scheduled-task

Pass

Audited by Gen Agent Trust Hub on Mar 8, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses local shell scripts (create-task.sh, update-task.sh, list-tasks.sh, delete-task.sh, toggle-task.sh) to interact with an internal management API via LOBSTERAI_API_BASE_URL.
  • [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface by design. It accepts arbitrary natural language instructions to be stored in a prompt field and executed at a later time.
  • Ingestion points: The prompt field in the task configuration is populated based on user requests (e.g., "每天早上9点帮我查一下AI新闻").
  • Boundary markers: The skill does not implement delimiters or safety instructions to prevent the agent from obeying malicious commands embedded within these stored prompts.
  • Capability inventory: Scheduled tasks are executed in a "Cowork session" where tool calls are "auto-approved" (as stated in SKILL.md), allowing the stored instructions to perform actions without human oversight.
  • Sanitization: No sanitization or validation logic is present to filter executable or harmful content within the stored prompt fields.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 8, 2026, 04:52 PM