The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill incorporates user-provided text from the person_description parameter directly into prompts for image and video generation tools.
Ingestion points: The person_description input in Phase A.
Boundary markers: No delimiters or negative prompting instructions are used to isolate user input.
Capability inventory: The skill utilizes muapi image edit and muapi video from-image which interface with external AI models via shell commands.
Sanitization: No filtering or sanitization is performed on the user-supplied text before interpolation.
[COMMAND_EXECUTION]: The instructions direct the agent to execute shell commands using the muapi CLI and curl.
Executes muapi image edit and muapi video from-image to generate content.
Uses curl to interact with the MuAPI REST API at api.muapi.ai as a fallback mechanism.
Includes instructions for authenticating the environment using muapi auth configure.
[DATA_EXFILTRATION]: The skill directs the agent to send the MUAPI_API_KEY environment variable to api.muapi.ai via headers in curl requests. While this is the intended communication channel for the service, it involves the transmission of sensitive credentials to a non-standard domain.

muapi-giant-product-showcase