The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates the ingestion and processing of untrusted data from external blockchains via the Hyperlane message handler interface, creating a vulnerability surface for indirect prompt injection.
Ingestion points: The handle(uint32 _origin, bytes32 _sender, bytes calldata _body) function in HyperlaneMessageReceiver (templates/message-receiver.sol) and SecureRecipient (examples/custom-ism/README.md) acts as the entry point for untrusted cross-chain message payloads.
Boundary markers: The provided templates implement infrastructure-level validation by checking msg.sender == mailbox and verifying authorized senders, but they do not include markers or delimiters to isolate potential instructions within the message body from the data.
Capability inventory: The skill enables powerful capabilities that could be targeted by injection attacks, including cross-chain message dispatch and remote transaction execution through Interchain Accounts (ICA).
Sanitization: The documentation and templates focus on authorization logic but lack patterns for sanitizing or validating the application-level content of the message payload against adversarial instructions.

hyperlane