manage

Warn

Audited by Gen Agent Trust Hub on Mar 7, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses platform-specific utilities (systemctl, launchctl, schtasks.exe) to install persistent task schedules that survive system reboots.
  • [REMOTE_CODE_EXECUTION]: The orchestrator allows users to schedule 'script' tasks, which leads to the execution of arbitrary shell scripts via generated wrappers (.sh or .ps1) without manual intervention.
  • [CREDENTIALS_UNSAFE]: The generated wrapper scripts attempt to programmatically retrieve and export the ANTHROPIC_API_KEY from system secret stores (macOS Keychain, Linux GNOME Keyring, Windows Credential Manager) or local configuration files to facilitate non-interactive authentication.
  • [PROMPT_INJECTION]: The skill provides a 'bypass' permission preset that configures tasks to run with the --dangerously-skip-permissions flag, effectively disabling safety guardrails that usually require human approval for tool execution.
  • [COMMAND_EXECUTION]: The scheduler.py script generates executable wrapper files at runtime using string replacement in templates, which are subsequently executed by the platform backends.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 7, 2026, 03:57 AM