llm-trading-agent-security

Pass

Audited by Gen Agent Trust Hub on Jun 24, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: The skill provides defensive logic to detect and mitigate prompt injection attempts. It includes a sanitize_onchain_data function using regular expressions to identify common override patterns like 'ignore previous instructions'.
  • [CREDENTIALS_UNSAFE]: The skill demonstrates secure practices for handling sensitive information. It explicitly instructs users to load private keys from environment variables using os.environ.get('TRADING_WALLET_PRIVATE_KEY') and emphasizes that keys should never be hardcoded or logged.
  • [EXTERNAL_DOWNLOADS]: The skill references a well-known external service, Flashbots (https://rpc.flashbots.net), which is a standard industry tool for MEV protection. No automated or hidden downloads were detected.
  • [SAFE]: The instructional content focuses on preventative security measures such as transaction simulation, circuit breakers, and spend limits to reduce the risk of asset loss in autonomous agents.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 24, 2026, 12:59 PM
Security Audit — agent-trust-hub — llm-trading-agent-security