The Agent Skills Directory

Indirect Prompt Injection (HIGH): The skill implements a workflow where data fetched from an external API (sushi SDK) is directly used to perform high-stakes write operations on the blockchain.
Ingestion points: The response from the getSwap() function in REFERENCE.md.
Boundary markers: None present; the code assumes the integrity of the fetched transaction data.
Capability inventory: Use of walletClient.sendTransaction to execute arbitrary calldata.
Sanitization: No validation or simulation of the calldata is performed before execution.
Data Exposure & Exfiltration (MEDIUM): The reference code (REFERENCE.md) explicitly demonstrates accessing a PRIVATE_KEY from environment variables to sign transactions. This practice exposes sensitive credentials to the agent's runtime environment, making them susceptible to exposure via other injection vectors.
Unverifiable Dependencies (MEDIUM): The skill instructs users to install the sushi and viem packages. While viem is a common library, the sushi package belongs to an organization not included in the trusted source list, making it an unverifiable dependency.

sushiswap-sdk