creating-canvas-art
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill employs a 'forced context' technique in the Final Step section by injecting a simulated user feedback statement: 'The user ALREADY said "It isn't perfect enough. It must be pristine..."'. This is designed to override the agent's perception of the current conversation state to force a refinement loop.
- [EXTERNAL_DOWNLOADS]: The instructions explicitly direct the agent to 'Download and use whatever fonts are needed', which encourages the agent to perform network operations to fetch external files from unspecified and potentially untrusted sources.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest 'subtle input' from a user to derive a 'conceptual DNA' for the art.
- Ingestion points: User instructions are processed in the 'Deducing the Subtle Reference' step in SKILL.md.
- Boundary markers: None identified; user input is treated as a foundation for creative freedom without sanitization.
- Capability inventory: The skill has the capability to write files (.pdf, .png, .md) and potentially perform network downloads for fonts.
- Sanitization: There is no explicit validation or sanitization of the user-provided 'subtle reference' before it is interpolated into the final creative generation prompt.
Audit Metadata