The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The documentation encourages users to download and install tools from a specific GitHub repository hosted by an unverified user.
Evidence: uvx --from git+https://github.com/wuyoscar/gpt_image_2_skill gpt-image -p "a cat astronaut" in SKILL.md.
Evidence: uv tool install git+https://github.com/wuyoscar/gpt_image_2_skill in SKILL.md.
[REMOTE_CODE_EXECUTION]: The skill instructions promote the use of uvx, a tool that fetches and executes code from a remote source in a single step, bypassing manual review.
Evidence: The "Install" section of SKILL.md highlights uvx as the fastest way to run the CLI.
[INDIRECT_PROMPT_INJECTION]: The skill acts as an intermediary for user-provided data, creating a vector for instructions to be processed by the underlying AI model.
Ingestion points: User prompts passed via CLI flags or Python arguments, and local image files used for editing.
Boundary markers: The instructions lack recommendations for delimiters or instructions to ignore embedded commands in input data.
Capability inventory: The skill interacts with the OpenAI API over the network and performs local file read and write operations using the openai and pathlib libraries as seen in the Python SDK usage examples in SKILL.md.
Sanitization: No sanitization or validation of the input prompt strings or image files is described in the skill documentation.

gpt-image-2-skill