The Agent Skills Directory

EXTERNAL_DOWNLOADS (LOW): The README.md recommends installation via 'npx skills add melonask/viem-sweep-skills'. The author/organization 'melonask' is not on the trusted organizations list, meaning the source code is unverifiable and managed by an unknown entity.
INDIRECT_PROMPT_INJECTION (LOW): The skill identifies a significant attack surface for indirect prompt injection if the agent attempts to execute these strategies using external data.
Ingestion points: Functions in references/strategies.md accept untrusted inputs such as tokenAddress, recipient, and privateKey as arguments.
Boundary markers: Absent; there are no instructions or delimiters within the code to prevent the agent from misinterpreting inputs as code or commands.
Capability inventory: The skill utilizes walletClient.writeContract and adminWallet.sendTransaction to perform on-chain asset transfers.
Sanitization: None; the reference implementations do not include validation or sanitization for addresses or input hex strings before use in sensitive operations.
DATA_EXPOSURE (SAFE): While the code examples in references/strategies.md use privateKey parameters (e.g., privateKeyToAccount(privateKey)), these are standard implementation patterns for the Viem library and do not represent hardcoded secrets or exfiltration logic.

viem-sweep