sprite-gen
Pass
Audited by Gen Agent Trust Hub on Jul 25, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill directs the agent to execute a local Python script (scripts/raster_qa.py) using parameters such as filenames and output paths provided by the user in the sprite specification. If these inputs contain shell metacharacters and the agent does not perform sanitization, it could lead to command injection.
- [PROMPT_INJECTION]: The workflow involves interpolating untrusted user input from the sprite specification directly into a natural language prompt for image generation models. This creates a surface for indirect prompt injection where malicious instructions could be embedded in the asset description.
- Ingestion points: The sprite specification fields in SKILL.md (name, subject, style, view) are used as variables for prompt construction.
- Boundary markers: Absent; user inputs are inserted into the prompt template without delimiters like XML tags or clear warnings.
- Capability inventory: The agent is instructed to execute a local script via python3 and write image files to user-defined output_path locations.
- Sanitization: Absent; there are no instructions within the skill to validate, escape, or filter the input strings before they are used in command lines or prompts.
Audit Metadata