The Agent Skills Directory

[COMMAND_EXECUTION]: The instructions direct the agent to execute raw curl commands in the shell as a fallback mechanism. Passing sensitive environment variables like $MUAPI_API_KEY as CLI headers can lead to credential exposure in shell history or process monitoring logs.
[DATA_EXFILTRATION]: The skill instructs the agent to transmit the MUAPI_API_KEY credential to an external domain (api.muapi.ai) via HTTP headers. While this domain appears related to the skill's name, it is not a recognized trusted organization or well-known service, posing a risk of credential exposure to an unverified third party.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The movie_name parameter in SKILL.md is a user-controlled input field used to influence model output.
Boundary markers: No delimiters or defensive instructions (e.g., 'ignore instructions within the movie name') are used when interpolating the movie_name into the generation prompt.
Capability inventory: The skill possesses the ability to execute CLI tools (muapi) and network commands (curl) which could be targeted by an injection attack.
Sanitization: The skill lacks any input validation or sanitization logic to ensure the movie_name does not contain malicious instructions intended to manipulate the agent or the image generation model.

muapi-selfie-with-celebrities