meeting-minutes-taker

Pass

Audited by Gen Agent Trust Hub on Apr 12, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill follows a structured processing lifecycle that includes multi-pass generation to prevent content loss and mandatory human confirmation for speaker mapping and final minutes approval.
  • [EXTERNAL_DOWNLOADS]: Recommends the 'markitdown' utility from Microsoft for document conversion. This is a well-known service from a trusted organization and is documented neutrally as a pre-processing recommendation.
  • [COMMAND_EXECUTION]: Employs standard filesystem operations (directory creation, file writing) and the platform-native 'Task' tool for parallel processing. All operations are logically scoped to transcript-specific intermediate directories within the user's workspace.
  • [PROMPT_INJECTION]: The skill processes untrusted transcript data, which represents an indirect prompt injection surface.
  • Ingestion points: Reads raw transcript files provided by the user.
  • Boundary markers: The instructions do not define specific delimiters for separating transcript content from agent instructions in subagent prompts.
  • Capability inventory: Includes filesystem writes, directory management, and the ability to spawn subagents via the Task tool.
  • Sanitization: No explicit filtering of transcript text is performed.
  • Mitigation: The design mitigates potential instruction overrides through independent redundant passes, union-based merging of outputs, and rigorous human-in-the-loop validation.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 12, 2026, 01:01 AM