mirror of
https://github.com/microsoft/vscode.git
synced 2026-06-29 19:06:00 +01:00
e6967d8b45
Adds an end-to-end smoke test for the codex harness that uses the REAL codex app-server and fakes only the LLM. It reuses the existing Agents-Window smoke harness, which already drives: real Agents-Window UI -> agent host -> real spawned `codex` binary -> real CodexProxyService -> mock LLM server The mock LLM server (scripts/chat-simulation/common/mock-llm-server.ts) already speaks codex's OpenAI `/responses` API and ships a `gpt-5.3-codex` `/responses` chat-default model, so no LLM-side changes are needed. - test/smoke/.../agentsWindow.test.ts: new `Agents Window (Codex)` suite via the existing setupAgentHostSuite helper with `chat.agentHost.codexAgent.enabled` (codex is off by default). The test selects the `Codex` session type, submits a scenario prompt, and asserts the mocked reply plus a `chat/turnStarted` frame in the AHP transcript (proving the codex harness, not a renderer fallback, served it). Adds a warmUpCodexModel helper mirroring warmUpClaudeModel. - test/automation/src/agentsWindow.ts: new isSessionTypeAvailable(label) used to skip gracefully when codex isn't registered. Codex only registers when its SDK is resolvable (product.agentSdks.codex in packaged builds, or VSCODE_AGENT_HOST_CODEX_SDK_ROOT in dev); when absent the test skips rather than fails. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
VS Code Tests
Contents
This folder contains the various test runners for VS Code. Please refer to the documentation within for how to run them: