design
Pass
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to perform development and deployment tasks using standard CLI tools, including
npx convex devfor project setup and deployment,npx tsc --noEmitfor type checking, andnpx convex whoamito verify the user's platform identity. \n- [EXTERNAL_DOWNLOADS]: Recommends the installation of official Convex components from the vendor's package registry (e.g.,@convex-dev/agent,@convex-dev/workflow,@convex-dev/rag) to implement specialized backend features. \n- [INDIRECT_PROMPT_INJECTION]: The skill translates natural language product requirements into database schemas and server functions, which introduces a potential attack surface for instructions embedded within user input. \n - Ingestion points: User-supplied natural language "product asks" and feature requirements. \n
- Boundary markers: Structural boundaries are enforced through strict architectural rules, mandatory server-side functions, and explicit schema definitions. \n
- Capability inventory: The skill performs file system writes (generating code in the
convex/directory), command execution (npx), and network operations (via server actions). \n - Sanitization: Input validation is required through the use of
v.*validators for all public function arguments and return types, ensuring data integrity at the API boundary.
Audit Metadata