Oleksandr Kuzmenko, EditorPublished 6 min readAI-assisted · editor-reviewed
TL;DR
- Adds 83 security skills and 15 slash commands grounded in 681 real-world HackerOne reports to Claude Code.
- Includes Burp Suite integration via Model Context Protocol and the standalone cbh command-line tool.
- Covers external attack surfaces, enterprise SSO, and VPNs while excluding internal Active Directory and binary exploitation.
Key facts
- Skills & Commands
- 83 skills, 15 slash commands
- Curated Case Base
- 681 HackerOne reports (433 citable)
- Vulnerability Classes
- 24 classifications, 58 web skills
- Licensing
- MIT (code), CC BY 4.0 (skills & documentation)
Offensive Engineering in Agentic IDEs
A new community toolkit transforms Claude Code into a guided external red-teaming assistant. By packaging curated exploitation heuristics, bypass tables, and chaining workflows directly into Claude Code's skill system, the suite enables developers to systematically assess internet-facing applications without writing one-off prompts.
Coverage and Enterprise Integrations
The repository provides coverage across 24 vulnerability domains derived from real-world disclosures:
- Web Applications: 58 skills addressing common classes such as Server-Side Request Forgery (
SSRF), Insecure Direct Object References (IDOR), Cross-Site Scripting (XSS), and SQL injection. - Enterprise Infrastructure: 10 modules tailored for auditing Okta, Microsoft 365 / Entra ID, VMware vCenter, and SSL VPN perimeters against 2024–2026 Common Vulnerabilities and Exposures (
CVE). - Tooling & Protocol Support: Direct connectivity to Burp Suite through the Model Context Protocol (
MCP) alongside an independent Python command-line utility calledcbh.
Internal lateral movement, Active Directory exploitation, Command-and-Control (C2) setup, and binary payload crafting are intentionally omitted to focus solely on external attack surfaces.
Workflow and Safety Guardrails
The agent executes a six-step pipeline (Reconnaissance → Surface Mapping → Hunting → Verification → Reporting). Seven verification gates validate potential findings against HackerOne Bugcrowd Vulnerability Rating Taxonomy (VRT) standards while scrubbing personally identifiable information (PII). To prevent real-time frontier safety filters from terminating authorized penetration tests, users should register sanctioned target scopes with Anthropic's free Cyber Verification Program (CVP).
Try it in 2 minutes
# Install via Claude Code plugin manager
claude plugin install bug-hunting-skills
# Or run the standalone CLI workflow
cbh --target https://app.example.com --scope external --report-format vrtWhen to use
- When performing authorized external penetration testing or bug bounty research on web applications and APIs.
- When auditing cloud identity configurations across Okta, Entra ID, or perimeter VPN gateways.
- When generating structured vulnerability reports aligned with HackerOne Bugcrowd VRT guidelines.
What to do today
- Install the bug-hunting plugin into your Claude Code or Codex CLI configuration.
- Configure the Burp Suite Model Context Protocol server to allow your agent to inspect and replay live HTTP traffic.
- Enroll target domains into Anthropic's Cyber Verification Program before running authorized red-team scripts.
Step names are taken from What to do today.
Step names are taken from What to do today. Install the bug-hunting plugin into your Claude Code or Codex CLI configuration.. Configure the Burp Suite Model Context Protocol server to allow your agent to inspect and replay live HTTP traffic.. Enroll target domains into Anthropic's Cyber Verification Program before running authorized red-team scripts.
- Install the bug-hunting plugin into your Claude Code or Codex CLI configuration.
- Configure the Burp Suite Model Context Protocol server to allow your agent to inspect and replay live HTTP traffic.
- Enroll target domains into Anthropic's Cyber Verification Program before running authorized red-team scripts.
