epic-fantasy-scifi-spectacle-by-1g5goov
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill requires the agent to access and analyze media and descriptions from external domains including goodcase.ai, media.goodcase.ai, and x.com. These references are used as visual evidence to ground the video prompt generation process in established styles.
- [INDIRECT_PROMPT_INJECTION]: By fetching content from external URLs (video summaries and media metadata), the skill creates a potential attack surface for indirect prompt injection where instructions embedded in external content could influence the agent. The skill mitigates this through several structural measures:
- Ingestion points: The agent is directed to inspect specific URLs listed in references/cases.md using browser or media analysis tools.
- Boundary markers: The skill implements a 'reference contract' (Preserve, Replace, Avoid) that forces the agent to explicitly categorize and filter information extracted from the external source.
- Capability inventory: The agent utilizes image, video, and web browsing capabilities to gather data from the specified external sources.
- Sanitization: The instructions require the agent to confirm the evidence file was read, distinguish between observed evidence and original recommendations, and preserve creator attribution, preventing the accidental adoption of external instructions as primary rules.
Audit Metadata