skills/jeremylongshore/tons-of-skills-marketplace/fireflies-reference-architecture/Gen Agent Trust Hub
fireflies-reference-architecture
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a reference architecture for a pipeline that ingests meeting transcript data from external sources (Zoom, Google Meet, MS Teams, etc.). This represents an indirect prompt injection surface because meeting transcripts are untrusted data that could contain malicious instructions or adversarial content designed to influence downstream AI processing or CRM synchronization logic.
- Ingestion points: The
services/webhook-processor.tsscript ingest transcript data via thefirefliesQueryfunction using IDs received from webhooks. - Boundary markers: The instructions recommend verifying event signatures at ingress and placing IDs on an idempotent queue, though specific delimiters for the transcript content itself are not provided in the code examples.
- Capability inventory: The architecture includes components with write capabilities to a database (
db.meetings.upsertinservices/transcript-store.ts) and a task manager or CRM (taskManager.createinservices/action-items.ts). - Sanitization: The instructions explicitly mandate applying "field allowlists before any downstream sync" and enforcing access checks at storage and retrieval to mitigate risks from processed data.
Audit Metadata