The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process potentially untrusted external data from various sources (CSV, Parquet, JSON, log files) using parallel computing collections.
Ingestion points: dd.read_csv(), db.read_text(), and dd.read_parquet() identified in SKILL.md, references/dataframes.md, and references/bags.md.
Boundary markers: No explicit use of boundary markers or instructions to ignore embedded directives in the data are specified.
Capability inventory: The skill has access to Bash, Read, Write, and Edit tools, which could be leveraged if malicious instructions were successfully injected through processed data.
Sanitization: No specific sanitization or validation logic is detailed for the content of the files being read.
[EXTERNAL_DOWNLOADS]: The documentation includes standard installation instructions for well-known data science libraries from official package registries.
Evidence: uv pip install commands for dask, s3fs, gcsfs, and pyarrow in SKILL.md.
[COMMAND_EXECUTION]: Provides instructions for environment setup and connecting to distributed clusters.
Evidence: Usage of Client to connect to remote schedulers and deployment on HPC/Kubernetes clusters described in references/schedulers.md.

dask