Skip to main content
skill CX Codex

llm-ai-hunter — Claude Code skill for LLM and AI agent security testing

llm-ai-hunter is a Claude Code skill that takes the role of an LLM and agentic AI security specialist. It covers OWASP LLM Top 10 v2025 and OWASP Agentic AI Top 10 across the full attack surface.

no license ★ 1,298 updated 2026-06-12

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add H-mmer/pentest-agents --skill llm-ai-hunter -a claude-code
Open on GitHub

What llm-ai-hunter does

On activation it reads SKILL.md as the testing methodology, then calls search_techniques and search_payloads to pull proven exploitation techniques before building an attack plan.

A subtype arrives with the dispatched task — prompt-injection, indirect-injection, tool-abuse, rag-poisoning, vector-idor, mcp, model-server, output-handling, or ascii-smuggling — and the skill routes to the matching CVE patterns. Among them: LangChain PythonREPLTool semantic RCE (CVE-2025-68613, CVSS 9.8), BentoML pickle family (CVE-2025-27520), Ollama RCE (CVE-2024-37032), Open WebUI Direct Connections SSE injection (CVE-2025-64496). When no subtype is given, it infers one from the target description.

The methodology is scoped to authorized bug bounty programs. RCE-class findings are demonstrated with id or an OOB DNS callback — no destructive commands. Pickle payloads emit a single ping to a controlled endpoint. The safety constraints are part of the methodology file, so the skill will not chain beyond primitive demonstration unless program scope explicitly permits it.

What it helps with

  • test a chatbot or RAG pipeline for prompt injection and indirect injection
  • find vulnerabilities in an MCP server through tool description poisoning
  • probe a model server for pickle deserialization or path traversal
  • detect data exfiltration via hidden Unicode in a chat client output
  • check a multi-tenant vector DB for IDOR across namespace and collection IDs

Example prompts for the agent

  • Test this chatbot for indirect prompt injection through uploaded documents
  • Run llm-ai-hunter with subtype tool-abuse on the agent that has access to PythonREPLTool
  • Audit the MCP server for tool description poisoning and SSE injection

How to install llm-ai-hunter

1Skills CLI (Vercel)

npx skills add H-mmer/pentest-agents --skill llm-ai-hunter -a claude-code

2Manually: put the files in place

~/.claude/skills/agent-llm-ai-hunter/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/agent-llm-ai-hunter/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about llm-ai-hunter

How do I install llm-ai-hunter in Claude Code?

Run npx skills add H-mmer/pentest-agents --skill agent-llm-ai-hunter -a claude-code. The methodology file lands at ~/.claude/skills/agent-llm-ai-hunter/SKILL.md.

What does llm-ai-hunter do?

It acts as an LLM and agentic AI vulnerability specialist, covering all ten categories of OWASP LLM Top 10 v2025 and all ten categories of OWASP Agentic AI Top 10 — from prompt injection to model server attacks and ASCII smuggling.

Does llm-ai-hunter work in Cursor and GitHub Copilot?

Yes. For Cursor run npx skills add H-mmer/pentest-agents --skill agent-llm-ai-hunter -a cursor; the file goes to .cursor/skills/agent-llm-ai-hunter/SKILL.md. For GitHub Copilot run gh skill install H-mmer/pentest-agents agent-llm-ai-hunter; the file goes to .github/skills/agent-llm-ai-hunter/SKILL.md.

Similar skills

All skills for Security →

From the same set43

The whole set →
validator validator — a skill for vetting bug bounty findings before writing the report. Runs a 7-question system and 4-gate checklist, filters weak and theoretical bugs in minutes, and outputs a decision: PASS, KILL, DOWNGRADE, or CHAIN REQUIRED. CX Codex +1 ★ 1,298repo stars sast-flow-tracer The sast-flow-tracer skill maps paths from entry points to dangerous operations, following function calls, checks, and data transformations. CX Codex +1 ★ 1,298repo stars sast-flow-tracer sast-flow-tracer traces data propagation from attack entry points to dangerous operations and maps every validation step in between. CX Codex +1 ★ 1,298repo stars llm-ai-hunter llm-ai-hunter is a subagent that tests chatbots, RAG, MCP servers, and agent tools for security flaws within an authorized scope. CX Codex +1 ★ 1,298repo stars validator validator — a triage subagent for security researchers that kills weak findings before report writing using a 7-question gate and 4-gate validation checklist. Runs real HTTP proofs, not code review. CX Codex +1 ★ 1,298repo stars business-logic Guides coding assistants in auditing business logic flaws like price tampering, coupon race conditions, and MFA bypasses. Designed to prove impact for bug bounties. CX Codex +1 ★ 1,298repo stars browser-stealth-agent The browser-stealth-agent skill drives a local camofox-browser server to get past Cloudflare, Akamai, DataDome and PerimeterX bot checks, then verifies client-side bugs in a real browser. CX Codex +1 ★ 1,298repo stars sast-hunter The sast-hunter skill builds and runs a PoC for a specific suspected vulnerability. Use it after the candidate has passed code and condition checks. CX Codex +1 ★ 1,298repo stars

Project sponsors

Companies that keep this analytics open