web3-ai-tools

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (critical risk: 1.00). The skill contains explicit, actionable offensive guidance (automated exploit tooling, credential-stuffing wordlists, SQLi/XSS payload tools, prompt-injection techniques and exploit PoC runners) that can be used to perform malicious attacks.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill includes git clone commands that fetch and run remote code (e.g., https://github.com/KeygraphHQ/shannon, https://github.com/SanMuzZzZz/LuaN1aoAgent, https://github.com/swisskyrepo/PayloadsAllTheThings, https://github.com/advaitbd/smartguard) and the PayloadsAllTheThings clone is explicitly loaded into a RAG knowledge base that will control agent behavior—therefore these URLs are runtime dependencies that execute remote code or directly influence prompts.