The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion and processing of data from external files and network locations, which serves as a potential surface for indirect prompt injection.
Ingestion points: Data is loaded via pq.read_table, pcsv.read_csv, and ds.dataset in the Python examples, and via fetch(url) in the JavaScript client code in SKILL.md.
Boundary markers: The provided instructions do not include boundary markers or explicit warnings to the agent to ignore instructions embedded within the data.
Capability inventory: The skill includes scripts for writing files to disk (pq.write_table, ds.write_dataset) and transmitting data to external APIs (fetch POST), creating a potential chain for data manipulation or exfiltration if malicious data is ingested.
Sanitization: No data validation or sanitization logic is present in the provided snippets.
[COMMAND_EXECUTION]: The skill provides instructions for installing well-known, official libraries from standard package registries.
Evidence: pip install pyarrow and npm install apache-arrow in the Installation section of SKILL.md.
[EXTERNAL_DOWNLOADS]: The JavaScript code snippets demonstrate fetching data from remote URLs.
Evidence: The fetchArrowData function in SKILL.md uses await fetch(url) to retrieve Arrow IPC data from a provided URL endpoint.

apache-arrow