The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documents installation commands for the openai-agents package and its optional extras from PyPI, which is the official registry for this library. References to remote content (GitHub, OpenAI documentation) all target trusted and well-known service domains including openai.github.io, github.com/openai, and pypi.org.
[CREDENTIALS_SAFE]: The documentation explicitly instructs users to set their API keys via environment variables (e.g., export OPENAI_API_KEY=sk-...) rather than hardcoding them into scripts. It provides placeholders like sk-... and <token> for illustrative purposes, which is a standard safety practice.
[COMMAND_EXECUTION]: The skill includes shell command examples for package installation (pip install, uv add) and project setup. These are transparent, standard developer workflows for the library's intended use and do not contain hidden or malicious payloads.
[INDIRECT_PROMPT_INJECTION]: The skill includes documentation on building guardrails (Category 8) to mitigate indirect prompt injection. It describes ingestion points (user input, tool outputs) and provides architectural patterns for sanitizing and validating data before it reaches the LLM context, which serves to enhance the security of agents built using the SDK.
[DATA_EXFIILLTRATION]: While the SDK facilitates network operations (calling OpenAI APIs), all network activity is directed toward established OpenAI endpoints and configured MCP servers as requested by the user. There are no patterns suggesting unauthorized harvesting of local files or environment secrets.
[NO_CODE]: The provided Python script check-updates.py is a maintenance utility that performs version checks against PyPI and verifies local file integrity. It uses standard libraries (urllib, json) and does not perform any suspicious dynamic execution or obfuscated operations.

openai-agents-sdk