The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/gpt_image_2.py invokes several external command-line utilities via subprocess.run. These include the sops tool for decrypting API credentials from secrets.enc.yaml files, the llm CLI for automated prompt validation, and magick (ImageMagick) for performing image resizing and generating contact sheets.
[INDIRECT_PROMPT_INJECTION]: The skill features a "Preflight" validation mechanism that is susceptible to indirect prompt injection attacks.
Ingestion points: The prompt argument passed to scripts/gpt_image_2.py (file: scripts/gpt_image_2.py).
Boundary markers: The validation query uses a PROMPT: delimiter but does not include explicit instructions for the model to ignore embedded commands (file: scripts/gpt_image_2.py).
Capability inventory: The skill can perform network requests, execute external commands, and perform file system writes (file: scripts/gpt_image_2.py).
Sanitization: No sanitization or escaping is applied to the user-supplied prompt before it is processed by the external validation model.
[EXTERNAL_DOWNLOADS]: The skill establishes network connections to well-known service providers, including api.openai.com and openrouter.ai, to transmit image generation requests and receive binary data.
[DATA_EXFILTRATION]: The script is configured to read from local files (secrets.enc.yaml) to retrieve API keys using the sops utility. These credentials are subsequently transmitted to the configured API providers in the Authorization header of network requests.

gpt-image-2