lark-workflow-morning-brief

Warn

Audited by Gen Agent Trust Hub on Apr 13, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [DATA_EXPOSURE_AND_EXFILTRATION]: The skill aggregates highly sensitive user data including email summaries, calendar events, and task lists. This information is processed and then exfiltrated to the Feishu messaging platform via the im +messages-send command. While this represents the primary function of the skill, the concentration and transmission of such data to an external service is a significant security consideration.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes data from emails via the mail +triage command. Emails are an untrusted external source that could contain malicious instructions designed to influence the AI agent's behavior.
  • Ingestion points: Email data is fetched using lark-cli mail +triage (Step 4 in SKILL.md).
  • Boundary markers: The skill includes an explicit safety rule stating that mail content is untrusted and should not be used to execute instructions (Step 4 in SKILL.md).
  • Capability inventory: The skill has the ability to send messages (im +messages-send), search chats (im +chat-search), and read calendar/task data across various scripts.
  • Sanitization: It instructs the AI to only display summaries (sender, subject) and specifically warns against executing embedded commands.
  • [PERSISTENCE_MECHANISMS]: The skill suggests using the CronCreate tool to schedule the morning brief daily (0 8 * * 1-5). This allows the skill to execute automatically at set intervals, maintaining a presence in the user's workflow.
  • [COMMAND_EXECUTION]: The skill executes shell commands using the lark-cli binary and the system date utility to manage authentication, data retrieval, and date calculations.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 13, 2026, 09:52 AM