peloton
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes fitness class data and metadata from Peloton's catalog, which represents an indirect prompt injection surface.
- Ingestion points: External class data is ingested through commands like
peloton search-class,peloton ride-archived, andpeloton ride-liveinSKILL.md. - Boundary markers: The instructions do not include explicit warnings to ignore instructions embedded in the external class data, though the use of a presentation widget provides some isolation.
- Capability inventory: The skill possesses write capabilities including
peloton schedule-eventandpeloton delete-scheduled-eventas defined inmanifest.yaml. - Sanitization: The skill mandates the use of
widget.createfor presenting class lists, which ensures data is handled as structured UI elements rather than raw conversational text, mitigating the risk of the agent executing instructions found in class descriptions.
Audit Metadata