twinmind-hello-world
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a workflow that ingests untrusted data from live meeting transcriptions and voice memos (Step 1 and Step 2). It lacks explicit boundary markers or delimiters to distinguish between transcribed speech content and system instructions when the agent processes the transcripts for summaries or action items. The skill operates in an environment where
Read,Write, andEdittools are permitted. No sanitization or filtering is prescribed for the transcribed content, making the workflow potentially susceptible to indirect prompt injection if a meeting participant speaks malicious commands intended for the AI agent.
Audit Metadata