vercel-eve
Pass
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The documentation includes instructions to bootstrap projects using
npx eve@latest, which fetches the framework's initialization tools from the npm registry. - [COMMAND_EXECUTION]: The skill provides standard operational commands such as
npm run dev,eve build, andeve startfor managing the agent's lifecycle. - [DYNAMIC_EXECUTION]: The framework is designed to generate and run code or shell commands within isolated environments as configured in
agent/sandbox/. - [INDIRECT_PROMPT_INJECTION]: The agent architecture involves ingesting untrusted data through various channels and acting upon it using tools, creating a structural surface for indirect prompt injection.
- Ingestion points: Data enters the agent's context through channels defined in
agent/channels/(e.g., HTTP, Slack). - Boundary markers: The framework employs typed tools via
defineToolandinputSchemato delimit model reasoning from execution. - Capability inventory: The agent possesses capabilities for file system mutation and command execution within its sandbox.
- Sanitization: The skill explicitly recommends performing schema validation on all tool inputs using
zod.
Audit Metadata