schedule-job

Fail

Audited by Gen Agent Trust Hub on Jun 21, 2026

Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: Provides instructions for establishing system persistence by modifying crontab files and Windows Scheduled Tasks (schtasks).
  • [COMMAND_EXECUTION]: Encourages the use of the --dangerously-skip-permissions flag when scheduling tasks, which explicitly instructs the agent to bypass security guardrails and user consent prompts.
  • [COMMAND_EXECUTION]: Includes instructions for modifying files in /etc/logrotate.d/, a system-level directory that typically requires root or sudo privileges, indicating a privilege escalation pattern.
  • [PROMPT_INJECTION]: Indirect prompt injection surface via natural language task interpolation.
  • Ingestion points: User-provided natural language task descriptions in SKILL.md.
  • Boundary markers: Absent in command interpolation examples (e.g., claude -p "$TASK").
  • Capability inventory: Uses Bash tool to execute crontab and the claude CLI tool.
  • Sanitization: Absent; instructions suggest direct interpolation into shell command strings without escaping.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Jun 21, 2026, 10:49 PM
Security Audit — agent-trust-hub — schedule-job