meeting-transcript
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted external data (meeting transcripts and notes), creating a vector for indirect prompt injection if a transcript contains adversarial instructions.
- Ingestion points: Meeting transcripts, recording notes, and external text files like recent braindumps or project notes are ingested into the agent context via the
/meeting-transcriptcommand. - Boundary markers: Absent. There are no delimiters or strict prompts specifying that the transcript text must be treated as passive data or that embedded system commands should be ignored.
- Capability inventory: The orchestrator spawns multiple parallel sub-agents (
content-extractor,dynamics-analyst, andcontext-enricher) using the Task tool withrun_in_background: true, reads local configuration metadata from00-inbox/MY-PROFILE.md, and interacts with GitHub issues/PRs. - Sanitization: Absent. The content filtering guidelines focus purely on semantic noise reduction (filtering out small talk, side chats, or technical glitches) rather than escaping or neutralizing malicious instructions embedded within the text.
Audit Metadata