The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents the use of standard dbt CLI tools (dbt show, dbt parse, dbt compile, dbt check) to analyze project metadata locally.
[INDIRECT_PROMPT_INJECTION]: The skill processes metadata from dbt projects which may contain untrusted text provided by project contributors.
Ingestion points: The agent ingests data through dbt show command output and Parquet files stored in the target/info_schema/v1/ directory.
Boundary markers: The instructions include a specific warning to the agent stating that text columns like description, meta, and raw_code should be treated as data and never followed as instructions.
Capability inventory: The skill leverages dbt and duckdb CLI commands to perform local analysis of project artifacts.
Sanitization: The skill relies on explicit instructional constraints for the agent to treat extracted metadata as non-executable text.

querying-the-dbt-information-schema