The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to execute local scripts using the uv run command. Specifically, it invokes tools/ideogram4.py to process JSON captions and communicate with the Ideogram API. The skill also references other local tools in the same directory, such as tools/flux2.py and tools/ltx2.py for integrated workflows.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to take untrusted user natural language and interpolate it into a structured JSON payload for an image generation service. This creates a surface where malicious user instructions could potentially influence the agent's behavior during the expansion process.
Ingestion points: User-provided text for titles, thumbnails, and descriptions is processed in SKILL.md and examples.md to populate the text and desc fields in the JSON schema.
Boundary markers: There are no specific delimiters or instructions defined to separate user-provided content from the rest of the JSON structure or to prevent the agent from obeying instructions embedded within the user text.
Capability inventory: The skill possesses the capability to read local brand configuration files (brands/<brand>/brand.json), write files (caption.json), and execute network-enabled tools (tools/ideogram4.py) that interface with external hosted APIs.
Sanitization: The instructions do not specify any sanitization, escaping, or validation logic for the user's input before it is used to build the tool's payload.

ideogram4