llm-trading-agent-security
Pass
Audited by Gen Agent Trust Hub on Aug 12, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill addresses the risk of untrusted data (token names, webhooks, social feeds) influencing agent behavior, which serves as an ingestion point for potentially malicious instructions.\n
- Ingestion points: The
sanitize_onchain_datafunction and general instructions highlight external data sources like webhooks or social feeds entering the LLM context inSKILL.md.\n - Boundary markers: The skill recommends architectural isolation, prompt hygiene, and independent spend policies as logical boundaries.\n
- Capability inventory: The agent has the capability to sign and send on-chain transactions using
eth-accountand web3 providers.\n - Sanitization: Implements regex-based filtering (
INJECTION_PATTERNS) to detect and block common injection strings such as instruction overrides or unauthorized transfer commands.\n- [EXTERNAL_DOWNLOADS]: The documentation referenceshttps://rpc.flashbots.netas a recommended RPC endpoint for MEV protection.
Audit Metadata