artifact-template-soft-focus-color-haze
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The workflow instructions direct the agent to execute a shell command (
od -An -N2 -tu2 /dev/urandom) to obtain entropy for subject placement randomization. This use of shell access is limited to a benign utility function for procedural logic. - [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied text briefs and images to drive its image generation logic, which creates a potential surface for indirect prompt injection via untrusted data.
- Ingestion points: User-supplied brief and images referenced in workflow step 2.
- Boundary markers: The instructions do not specify explicit delimiters or markers to isolate user input from the system instructions.
- Capability inventory: The agent is authorized to invoke the
$imagegentool as described in steps 3 and 5. - Sanitization: The skill provides extensive rules for abstracting user input (Step 4), requiring the agent to discard anatomy, taxonomy, and surface details, which serves as a functional layer of sanitization for the generated prompt.
Audit Metadata