fix(runtime): harden provider failure handling

5 months ago

1074ce4

Authored

mweinbach5/20/2026, 5:08:21 AM

Preserve OpenAI failure details, normalize reasoning effort sentinels, and tighten Codex app-server usage/file-change projection.

Avoid double-counting provider failure usage and add focused runtime/session coverage.

Parent2ee90f5

15 files changed

src/runtime/codexAppServerRuntime.ts+66−11
src/runtime/googleInteractionsRuntime.ts+27−3
src/runtime/openaiNativeResponses.ts+26−4
src/runtime/openaiResponsesProjector.ts+88−2
src/runtime/openaiResponsesRuntime.ts+56−5
src/runtime/piRuntime.ts+27−2
src/server/session/TurnExecutionManager.ts+37−1
src/shared/openaiContinuation.ts+2−0
src/shared/providerContinuation.ts+27−0
test/runtime.codex-app-server.test.ts+196−23
test/runtime.google-interactions.test.ts+58−0
test/runtime.openai-responses-projector.test.ts+32−0
test/runtime.openai-responses-runtime.test.ts+90−0
test/runtime.pi-runtime.test.ts+70−0
test/session.test.ts+65−0

mweinbach / agent-coworker

fix(runtime): harden provider failure handling