The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script scripts/generate_image.py to interface with the Azure AI API. The script is invoked by the agent with parameters for the prompt, image dimensions, and output path. The instructions specify using separate shell tokens and quoting user-provided values to prevent command injection.
[EXTERNAL_DOWNLOADS]: The Python script makes network requests to Azure AI image generation endpoints to retrieve image data in Base64 format. This operation targets a well-known service and is the primary function of the skill.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection through the user prompt and the API response data, but it implements sufficient controls for a safe verdict.
Ingestion points: User prompt input (processed in SKILL.md) and the JSON API response (specifically the b64_json field in scripts/generate_image.py).
Boundary markers: SKILL.md instructs the agent to pass arguments as separate shell tokens and quote all user-provided values when calling the script.
Capability inventory: The skill uses urllib.request.urlopen for network communication and pathlib for writing the resulting image to the filesystem.
Sanitization: The script performs strict JSON decoding and uses base64.b64decode with validation enabled to verify the integrity of the image data before writing it to a file.

generate-images-mai