monitor-experiment

Pass

Audited by Gen Agent Trust Hub on Apr 10, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: Executes shell commands and Python scripts on remote servers via SSH to monitor experiment progress, capture screen session output, and retrieve JSON results.
  • [PROMPT_INJECTION]: Vulnerable to indirect prompt injection because the skill ingests and interprets data from remote screen sessions, JSON result files, and third-party metrics without sanitization or boundary markers.
  • Ingestion points: Remote screen output (tail), experiment JSON result files, and Weights & Biases API metrics data (SKILL.md).
  • Boundary markers: Absent.
  • Capability inventory: Shell command execution via SSH, file read/write/edit across the workflow.
  • Sanitization: Absent.
  • [EXTERNAL_DOWNLOADS]: Fetches experiment metrics and training curves from Weights & Biases, which is a well-known third-party service for machine learning experiment tracking.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 10, 2026, 02:14 AM