fix: handle Copilot GPT-5 reasoning streams by EntropyParadigm · Pull Request #577 · Nano-Collective/nanocoder

EntropyParadigm · 2026-06-16T12:39:44Z

Description

Fixes #576.

This PR makes GitHub Copilot GPT-5 / Responses streams more tolerant of Copilot's reasoning event ordering and preserves streamed text when the AI SDK final text accessor reports no output.

Changes:

patch @ai-sdk/openai@3.0.71 so Responses reasoning completion / summary events can arrive without a pre-existing activeReasoning entry
accumulate streamed text in Nanocoder's chat handler and use it as the final assistant content fallback
restore compact tool error punctuation expected by existing tests
make file watcher path-filter coverage deterministic via the existing _watchFn test hook
clear shutdown timeout handles and reset directly constructed shutdown managers in tests
add regression tests for both the streamed-text fallback and Copilot-style Responses reasoning events

Type of Change

Bug fix
New feature
Breaking change
Documentation update

Testing

Automated Tests

New features include passing tests in .spec.ts/tsx files
All existing tests pass (pnpm test:all completes successfully)
Tests cover both success and error scenarios

Commands run in a clean WSL-native verification clone:

corepack pnpm run build
corepack pnpm run test:all

pnpm test:all completed successfully with the repo script's Semgrep branch skipped because semgrep is not installed in the local environment:

✅ AVA tests passed
✅ Knip check passed
✅ Audit passed
⚠️  Semgrep not installed - skipping security scan
✅ Everything passes!

Additional focused regression command:

corepack pnpm run test:ava source/events/sources/file-watcher.spec.ts source/utils/shutdown/shutdown-manager.spec.ts source/utils/tool-result-display.spec.tsx source/ai-sdk-client/chat/chat-handler.spec.ts

Manual Testing

Tested with Ollama
Tested with OpenRouter
Tested with OpenAI-compatible API
Tested MCP integration (if applicable)

Manual/local repro validation:

configured GitHub Copilot provider with gpt-5.5
verified nanocoder --provider "GitHub Copilot" --model gpt-5.5 run "reply with exactly ok" returns ok with the local equivalent patch

Checklist

Code follows project style guidelines
Self-review completed
Documentation updated (if needed)
No breaking changes (or clearly documented)
Appropriate logging added using structured logging (see CONTRIBUTING.md)

Copilot

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

This PR improves resilience when the AI SDK fails to provide a final aggregated response, by falling back to streamed text captured from the token stream, and patches @ai-sdk/openai to tolerate certain reasoning/summary SSE sequences.

Changes:

Accumulate streamed text deltas and use them as a fallback when result.text is empty/unavailable.
Add AVA tests for streamed fallback behavior and for OpenAI Responses parser tolerance.
Apply a pnpm patch to @ai-sdk/openai@3.0.71 to avoid crashes on missing reasoning state.

Reviewed changes

Copilot reviewed 4 out of 5 changed files in this pull request and generated 6 comments.

File	Description
source/ai-sdk-client/chat/chat-handler.ts	Captures streamed text and uses it as a fallback for the final response content / empty-output handling.
source/ai-sdk-client/chat/chat-handler.spec.ts	Adds tests for streamed fallback and parser tolerance scenarios.
pnpm-workspace.yaml	Registers a patched dependency for `@ai-sdk/openai@3.0.71`.
patches/@ai-sdk__openai@3.0.71.patch	Patches OpenAI Responses language model parsing to handle missing reasoning state safely.

