portfolio-rebalance-review
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- Indirect Prompt Injection Surface: The skill ingests data from external portfolio platforms, user-uploaded documents, and meeting intelligence transcripts. While necessary for its core functionality, processing untrusted external content represents a potential surface for instructions embedded in data to influence the agent's context.
- Ingestion points: Step 1 (Portfolio APIs, Document Store, User Uploads) and Step 4 (Meeting Intelligence).
- Boundary markers: The skill uses specific markdown templates to structure its outputs, providing context separation.
- Capability inventory: The skill possesses capabilities to write markdown files, call subagents, and save data to external platforms.
- Sanitization: A mandatory 'Data-sanity gate' is implemented in Step 1, using a subagent to validate data.
- External Service Integration: The skill interacts with financial analytics platforms to perform risk modeling. This involves transmitting data to the service, which is mitigated by explicit user consent requirements.
- Subagent Delegation: The skill delegates validation tasks to a dedicated subagent, a design that ensures data integrity is verified by a purpose-built component.
Audit Metadata