web3-smart-contracts

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). The skill’s required runtime workflow is driven by user prompts for Solidity/code review (e.g., the evals.json prompts), meaning outsider-authored free text is directly ingested at runtime.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 0.70). The companion-check instructs the agent to run shell commands that inspect hidden skill directories and to offer installing missing skills via npx skills add, which involves filesystem access and running installers that can modify the host and execute code.