meeting-intelligence-system
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests arbitrary untrusted text datasets provided by users (meeting transcripts, recordings, and text notes) in the
SKILL.mdworkspace instructions. There are no explicit boundary markers, delimiter protocols, or content sanitization guidelines to handle embedded instructions within the source transcript data. The skill possesses capabilities to interpolate this data directly into Markdown formats, structured tables, and automated email drafts, which creates an exposed trust surface for downstream indirect injection payloads if a user processes a malicious transcript.
Audit Metadata