pyth

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

• Third-party content exposure detected (medium risk: 0.30). In the required runtime workflow, the client fetches/parses free-text-from-HTTP data at runtime from the Hermes API (e.g., JSON/SSE “event.data”) into LLM-consumed fields such as price, conf, and publish_time, creating an indirect prompt-injection style exposure via outsider-controlled API responses.

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

• Direct money access detected (high risk: 1.00). The skill explicitly includes crypto/blockchain transaction operations: it shows creating a Solana Keypair from a secret key, building/posting price-update transactions via the Pyth Solana receiver SDK, and calling connection.sendTransaction(...) to submit signed transactions. These are specific wallet/signing and on-chain transaction actions (crypto/blockchain execution), which qualify as Direct Financial Execution under the crypto/blockchain criterion.