ai-agent-framework-attack

ai-agent-framework-attack is a Penetration testing & red team skill for AI agents, published by brucesongs in the kali-claw repository on GitHub. Skillspace rates ai-agent-framework-attack F (45/100) and the safety scan rated it risky; the rating is based on static analysis only.

AI agent framework attack surface — orchestration-layer compromise of LangChain (Python/JS), LangGraph, CrewAI, Microsoft AutoGen, OpenAI Assistants API v2, Anthropic Claude Agent SDK, LlamaIndex, Microsoft Semantic Kernel, Google ADK, SmolAgents, and MCP-integrated agent runtimes. Covers tool poisoning (Simon Willison 2025 research), indirect prompt injection via untrusted tool outputs, tool rug-pull attacks (npm/PyPI dependency confusion in agent toolkits), LangChain SSRF via RequestsTool / SqlQueryTool / PythonREPLTool (CVE-2024-21514 area, CVE-2024-43480 area), CrewAI tool RCE via decorators (CVE-2024-10231 area), AutoGen code executor escape (Docker/code execution services), OpenAI Assistants API code_interpreter abuse, Claude Agent SDK MCP injection, LangGraph state injection via checkpoint poisoning, LlamaIndex query engine injection, MCP server poisoning (Anthropic/rug-pull.com research Nov 2024), tool description injection, agent memory poisoning, Retrieval-Augmented…

Grade
F
Quality
45 / 100
Safety
Risky
GitHub stars
39

Unproven: rated by static analysis only, not yet compared head-to-head.

Our two independent classifiers differ on this skill's category, so treat the placement as provisional.

This rating describes the version of the repository we crawled; it may have changed since.

View source on GitHubbrucesongs/kali-claw

Our review

Architecture
89 / 100
Maturity
Structured: ships references or scripts
Security scan
High-risk patterns
License
MIT
Adoption
Adopted by 0 other repositories, in 0 collections

What the scanner flagged

  • prompt-injection: ignore-previous
  • exfiltration language

Not checked yet

  • Not judged on the rubric yet (not a contender in its category)
  • No blind task eval yet
  • Scope (portable / vendor / repo-internal) not assessed; only the top of each category is read for scope
  • Sibling files (scripts, references) not fetched; security scan covers SKILL.md only

Better-rated skills in Penetration testing & red team

Skillspace indexes and rates this skill; it does not host or endorse it. Run third-party skills at your own risk.