The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exposes an attack surface for indirect prompt injection as it processes external data to execute financial transactions.
Ingestion points: Reference scripts ingest environment variables (e.g., CIRCLE_API_KEY, EVM_PRIVATE_KEY) and transaction parameters such as recipient addresses and amounts from the execution environment.
Boundary markers: The skill's primary instructions (SKILL.md) mandate explicit user confirmation for all transfers and require input validation before transaction submission.
Capability inventory: The skill possesses the capability to sign and broadcast cryptographic transactions across 11 EVM chains and Solana.
Sanitization: Instructions include explicit warnings against hardcoding secrets and require validation of all addresses and amounts to prevent unintended fund movements.
[COMMAND_EXECUTION]: The provided reference scripts interact with the host environment to retrieve sensitive configuration data.
Evidence: Use of process.env to load private keys and API secrets in files such as references/deposit-evm.md and references/transfer-evm-circle-wallet.md.
Context: This behavior is consistent with standard secure development practices for blockchain tooling to avoid hardcoding credentials in source code.

use-gateway