workflow-volume-design
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill describes architectural patterns for workflows that ingest and process untrusted external data.
- Ingestion points: The design patterns involve processing data from 'uploads', 'concurrent requests', and external 'subject views' (SKILL.md).
- Boundary markers: The guidance recommends using entry steps for validation, stable subject keys, and unique record paths to maintain isolation.
- Capability inventory: Workflows utilizing these designs are expected to perform filesystem writes to volumes, update embedded databases (SQLite/DuckDB), and execute connector calls.
- Sanitization: The skill advises on incorporating request validation, response signing, and bounding background processing to limit exposure.
- [SAFE]: The skill promotes security best practices by explicitly prohibiting the storage of secrets or sensitive credentials within volumes.
- [SAFE]: Architectural recommendations include the use of read-only mounts ('VOLUME ["views:ro"]') and isolation between draft and production environments to enforce the principle of least privilege.
Audit Metadata