# llm-ai-hunter

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.

- skill · `type=skill` · Claude Code, Cursor, Codex, GitHub Copilot, Gemini CLI, Windsurf (Devin Desktop), Claude.ai · ★ 1298 repo stars
- Author: H-mmer
- License: not stated · policy=link-only · description only; the file stays with the author
- Source: https://github.com/H-mmer/pentest-agents/tree/HEAD/providers/cursor/.cursor/skills/agent-llm-ai-hunter/SKILL.md
- Updated: 2026-06-12
- Security check: automated, 2026-11-16 (text) — no signs of malicious code found.
- HTML: https://affpapa.org/en/nevibecode/skills/llm-ai-hunter

## What it 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.

## Install

### Claude Code (native platform)

1. **Skills CLI (Vercel)**
```bash
npx skills add H-mmer/pentest-agents --skill llm-ai-hunter -a claude-code
```

2. **Manually: put the files in place**
- File: `~/.claude/skills/agent-llm-ai-hunter/SKILL.md` — Copy the whole skill folder; for one project use .claude/skills/agent-llm-ai-hunter/

### Codex

1. **Skills CLI (Vercel)**
```bash
npx skills add H-mmer/pentest-agents --skill llm-ai-hunter -a codex
```

2. **Manually: put the files in place**
- File: `.agents/skills/agent-llm-ai-hunter/SKILL.md` — For all projects use ~/.agents/skills/

### Cursor

1. **Skills CLI (Vercel)**
```bash
npx skills add H-mmer/pentest-agents --skill llm-ai-hunter -a cursor
```

2. **Manually: put the files in place**
- File: `.cursor/skills/agent-llm-ai-hunter/SKILL.md` — Cursor also reads .agents/skills/ and .claude/skills/

### GitHub Copilot

1. **GitHub CLI**
```bash
gh skill install H-mmer/pentest-agents providers/cursor/.cursor/skills/agent-llm-ai-hunter
```

2. **Manually: put the files in place**
- File: `.github/skills/agent-llm-ai-hunter/SKILL.md`

### Gemini CLI

1. **Skills CLI (Vercel)**
```bash
npx skills add H-mmer/pentest-agents --skill llm-ai-hunter -a gemini-cli
```

2. **Platform CLI**
```bash
gemini skills install https://github.com/H-mmer/pentest-agents --path providers/cursor/.cursor/skills/agent-llm-ai-hunter
```

### Windsurf (Devin Desktop)

**Manually: put the files in place**
- File: `.devin/skills/agent-llm-ai-hunter/SKILL.md` — The legacy .windsurf/skills/ path still works

### Claude.ai

**Claude.ai**
- Zip the folder and upload it in Settings → Capabilities → Skills

Where a file is given instead of a command, copy the files from the source to that path: https://github.com/H-mmer/pentest-agents/tree/HEAD/providers/cursor/.cursor/skills/agent-llm-ai-hunter/SKILL.md

## When to use it
- 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
- 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

## Similar
- [llm-security](https://affpapa.org/en/nevibecode/skills/llm-security.md) — skill · 1800 installs
- [llm-risk-assess](https://affpapa.org/en/nevibecode/skills/llm-risk-assess.md) — skill · 186 installs
- [llm-app-security](https://affpapa.org/en/nevibecode/skills/llm-app-security.md) — skill · 365 installs
- [16-ai-llm-security](https://affpapa.org/en/nevibecode/skills/16-ai-llm-security.md) — skill · ★ 71 repo stars
- [prompt-injection](https://affpapa.org/en/nevibecode/skills/prompt-injection.md) — skill · 481 installs
- [llm-redteam-specialist](https://affpapa.org/en/nevibecode/agents/llm-redteam-specialist.md) — subagent · 139 installs

From the same set: [h-mmer/pentest-agents](https://affpapa.org/en/nevibecode/by/h-mmer/pentest-agents.md) — 43 cards

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md
