The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses scripts/dataplex_scanner.py which executes bq query and gcloud dataplex datascans commands to profile datasets. These are legitimate operations for a data engineering tool within the Google Cloud ecosystem.
[COMMAND_EXECUTION]: The instructions mandate using bq commands to fetch schemas and samples, and require the agent to generate SQL transformations for data cleaning in BigQuery.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data profiles (JSON) and schema information from BigQuery/GCS. It implements boundary markers by requiring a 'Profiling Evidence' section in the implementation plan and specific comparison protocols to ensure transformations address anomalies without introducing regressions.
[DATA_EXPOSURE]: The skill processes dataset schemas and metadata. It does not access credentials, SSH keys, or environment secrets. Data movement is confined to the user's BigQuery/GCS environment.
[DYNAMIC_EXECUTION]: The Python script dataplex_scanner.py uses asyncio.create_subprocess_shell to run Google Cloud CLI tools. The inputs (table IDs and locations) are derived from the user's environment and passed via command-line arguments to official CLIs.

data-autocleaning