twist-comedy-skit
Pass
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the agent to visit external websites, including
goodcase.aiandmedia.goodcase.ai, to inspect video media, posters, and case summaries. While it also references the well-known servicex.com, the use of less-common external domains for data retrieval is noted. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content, specifically prompt excerpts and scene descriptions within
references/cases.md, to guide its output. This creates a surface where embedded instructions in the source material could potentially influence the agent's behavior. - Ingestion points: The agent reads
references/cases.mdand is directed to visit external URLs to analyze video evidence and prompt examples. - Boundary markers: The instructions include a mandatory "Preserve / Replace / Avoid" reference contract designed to filter which traits from the external evidence are applied to the final artifact.
- Capability inventory: The agent uses browser, image, or video tools to fetch and inspect content from the provided external links.
- Sanitization: The workflow includes verification steps requiring the agent to distinguish observed evidence from its own recommendations and to confirm that the artifact maps back to specific structural traits.
Audit Metadata