lark-workflow-meeting-summary
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted content from meeting titles, notes, and transcripts via the Lark CLI, creating an attack surface for indirect prompt injection.
- Ingestion points: Content is ingested from external Lark systems via
lark-cli vc +search,lark-cli vc +notes, andlark-cli drive metas batch_queryinSKILL.md. - Boundary markers: Absent. There are no instructions or delimiters defined to isolate the untrusted meeting data or prevent the agent from executing instructions embedded within the meeting text.
- Capability inventory: The skill utilizes
lark-clito create and modify documents (lark-cli docs +createandlark-cli docs +update) and relies on executing system commands. - Sanitization: Absent. The skill contains no instructions to sanitize or validate the external text content before interpolating it into markdown summaries or uploading it back to cloud documents.
- [COMMAND_EXECUTION]: The skill explicitly instructs the agent to run local system shell commands (such as the
datecommand) to calculate and format dates instead of performing date calculations within the model context.
Audit Metadata