AgentRunProof v0.2: provider-free Runner regression checks

[FU-max-boop]

AgentRunProof v0.2 is available on PyPI and now has a 30-second first run, a real provider-free Runner tool-loop example, and a five-minute RunState case study.

What it does

It drives the real OpenAI Agents SDK Runner with deterministic public-Model responses, compares run and run_streamed, checks state and exactly-once side effects, and writes content-addressed JSON records. Built-in scenarios need no API key and make no model-provider request.

Reports backed by this harness were referenced by merged upstream fixes #4413 and #4414. That is diagnostic impact, not OpenAI adoption or endorsement.

Try it

pip install agentrunproof==0.2.0
agentrunproof probe basic-tool-session-parity --certificate proof.json
agentrunproof check-certificate proof.json

Which deterministic runtime contract would be most useful next: handoffs, retries, max turns, or cancellation?