action-continuity-choreography
Pass
Audited by Gen Agent Trust Hub on Oct 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external prompt excerpts and media content from
references/cases.mdto guide its video choreography workflow. - Ingestion points: Data is read from
references/cases.md, which contains various prompt snippets and links to external sites includinggoodcase.ai,x.com, andmp.weixin.qq.com. - Boundary markers: The skill uses a structured 'Reference contract' with 'Preserve', 'Replace', and 'Avoid' sections to delimit how external evidence influences the final output.
- Capability inventory: The agent is directed to use browser or media inspection tools to view content from external URLs to verify visual traits.
- Sanitization: The workflow includes explicit instructions for the agent to differentiate between observed evidence and its own recommendations and to avoid inventing original details.
Audit Metadata