The Agent Skills Directory

[SAFE]: The skill documentation outlines secure credential management by specifying that ANTA_USERNAME and ANTA_PASSWORD must be provided through environment variables and are never passed as tool arguments or displayed in user-facing output.
[INDIRECT_PROMPT_INJECTION]: This skill possesses a potential attack surface for indirect prompt injection due to its interaction with external network devices.
Ingestion points: Device state information is ingested from Arista EOS switches via the anta_run_tests tool (SKILL.md).
Boundary markers: The instructions do not define explicit boundary markers or delimiters for raw device output, though results are processed into structured verdicts.
Capability inventory: The skill includes tools for network connectivity (anta_run_tests) but is strictly defined as read-only and does not permit configuration changes.
Sanitization: No specific output sanitization or filtering is mentioned, but the focus on structured pass/fail responses reduces the likelihood of the agent interpreting raw device data as instructions.

anta-validation