The Agent Skills Directory

[PROMPT_INJECTION]: The skill presents an indirect prompt injection risk by processing external data sources that could contain malicious instructions.
Ingestion points: Processes dataset files, metadata in dag/main.yml, and workbench artifacts.
Boundary markers: Lacks clear delimiters between untrusted data and system instructions.
Capability inventory: Can perform file writes, git commits/pushes, and SQL queries via subagents like step-fixer.
Sanitization: No sanitization of ingested dataset content is evident before it is used to drive agent actions.
[COMMAND_EXECUTION]: Executes various shell-based tools and subagents through command-line interfaces.
Evidence: Directly executes date, etl update, and make query commands.
Evidence: Orchestrates multiple subagents (etl-pr, snapshot-runner, step-fixer) that interact with the system shell.

update-dataset