The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to download the OpenAI CLI from official sources, including the OpenAI Homebrew tap and the official OpenAI GitHub repository (github.com/openai/openai-cli). These are well-known, trusted sources for the specified tool.
[COMMAND_EXECUTION]: Provides a comprehensive guide for executing shell commands using the openai tool to interact with various API endpoints including files, models, and chat responses.
[CREDENTIALS_UNSAFE]: References standard environment variables (OPENAI_API_KEY, OPENAI_ADMIN_KEY) for authentication. It correctly uses placeholders (sk-...) for examples and does not contain any hardcoded secrets.
[PROMPT_INJECTION]: The skill identifies a surface for indirect prompt injection where untrusted user input is passed to the --input flag of the CLI tool. This is a characteristic of tools designed to process natural language inputs.
Ingestion points: User-supplied strings used in commands like openai responses create --input "$input" in SKILL.md.
Boundary markers: No explicit delimiters or instructions are provided to the agent to treat input as data rather than instructions.
Capability inventory: The agent can execute shell commands and perform network operations via the CLI.
Sanitization: No sanitization or validation of the input strings is performed by the skill instructions.

openai-cli