post-meeting

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [Indirect Prompt Injection Surface]: The skill processes untrusted meeting transcripts from external sources (Zocks, Wealthbox, or manual pastes) to generate CRM entries. This creates a potential vector where instructions embedded within a transcript could attempt to influence the agent's output.
  • Ingestion points: Meeting content is pulled from Zocks AI results, Wealthbox summaries, or advisor-pasted transcripts as described in the Inputs section of SKILL.md.
  • Boundary markers: The instructions do not define explicit structural delimiters for the transcript content, though they emphasize only including items actually said in the meeting.
  • Capability inventory: The skill has the ability to write records (Notes, Tasks, Opportunities) to connected CRMs (Wealthbox, Redtail) and file documents in Google Drive.
  • Sanitization: There are no specific instructions for sanitizing or escaping the content of the transcripts before processing.
  • Mitigation: A mandatory 'human-in-the-loop' step is implemented in Step 4, where the advisor must approve the entire batch in a schema-shaped table before any write operations occur, which significantly reduces the risk of unintended actions.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 03:47 PM
Security Audit — agent-trust-hub — post-meeting