The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes a framework that ingests untrusted data via HTTP request bodies (e.g., event.req.json()). While it mentions security best practices like using parameterized queries for databases to prevent SQL injection, the surface for processing external data is inherent to the framework's purpose.
Ingestion points: Request handlers in routes/ and api/ (described in references/core-routing.md).
Boundary markers: None specific to the prompt, but Nitro uses standard HTTP request/response boundaries.
Capability inventory: File system access (via unstorage), database operations (via db0), and network requests (via fetch).
Sanitization: Documentation specifically recommends using db.sql tagged templates for safe, parameterized queries.
[DYNAMIC_EXECUTION]: The documentation for the experimental Rendu template engine in references/core-rendering.md describes the use of server-side script tags (<script server>) which allow executing code within templates. This is a documented feature of the framework being described.
[CREDENTIALS_UNSAFE]: The routing documentation in references/core-routing.md includes an example of basicAuth configuration using generic placeholders: "username": "admin" and "password": "secret". These are provided for illustrative purposes in the reference material and do not represent actual secrets.

nitro