openai-agents

Scope

Nous Research の hermes-agent(third-party AI CLI) is a different thing. The Responses API itself is handled by the openai-api-core Skills, Codex CLI is openai-codex skill (this skill specializes in Agents SDK / built-in tools / MCP / orchestration). This skill's MCP is on the consuming side, consuming MCP servers and connectors from the Agents SDK, while the side that exposes MCP servers as ChatGPT apps (window.openai, registerAppTool, ChatGPT UI components) are openai-apps-sdk is handled by the skill. Cross-skill references are by name only, and no relative links are made.

Directory structure

skills/openai-agents/
  SKILL.md
  references/
    agents-sdk/
      README.md
      overview.md
      quickstart.md
      define-agents.md
      models-and-providers.md
      running-agents.md
      results-and-state.md
      sandboxes.md
    tools/
      README.md
      tools-overview.md
      web-search.md
      file-search.md
      vector-stores.md
      code-interpreter.md
      computer-use.md
      image-generation.md
      shell.md
      local-shell.md
      apply-patch.md
      function-tools.md
      tool-search.md
      programmatic-tool-calling.md
      skills.md
    mcp/
      README.md
      mcp-and-connectors.md
      agents-sdk-integration.md
      realtime-sessions.md
      secure-mcp-tunnel.md
    orchestration/
      README.md
      handoffs-and-agents-as-tools.md
      multi-agent.md
      guardrails-and-human-review.md
      background-mode.md
      tracing.md
  samples/
    README.md
    single-agent-function-tool.md
    web-search-tool.md
    file-search-tool.md
    remote-mcp-server.md
    multi-agent-handoffs.md
    input-guardrails.md
    streaming-output.md
  scripts/
    README.md
    install.md
    environment.md
    tracing.md
    run.md

Exploration procedure

Derive the category from the task, and identify the target page in the category's README.md:

1. Find the category corresponding to the task in the mapping table below
2. For that category, references/{category}/README.md Refer to it to identify the target page
3. For the relevant page, .md Read to check the details

Task → Category mapping

Task	Category	Reference README
Want to know about Agent definition, execution, conversation state, and streaming	agents-sdk	references/agents-sdk/README.md
Want to know about model/provider selection and the overall SDK picture	agents-sdk	references/agents-sdk/README.md
Want to use isolated execution environments such as filesystem/shell with SandboxAgent	agents-sdk	references/agents-sdk/README.md
Want to use built-in tools such as web search / file search / code interpreter / computer use	tools	references/tools/README.md
Want to know about defining function tools and creating/searching vector stores	tools	references/tools/README.md
I want to attach Agent Skills (SKILL.md bundles) to the shell tool	tools	references/tools/README.md
Want to know about connecting to and authenticating with remote MCP servers / connectors	mcp	references/mcp/README.md
Want to know about MCP tool integration from the Agents SDK and using MCP in Realtime sessions	mcp	references/mcp/README.md
Want to build multi-agent configurations with handoffs / agents-as-tools	orchestration	references/orchestration/README.md
Want to configure guardrails, human review, tracing, and background mode	orchestration	references/orchestration/README.md
Want to know typical usage	samples	samples/README.md
Want to know about installation, environment variables, tracing configuration, and execution commands	scripts	scripts/README.md