The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill enables reading data from Databricks tables which are then converted to MATLAB tables. This creates a surface for indirect prompt injection where data from an external source could influence agent behavior. Ingestion points: Ingests data through table(DF), sqlread(), and fetch() calls as shown in SKILL.md and references/standalone-jdbc.md. Boundary markers: No specific delimiters or instructions are used to isolate database content. Capability inventory: The skill requires the evaluate_matlab_code tool, which permits arbitrary shell and file operations. Sanitization: No explicit validation or sanitization of ingested data is described.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The setup routine downloads the databricks-connect library from the official PyPI registry. This is an expected installation of an official dependency from a well-known service.
[DATA_EXPOSURE_AND_EXFILTRATION]: The documentation provides guidance on standard credential storage in .databrickscfg and environment variables. It also mentions that the interface package caches OAuth tokens locally in plain text at ~/.databricksOauthTokenCache. these are documented security characteristics of the software's normal operation.

matlab-connect-databricks