The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided text from the caption_idea field and interpolates it directly into the prompt used for image editing. This creates a surface where malicious instructions embedded in the caption could potentially influence the behavior of the image generation model.
Ingestion points: The caption_idea input field in SKILL.md is the primary entry point for untrusted data.
Boundary markers: The skill uses single quotes as delimiters (e.g., '<caption_idea>'), but lacks explicit instructions for the model to ignore potential commands within that text.
Capability inventory: The skill utilizes shell commands (muapi CLI) and network operations (curl) to interact with the external muapi.ai service.
Sanitization: No sanitization, escaping, or validation logic is present to handle the caption_idea input before interpolation.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using the muapi CLI and curl for API communication. These commands are necessary for the skill's core functionality of reframing and processing images.

muapi-social-pack