Responses API: AssistantReasoningDeltaEvent and AssistantReasoningEvent triggered with empty content via BYOK providers (Copilot CLI)
- Langage dominant
- Shell
- Étoiles
- 11.2k
- Forks
- 1.9k
- Merge moyen
- 14 h 16 min
- PR mergées (30 j)
- 6
Description
### Describe the bug
When using GitHub Copilot CLI (or SDK) with BYOK—either OpenAI or vLLM—for Responses API, both AssistantReasoningDeltaEvent and AssistantReasoningEvent are triggered but with empty content, even if the provider's response includes reasoning content.
### Affected version
1.0.43
### Steps to reproduce the behavior
1. Set up Copilot CLI to use BYOK, OpenAI or vLLM provider (Windows/WSL). Sample .NET code for SDK (works similarly with CLI):
```csharp
await using var session = await client.CreateSessionAsync(new SessionConfig
{
Provider = new ProviderConfig
{
Type = "openai",
BaseUrl = "https://api.openai.com/v1",
ApiKey = "your_api_key",
WireApi = "responses",
},
ReasoningEffort = "high",
Model = "gpt-5.4-2026-03-05",
OnPermissionRequest = PermissionHandler.ApproveAll,
Streaming = true
});
```
2. Trigger completions or calls that should generate reasoning via the Responses API.
3. See that the events are always empty.
### Expected behavior
These events should present the reasoning/explanation content if the provider response includes it. Content field should not be blank if a reasoning field is present.
### Additional context
- Environment: Windows, WSL
- Copilot CLI version: 1.0.43
- Copilot SDK version: 0.3.0
- BYOK provider: OpenAI or vLLM
Guide de contribution
Ouvrir le guide de contribution
Piste de recherche
Aucun fichier d’implémentation ni test n’est indiqué. Reproduisez le problème avec la configuration de Responses API indiquée pour un fournisseur OpenAI ou vLLM BYOK, puis suivez le traitement des événements de reasoning. C’est terminé lorsque AssistantReasoningDeltaEvent et AssistantReasoningEvent exposent le contenu de reasoning du fournisseur au lieu d’un contenu vide.
Rédigé par le modèle d'indexation à partir du texte de l'issue.
Évaluation
- Stack technique
- github, shell
- Domaine
- api, backend
- Type d'issue
- Bug
- Difficulté
- 4/5
- Temps estimé
- 3-5 jours
- Activité
- Calme
- Clarté
- Plutôt claire
- Accessibilité débutants
- 45/100