The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches query results from the vendor's official platform (datadata.com). These downloads are restricted to data results (NDJSON/CSV) generated by user-initiated queries.
[COMMAND_EXECUTION]: Uses common terminal utilities like "curl", "head", "tail", "jq", and "wc" to download and preview query results stored in temporary local files. This is a standard practice for handling large datasets without overwhelming the agent's context.
[PROMPT_INJECTION]: The skill processes data from external sources (database query results), which is a common surface for indirect prompt injection. However, the skill includes explicit rules to minimize risk, such as the 'Minimal Operation Principle' and instructions to only preview data using "head" or "tail" rather than reading entire files into the conversation.
Ingestion points: External data results downloaded from the Datadata execution engine to the "/tmp/" directory.
Boundary markers: The skill does not currently use specific XML or markdown delimiters when previewing data contents.
Capability inventory: The agent can execute terminal commands for file management ("curl") and data processing ("jq", "grep", "awk").
Sanitization: No automated sanitization of data content is performed; the agent is instructed to use standard command-line tools for manual inspection.

datadata-mcp