stable-baselines3
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes data from external vendor tools and local files, which constitutes a surface for indirect instructions.
- Ingestion points: Data is ingested through tool outputs from
sandbase_run_get(SKILL.md) and local data files mentioned in the workflow. - Boundary markers: While explicit prompt delimiters are not defined, the skill requires strict validation of data shapes, types, and schemas before any processing occurs.
- Capability inventory: The skill utilizes
sandbase_run(references/sandbase-api-map.md) for external API execution and performs local model training and artifact creation. - Sanitization: The skill mandates 'Quality gates' for data inspection and requires using only 'schema-defined arguments' for tool invocations (SKILL.md).
Audit Metadata