The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided prompts and reference images that are transmitted to an external API service.\n
Ingestion points: The skill accepts free-form text via the --prompt argument and user-specified files via the --input argument in gpt_image.py.\n
Boundary markers: The script does not implement explicit delimiters or instructions to isolate user-provided content from the system's operational parameters when calling the API.\n
Capability inventory: The skill utilizes tool access to execute shell commands (python3), performs network communication with the OpenAI API and other URLs via the openai and httpx libraries, and writes files to the local system using the Pillow library.\n
Sanitization: There is no validation or filtering performed on the user-supplied prompt string before it is passed to the script and subsequent API calls.\n- [COMMAND_EXECUTION]: The skill instructs the agent to run a bundled Python script with command-line arguments generated from the user's instructions.\n
The --prompt and --output flags are populated with data from the conversation and used to execute a shell command.\n
The script also dynamically manages local file paths based on these user-provided arguments.\n- [EXTERNAL_DOWNLOADS]: The script gpt_image.py downloads generated image files from URLs provided by the OpenAI API service response.\n
The save_results function uses the httpx library to fetch content from the image_data.url returned by the OpenAI service and writes it to the local disk.

gpt-image-skill