The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines an application architecture with several ingestion points for untrusted data that could be exploited if not properly handled by the implementation.\n
Ingestion points: Data enters the system via req.body in routes/datasets.ts (specifically the datasetName and transformations fields) and through file uploads managed by the multer middleware.\n
Boundary markers: There are no explicit boundary markers or "ignore instructions" directives for data processed by the transformation services.\n
Capability inventory: The application possesses the capability to execute database queries via the db.execute and db.query methods in utils/dbHelpers.ts, and to write files to the filesystem via the multer storage configuration.\n
Sanitization: While the datasetService.delete method correctly uses parameterized queries, the datasetService.save method relies on a queryBuilder.buildTransformQuery utility (not fully implemented) which suggests a pattern of dynamic SQL construction from user-provided transformations. This creates a vulnerability surface for SQL injection.\n- [DATA_EXFILTRATION]: The error handling logic in middleware/errorHandler.ts is configured to include the full err.stack trace in the JSON response when process.env.NODE_ENV is set to 'development'. If the environment is misconfigured, this can leak sensitive information about the server's file structure and internal logic to remote clients.\n- [COMMAND_EXECUTION]: The utils/dbHelpers.ts file facilitates the execution of arbitrary SQL commands through a wrapper around the duckdb library. This capability is integrated into the service layer for performing data persistence and retrieval operations.

nodejs-typescript-app