Files not reviewed (1)

pnpm-lock.yaml: Generated file

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

 					: [];

-			const content = fullText;
+			const content = fullText || accumulatedText;


+						controller.enqueue({type: 'text-start', id: '0'});
+						controller.enqueue({type: 'text-delta', id: '0', delta: 'ok'});
+						controller.enqueue({type: 'text-end', id: '0'});
+						controller.enqueue({
+							type: 'finish',
+							finishReason: 'stop',
+							usage,
+						});


+                const activeReasoningPart = activeReasoning[value.item_id] ??
+                  (activeReasoning[value.item_id] = {
+                    encryptedContent: null,
+                    summaryParts: {},
+                  });


 	// reasoning-aware nudge depends on this for the GPT-5 case where the
 	// SDK throws AI_NoOutputGeneratedError after a reasoning-only stream.
 	let accumulatedReasoning = '';
+	let accumulatedText = '';


 						}
 						break;
 					case 'text-delta':
+						accumulatedText += chunk.text;


 					: [];

-			const content = fullText;
+			const content = fullText || accumulatedText;


EntropyParadigm · 2026-06-16T13:12:56Z

Follow-up pushed in ddd24966 to make the repo's full local gate pass strictly:

restored compact tool error punctuation expected by existing tests
made file watcher path-filter coverage deterministic with _watchFn
cleared shutdown timeout handles and reset directly constructed shutdown managers in tests

Verified in a clean WSL-native clone:

corepack pnpm run build
corepack pnpm run test:all

Result: ✅ Everything passes! (semgrep skipped by scripts/test.sh because it is not installed locally).

will-lamerton · 2026-06-16T14:58:21Z

restored compact tool error punctuation expected by existing tests

made file watcher path-filter coverage deterministic with _watchFn

cleared shutdown timeout handles and reset directly constructed shutdown managers in tests

Hey @EntropyParadigm - really appreciate this PR. Looking good and great to merge! On the stabilisation of the test gates, are you okay to revert this commit? These issues have been addressed on main now and are out of scope for this PR. Thanks :)

EntropyParadigm · 2026-06-16T15:20:48Z

restored compact tool error punctuation expected by existing tests

made file watcher path-filter coverage deterministic with _watchFn

cleared shutdown timeout handles and reset directly constructed shutdown managers in tests

Hey @EntropyParadigm - really appreciate this PR. Looking good and great to merge! On the stabilisation of the test gates, are you okay to revert this commit? These issues have been addressed on main now and are out of scope for this PR. Thanks :)

Of course, my pleasure and totally good to revert the commit. Thank you :)

This reverts commit ddd2496.

will-lamerton · 2026-06-16T15:35:42Z

Thanks for this PR @EntropyParadigm - feel free to add yourself as a contributor to our website via a PR which I will approve :)

https://nanocollective.org/contributors
https://github.com/Nano-Collective/organisation

fix: handle copilot gpt-5 reasoning streams

d212c78

Copilot AI review requested due to automatic review settings June 16, 2026 12:39

EntropyParadigm requested review from Avtrkrb and will-lamerton as code owners June 16, 2026 12:39

Copilot AI reviewed Jun 16, 2026

View reviewed changes

fix: stabilize full test gate

ddd2496

Revert "fix: stabilize full test gate"

7591a93

This reverts commit ddd2496.

will-lamerton merged commit 19d238b into Nano-Collective:main Jun 16, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix: handle Copilot GPT-5 reasoning streams#577

fix: handle Copilot GPT-5 reasoning streams#577
will-lamerton merged 3 commits into
Nano-Collective:mainfrom
EntropyParadigm:fix/copilot-gpt5-reasoning-stream

EntropyParadigm commented Jun 16, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

EntropyParadigm commented Jun 16, 2026

Uh oh!

will-lamerton commented Jun 16, 2026

Uh oh!

EntropyParadigm commented Jun 16, 2026

Uh oh!

will-lamerton commented Jun 16, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

EntropyParadigm commented Jun 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Type of Change

Testing

Automated Tests

Manual Testing

Checklist

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

EntropyParadigm commented Jun 16, 2026

Uh oh!

will-lamerton commented Jun 16, 2026

Uh oh!

EntropyParadigm commented Jun 16, 2026

Uh oh!

will-lamerton commented Jun 16, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

EntropyParadigm commented Jun 16, 2026 •

edited

Loading