1. Recon — Subdomain enum, tech fingerprinting, surface ranking
  2. Map & Rank — Identify high-value targets, prioritize attack vectors
  3. Hunt — Execute vuln-class-specific techniques (the 58 hunt-* skills)
  4. Validate — 7-Question Gate before any finding is submitted
  5. Report — Platform-specific formatting (H1, Bugcrowd, Intigriti, Immunefi)
  6. Iterate — Non-linear, return to earlier phases as needed

The 7-Question Gate is critical for quality:

  • Is this in scope?
  • Is the impact on the accepted list?
  • Is the evidence reproducible?
  • Is PII properly redacted?

Enterprise Platform Coverage

This isn’t just web app testing. The bundle includes attack matrices for enterprise infrastructure:

Identity & Access:

  • Microsoft 365 / Entra ID attack chains
  • Okta-as-IdP exploitation
  • Cloud IAM (AWS STS, cross-account confused-deputy)

Infrastructure:

  • VMware vCenter / Workspace ONE
  • SSL VPN appliances (Cisco, Fortinet, Citrix, Palo Alto, Pulse, SonicWall, F5)
  • SharePoint (ToolShell + legacy SOAP)

Mobile:

  • Android APK red-team pipeline
  • Supply-chain recon

Cloud:

  • Public S3 misconfig
  • IMDS chains
  • STS AssumeRole abuse

Burp Suite Integration

The bundle integrates with Burp via MCP (Model Context Protocol):

bash scripts/install.sh --burp-mcp

This wires up the Burp MCP server so Claude can interact with Burp’s proxy, scanner, and repeater during testing.

What It Doesn’t Cover

The bundle explicitly excludes internal network attacks:

  • No AD attacks — BloodHound, Kerberoasting, DCSync, Pass-the-Hash
  • No C2 frameworks — Cobalt Strike, Sliver, Mythic
  • No post-exploit — Mimikatz, golden tickets, persistence mechanisms
  • No evasion — AMSI bypass, AV/EDR bypass
  • No binary exploitation — Kernel pwn, browser internals

Why? Different operational risk profile. These require different judgment and tooling. A future bundle may cover internal red team work, but this one stays external-only.

Legal & Ethical Use

This is for authorized testing only:

  • Bug bounty programs where the target is in scope
  • Penetration tests with signed engagement letters
  • Your own infrastructure
  • CTF challenges and training platforms (DVWA, Juice Shop, Hacker101)

The skills include built-in validation gates that trigger when you point Claude at unverified targets. The 7-Question Gate explicitly asks about scope and authorization.

Anthropic’s Cyber Verification Program

Anthropic’s models have safeguards that can block offensive security work — even when authorized. If you’re doing legitimate bug bounty or pentest work, enroll in Anthropic’s Cyber Verification Program (CVP) to get safeguards adjusted.

Note: On Opus 5, some higher-risk cyber requests fall back to Opus 4.8 silently. Check the model label in responses if something seems off.

Slash Commands

The bundle includes 15 slash commands for structured workflows:

CommandPurpose
/huntStart an engagement with scaffolding
/reconRun reconnaissance workflow
/reportGenerate platform-specific report
/triageRun 7-Question validation gate
/scopeCheck/set scope boundaries

Plus the cbh CLI for terminal-native orchestration:

pipx install git+https://github.com/elementalsouls/Claude-BugHunter
cbh recon acme.com --depth full
cbh classify findings.json
cbh report --format h1

Getting Started

For a HackerOne target:

You: I have authorization for acme.com on HackerOne. 
     The scope includes *.acme.com except staging.acme.com.
     Help me find valid security issues.

Claude: [loads bb-methodology, web2-recon, triage-validation]
        Starting engagement for acme.com...
        
        Phase 1: Recon
        - Subdomain enumeration
        - Tech stack fingerprinting  
        - Surface ranking
        
        [proceeds with structured workflow]

For enterprise red team:

You: External red team engagement for BigCorp. 
     Targets: *.bigcorp.com, mail.bigcorp.com, vpn.bigcorp.com
     Looking for initial access vectors.

Claude: [loads enterprise skills, cloud-iam-deep, M365/Entra chains]
        Analyzing external attack surface...
        
        Identified:
        - Azure AD tenant (bigcorp.onmicrosoft.com)
        - GlobalProtect VPN (Palo Alto)
        - Exchange Online + legacy on-prem
        
        Priority vectors: [ranked list]

Why This Matters

Bug hunting has always required deep knowledge across dozens of vulnerability classes, platform-specific quirks, and reporting standards. This bundle encodes that knowledge:

  • 681 patterns from real disclosed reports
  • Platform-specific chains (not generic advice)
  • Validation gates to avoid noise submissions
  • Proper evidence hygiene built into the workflow

It won’t replace experience, but it dramatically accelerates the process — especially for researchers expanding into new vulnerability classes or enterprise platforms.

Links: