llm-trading-agent-security
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEPROMPT_INJECTIONCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill contains defensive code examples that use regex patterns to identify and block known prompt injection keywords within input strings.
- [CREDENTIALS_UNSAFE]: Practical examples demonstrate how to securely load wallet private keys from environment variables, adhering to industry standards for secret management and avoiding hardcoded credentials.
- [EXTERNAL_DOWNLOADS]: Includes references to the Flashbots RPC endpoint (rpc.flashbots.net), a recognized and well-known service for protecting Ethereum transactions from frontrunning and MEV.
- [INDIRECT_PROMPT_INJECTION]: The skill identifies a data ingestion surface (external strings like token names) and provides defensive measures. It implements boundary markers via the
INJECTION_PATTERNSregex, identifies high-risk capabilities including transaction signing (account.sign_transaction) and on-chain submission (send_raw_transaction), and provides an explicit sanitization function (sanitize_onchain_data) to filter external content.
Audit Metadata