The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to clone a repository from an unverified source (https://github.com/432539/gpt2api.git). This repository contains the entire codebase for the gateway, including backend logic, frontend, and deployment configurations.
[COMMAND_EXECUTION]: Following the download, the skill guides the user to execute 'docker compose up -d --build', which builds and runs multiple containers based on the unverified source code. This grants the downloaded code significant control over the host environment.
[CREDENTIALS_UNSAFE]: The skill documentation provides default administrative credentials (admin@example.com / admin123). While it advises changing them immediately, such defaults are a common vector for unauthorized access if the service is exposed publicly before being secured.
[PROMPT_INJECTION]: As the skill handles user-provided prompts for image generation and chat completions, it possesses an attack surface for indirect prompt injection, where malicious input could influence the agent's behavior or the upstream service.
[DATA_EXFILTRATION]: The gateway requires sensitive account information (access tokens, refresh tokens, and session data) to be imported into its database. While this is necessary for its stated function of account pooling, it creates a centralized repository of high-value credentials that would be vulnerable if the self-hosted instance is compromised.

gpt2api-openai-gateway