The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided images (screenshots, diagrams) and generates code or diagnostic advice based on the contents. This represents a potential surface for indirect prompt injection if the visual data contains instructions designed to mislead the agent.
Ingestion points: image_path parameters in ui_to_artifact, diagnose_error_screenshot, and understand_technical_diagram tool calls in SKILL.md.
Boundary markers: Absent. The skill does not explicitly instruct the agent to ignore instructions embedded within the images.
Capability inventory: Vision analysis and code generation tools (zai-vision:*).
Sanitization: Absent. There is no mention of validating or sanitizing the content extracted from images before using it to generate code or suggestions.

vision-expert