muapi-product-campaign
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill interpolates user-controlled input fields, such as
product_nameandcampaign_message, directly into generative prompts for image and video models. - Ingestion points:
SKILL.md(Inputs section used in Step 1 through Step 5). - Boundary markers: Absent; there are no delimiters or instructions for the models to ignore potentially malicious commands embedded in user input.
- Capability inventory: The skill uses
muapiCLI tools andcurlfor network requests. - Sanitization: Absent; input is used verbatim in prompt templates.
- [COMMAND_EXECUTION]: The skill provides instructions for the agent to use
curlto interact with the MuAPI backend if CLI aliases are not available. This includes passing theMUAPI_API_KEYvia headers. This is a functional requirement for the skill but involves executing shell commands with sensitive environment variables. - [EXTERNAL_DOWNLOADS]: The skill initiates requests to
api.muapi.aito generate and download campaign assets (images and videos). These are directed to the vendor's primary service domain.
Audit Metadata