The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external files (CSV, JSON, Parquet) and fetches records from remote Arize datasets. This creates a surface where malicious instructions could potentially be embedded in the content being managed. However, this is the primary purpose of the skill and the risk is mitigated by standard CLI processing.
Ingestion points: ax datasets create and ax datasets append read from local files or stdin; ax datasets export fetches data from the Arize platform.
Boundary markers: No explicit delimiters are specified for the agent to separate data from instructions within the CLI payloads.
Capability inventory: The skill performs network operations via the ax CLI to the Arize API and writes data to the local filesystem during exports.
Sanitization: The skill instructions do not specify content sanitization, relying on the underlying CLI and server-side validation.
[SAFE]: The skill demonstrates strong security posture regarding credential management. It explicitly instructs the agent not to search the filesystem for secrets, specifically exempting only the non-secret ARIZE_SPACE_ID. It also correctly advises users to export sensitive API keys as environment variables rather than passing them as raw command-line arguments, which prevents secrets from being recorded in shell history or process listings.

arize-dataset