process-transformation-story
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the agent to access and analyze external media (MP4, JPG) and web content from
goodcase.aiandx.com(Twitter) to ground the visual style of the generated prompts. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data in the form of prompt excerpts and summaries from
references/cases.mdto influence output generation. This is mitigated by a strict 'Reference Contract' (Preserve/Replace/Avoid) and a verification checklist. - Ingestion points:
references/cases.mdcontains external prompt excerpts and descriptions. - Boundary markers: The 'Required evidence gate' and 'Output contract' in
SKILL.mddefine strict boundaries for how reference data is used. - Capability inventory: Uses browser, image, and video tools to inspect remote assets.
- Sanitization: The 'Verification' section requires the agent to distinguish observed evidence from its own recommendations and labels results as 'prompt-grounded' if media is inaccessible.
Audit Metadata