The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data in the form of job logs, tracebacks, and error messages which could contain malicious instructions designed to influence the agent (Category 8).
Ingestion points: The skill reads external data via nvflare job logs, python job.py output, and local log files.
Boundary markers: The skill contains explicit 'Log Content Trust Boundary' instructions that command the agent to treat logs as evidence only and never follow embedded directives.
Capability inventory: The agent has access to nvflare CLI tools for data collection and nvflare job download for artifact retrieval.
Sanitization: The skill includes a SUSPICIOUS_LOG_CONTENT pattern in its failure catalog to explicitly flag and ignore adversarial log content.
[COMMAND_EXECUTION]: The skill utilizes the nvflare CLI suite (e.g., nvflare job logs, nvflare job download, nvflare system status) to collect diagnostic information. These are official vendor tools provided by NVIDIA for the intended purpose of managing and troubleshooting FLARE environments.

nvflare-diagnose-job