route-handlers
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructions process untrusted user input from API requests (headers, body, query parameters, and dynamic route segments) and utilize it in sensitive operations like file system access and database queries.
- Ingestion points: Data enters the agent context through various API endpoint examples, specifically in SKILL.md, references/http-methods.md, and references/streaming-responses.md.
- Boundary markers: While the skill demonstrates structured data validation using Zod in examples/crud-api.md, other utility examples lack explicit boundary markers or warnings against executing embedded instructions in user content.
- Capability inventory: The skill utilizes powerful capabilities including local file system writes (fs.writeFile), file system reads (fs.createReadStream), and database interactions via Prisma.
- Sanitization: The examples/crud-api.md file correctly implements sanitization via Zod. However, the file-handling patterns in references/http-methods.md and references/streaming-responses.md directly use user-supplied parameters in path.join() calls without sanitization, which is a common vulnerability surface for path traversal attacks.
Audit Metadata