meeting-transcript-actions
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted meeting transcripts and attachments which may contain instructions intended to override agent behavior.
- Ingestion points: Reads meeting transcripts, attachments, and linked pages (
SKILL.md). - Boundary markers: Implements a specific XML-style delimiter (
<EXTERNAL_DATA>) and includes a detailed 'Untrusted content boundary' contract that instructs the agent to ignore embedded commands (SKILL.md). - Capability inventory: The skill performs searches and writes to external systems like note stores, task trackers, and contact logs (
SKILL.md). - Sanitization: Instructions mandate structured extraction, schema validation, and URL destination filtering (
SKILL.md).
Audit Metadata