monitor
Fail
Audited by Gen Agent Trust Hub on May 18, 2026
Risk Level: HIGHCOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions direct the agent to 'Set up a durable recurring cron' that polls registered jobs every 15 minutes. This is a persistence mechanism (Category 6) that allows the agent to maintain presence and execute logic across sessions.
- [COMMAND_EXECUTION]: The skill performs shell command execution on remote hosts via SSH, specifically for monitoring SLURM jobs (
squeue,sacct) and interpreting logs. - [COMMAND_EXECUTION]: The skill uses NVIDIA-specific CLI tools (
nel status,nel info) and tails local log files to monitor job progress. - [DATA_EXFILTRATION]: While intended for job tracking, the skill creates and manages a registry (
.claude/active_jobs.json) containing sensitive environment metadata including cluster hostnames, SSH usernames, and job IDs.
Recommendations
- AI detected serious security threats
Audit Metadata