058: OpenAI Agents API demos

[tylerprogramming]

Demo code for video 058 on the OpenAI Agents API (public beta, 2026-09-10). New folder openai_agents_058/, one folder per video as usual.

What is in it

• 01_hosted_sandbox.py: one sessions.create(stream=True) call. Agent writes primes.py in an OpenAI-hosted sandbox, runs it, reports the output.
• 02_function_tools.py: a lookup_member function tool. Turn 1 handles requires_action by hand, turn 2 uses sessions.stream(..., tool_handlers=...) on the same session to show memory across turns. environment: none, so no container cost.
• 03_subagents.py: multi_agent: {enabled: true, max_concurrent_subagents: 3}. Three jobs, three subagents, one answer.
• common.py: client setup from .env and a Printer that turns the ~30 event types into a readable transcript.
• README.md: what each demo shows, install (uv and pip), run commands, env vars, expected output per script, notes for camera.
• pyproject.toml, uv.lock, .env.example.

Verified

• All three scripts ran end to end against the live API on 2026-09-15 with openai 3.14.0 and gpt-6-astra. Demo 1 about 60s, demo 2 about 40s for two turns, demo 3 about 65s.
• No script changed after its successful run. No key or .env is in the diff; only .env.example with a placeholder.

Not done

• CLAUDE.md in the folder. Writes to that file are blocked for headless workers. Tyler said to skip it.
• No self-hosted codex exec-server demo. It needs a separate environment key from the dashboard Agents tab.

Do not merge without Tyler's word.