The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the gh CLI to search for and retrieve information from GitHub repositories. These executions in scripts/search_ml_datasets.py use list-based arguments with subprocess.run and do not invoke a shell, which is a secure method to prevent command injection.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources (HuggingFace, OpenML, Semantic Scholar, and GitHub) and displays it to the agent. While the script performs basic sanitization through truncation, malicious metadata in external datasets could theoretically influence agent behavior.
Ingestion points: External API responses fetched via requests and CLI output from gh search results in scripts/search_ml_datasets.py.
Boundary markers: Absent; external content is interpolated directly into markdown tables and generated README files without specific delimiters to isolate it from the agent's instructions.
Capability inventory: File writing (write_file), file reading (read_file), and terminal execution (run_terminal via gh CLI).
Sanitization: Dataset descriptions are truncated to 200 characters, and imported GitHub README content is truncated to 2000 characters in cmd_detail.

dataset-discovery