The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run within templates/linear_batch.py to execute its own design_image.py script. This is used to support sequential batch generation. The command is safely constructed as a list, preventing shell injection vulnerabilities.
[EXTERNAL_DOWNLOADS]: The skill interacts with api.apimart.ai to submit image generation tasks and poll for completion status. It downloads the resulting image files to a local output directory. These operations are core functionalities and target a documented API provider.
[DATA_EXFILTRATION]: The skill transmits user-provided prompts and optionally base64-encoded local images (from the case_library used as style references) to the external apimart.ai endpoint. This is the intended behavior for image-to-image and design-led generation workflows.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input and interpolates it into complex design prompts using the design_compiler.py module. While it lacks rigorous sanitization, it employs structured prompt construction logic that treats user input as a 'brief' within a larger design context, which reduces the impact of potential injection attempts.

multi-agent-image