The Agent Skills Directory

[PROMPT_INJECTION]: The skill documentation implements an interface for processing external data labels from the iOS Visual Intelligence system, which represents a surface for indirect prompt injection. If the system-provided labels contain malicious instructions, they could influence the agent's behavior.
Ingestion points: values(for input: SemanticContentDescriptor) in SKILL.md references external labels and pixel buffers.
Boundary markers: Absent; there are no instructions for the agent to treat these labels as untrusted or to use delimiters.
Capability inventory: The skill execution environment (specified in frontmatter) includes Bash, Write, and Edit tools.
Sanitization: Absent; the provided Swift examples demonstrate direct use of external labels to query internal data stores without validation.

visual-intelligence