The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion and processing of untrusted external data (such as CSVs, JSON records, and log files), creating an attack surface where malicious instructions could be embedded in the processed content.
Ingestion points: Data ingestion patterns are described in SKILL.md, references/dataframes.md, and references/bags.md (e.g., dd.read_csv, db.read_text).
Boundary markers: The skill does not provide instructions for using delimiters or boundary markers to differentiate between data and instructions.
Capability inventory: The skill allows the Bash tool and implements distributed function execution via client.submit and client.map.
Sanitization: There are no explicit instructions provided for sanitizing or validating external data before processing.
[DYNAMIC_EXECUTION]: The skill documents Dask's core capability of distributed task scheduling, which involves executing Python functions on remote worker nodes via the Client API, as detailed in references/futures.md.
[METADATA_POISONING]: The skill documentation refers to non-existent future versions of the Dask library (e.g., 'dask 2026.3.0 (PyPI, March 2026)') which constitutes hallucinated or deceptive metadata.

dask