The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to take user-provided text for brand names, industries, and stylistic preferences and place them directly into prompts for various image generation models. This creates a surface for indirect prompt injection, as malicious user input could potentially alter the behavior or output of the generation models.
Ingestion points: brand_name, industry, style_preference, color_preference, and mood inputs in SKILL.md.
Boundary markers: None present in the prompt templates.
Capability inventory: Execution of the muapi CLI and curl for remote API calls across all generation steps.
Sanitization: No sanitization or validation of user-provided strings is performed before they are interpolated into the prompt strings.
[COMMAND_EXECUTION]: The skill's operational flow relies on the execution of shell commands. It specifically calls the muapi CLI for core functionality (image generation and prediction status tracking) and specifies curl as a fallback method for direct API requests to the service endpoint.
[DATA_EXFILTRATION]: The instructions for the agent include using the MUAPI_API_KEY environment variable within HTTP headers for requests sent to the remote domain api.muapi.ai. This involves the transmission of sensitive authentication tokens to a remote service as part of the skill's intended operation.

muapi-logo-branding