desk-monitoring
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill describes routines that ingest external, potentially untrusted data from market APIs and research documents, which constitutes an indirect prompt injection surface.
- Ingestion points: Market data retrieved via
/infopolling, WebSocket subscriptions (e.g.,allMids,userFills), and external research files likeresearch/calendar.md. - Boundary markers: The skill mandates that briefings include facts with explicit sources and that any interpretation must be on a clearly labeled separate line.
- Capability inventory: Capabilities are limited to read-only market data fetching, file writing to
/workspace/trading-deskfor logging and briefing, and alert generation. The skill explicitly forbids automated order placement or modification. - Sanitization: The instructions emphasize data integrity by requiring alerts for "unavailable" data rather than allowing stale or missing data to be interpreted as a silent "all-clear" state.
Audit Metadata