prisma-cli
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides detailed instructions for executing shell commands via the Prisma CLI to manage databases, migrations, and client generation.
- [INDIRECT_PROMPT_INJECTION]: The skill documents tools like
db pulland the MCP server which ingest external data (database schemas and project files) into the agent's context, creating a surface for indirect prompt injection. - Ingestion points: Database introspection (
prisma db pull) and project configuration analysis via the MCP server (prisma mcp). - Boundary markers: The skill explicitly instructs the agent to use an 'AI safety checkpoint' workflow for destructive commands, requiring fresh, non-inferred user consent.
- Capability inventory: The skill allows for significant database modifications (SQL execution, resets, pushes), file system writes (initialization, generation), and utilizes the network via the Prisma CLI.
- Sanitization: Safety is enforced through a mandatory human-in-the-loop consent process, where the agent must verify specific user approval before proceeding with dangerous actions.
- [SAFE]: The skill incorporates defensive instructions designed to prevent accidental or malicious data loss by implementing a consent-based verification mechanism for destructive operations, matching official Prisma ORM security features.
Audit Metadata