coinbase-agentkit

BENIGN for publisher/source trust and purpose alignment, but HIGH RISK as an agent capability set because it grants autonomous financial and public-posting actions. No evidence of credential theft, hidden exfiltration, or malicious install behavior; the main risk is real-world action scope inherent to the skill.

No explicit malicious logic (backdoor, exfiltration, credential theft, or command execution) is evident in the provided TypeScript excerpt. However, the module is security-sensitive by design: an LLM agent can trigger runtime compilation and on-chain contract deployment using privileged wallet credentials, and the deployed contract’s true logic cannot be fully validated from the truncated Solidity excerpt. Key review items are the contract source completeness, strict constraints on agent-driven deployment parameters, replacement of runtime compilation with pinned/verified bytecode where possible, and confirmation that contract-creation transactions are formed correctly with `to: undefined` for the specific wallet provider.