The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the subprocess.run function to execute the kbagent command-line utility. This tool is used to call the query_data function for probing database column values and retrieving metadata.
[DATA_EXFILTRATION]: The skill performs authenticated network operations to Keboola's infrastructure (*.keboola.com) using urllib.request. It retrieves sensitive credentials (Storage API tokens) from environment variables such as KBC_TOKEN or from the local agent configuration file located at ~/Library/Application Support/keboola-agent-cli/config.json.
[PROMPT_INJECTION]: The skill features an attack surface for indirect prompt injection by ingesting data from external APIs and database queries into the agent's context.
Ingestion points: Data is ingested via api_get calls to the Keboola Metastore and through CSV results returned by the kbagent query tool.
Boundary markers: No specific delimiters or boundary markers are defined to isolate external data from instructions.
Capability inventory: The skill possesses the ability to execute shell commands (kbagent), make network requests (urllib.request), and write files to the /tmp/ directory.
Sanitization: Ingested content is parsed using standard JSON and CSV libraries without additional validation or filtering of the data content.

semantic-layer