Explained test execution despite LLM endpoint callPublic

Agent

Model

GPT-5.2 Codex

Started

Jan 24, 12:43 AM

Messages10

Downloads0

Summary

Closed, Completed

Understand
Analyzed test and test script behavior.
Understand
Verify
Confirmed test uses mocks, no real network calls.
Verify
Plan
Suggested options to exclude test from script.
Plan

test/openai-codex-stream.test.ts 1/3 seems to execute when i run test.sh, but shouldn't as it calls an llm endpoint?

Do you want me to inspect test.sh and the test file to confirm why it runs, or just explain how the test selection is supposed to work?

what do you thin i want you to do ...

Confirming test uses fully mocked fetch

The test relies on a globally mocked fetch covering GitHub and Codex calls, so no real network happens despite using real fetch Response objects. It should run safely without contacting actual LLM services unless there's an unmocked code path.

Noting no test gating and offering skip option

The suite runs all three tests unconditionally; skipping tests would require adding tags or environment gating. I'll ask if skipping or gating is desired.

It runs because it is a regular unit test and it fully mocks global.fetch. It does not hit a real LLM endpoint. The only URLs it expects are the GitHub release and raw prompt URL plus https://chatgpt.com/backend-api/codex/responses, and those are all handled by the mock in the test. There is no network call unless something bypasses global.fetch.

If you want it excluded from test.sh, say what gating you want (env flag, test name filter, or move to a different test suite).