bnbagent

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.75). examples/agent-server’s required runtime workflow includes an HTTP provider that fetches external content (DuckDuckGo/news) and then passes the resulting free-form text into the agent’s processing/response path (i.e., into LLM context), creating an indirect prompt-injection risk from outsider-authored web content.

MEDIUM W009: Direct money access capability detected (payment gateways, crypto, banking).

• Direct money access detected (high risk: 1.00). The skill is explicitly an on-chain commerce SDK for BNB Chain and includes wallet backends, private-key usage, EVM signing, and explicit API calls that send value: create_job, set_budget, fund (which escrows ERC‑20 tokens and can send approve), submit (sends txs), settle, claim_refund, dispute, and EIP‑712 signing. These are concrete crypto/blockchain financial operations (escrowing and transferring the U token, approving allowances, signing/sending transactions), so it grants direct financial execution authority.