design-direction
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data including 'existing UI/brand evidence', 'constraints', and 'validated prototypes' to generate a design document (
docs/design.md). - Ingestion points: The agent is instructed to 'Inspect the actual product and visual context' and process developer-supplied scope and prototype data.
- Boundary markers: The instructions do not specify any delimiters or warnings to ignore instructions embedded within the inspected visual or brand evidence.
- Capability inventory: The skill primarily performs file-write operations to
docs/design.mdand context retrieval via thevisual-tastetool. - Sanitization: No explicit sanitization or filtering of the ingested visual or textual evidence is mentioned.
- [EXTERNAL_DOWNLOADS]: The skill documentation references external design standards and guidelines hosted on GitHub repositories belonging to 'Leonxlnx' and 'pbakaus'. While these are presented as sources for the agent to follow, they originate from third-party sources that are not identified as trusted organizations, introducing a dependency on external content that is not under the author's direct control.
Audit Metadata