[BUG] Claude Sonnet 4.6 via VS LM API: Provider Error/Model Response Incomplete

[bosconian-dynamics]

Problem (one or two sentences)

Sending any message to Claude Sonnet 4.6 via the VS LM API and GitHub Copilot extensions invariably fails with a "Provider Error" error. The automatic retry fails with both a "Model Response Incomplete" and "Provider Error" error.

Claude Sonnet 4.6 is fully functional via the GitHub Copilot extensions themselves.

Context (who is affected and when)

When any message is sent to Claude Sonnet 4.6 by way of the VS LM API.

Reproduction steps

1. Configure a claude-code/claude-sonnet-4.6 profile with the VS Code LM API provider
2. Select it in any mode and send a message

Expected result

A response from the model.

Actual result

"Provider Error"/"Model Response Incomplete" errors.

Variations tried (optional)

• Inside of an Ubuntu 24.04 devcontainer
• Directly on Windows 11 Pro
• With proxies and AdGuard Home disabled
• Fresh installations of VSCode and RooCode stable on a separate machine

App Version

3.53.0 (44f6bba)

API Provider (optional)

VS Code Language Model API

Model Used (optional)

Claude Sonnet 4.6 (claude-code/claude-sonnet-4.6)

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

Error after initial task message sent:

// Please share this file with Roo Code Support (support@roocode.com) to diagnose the issue faster
// Just make sure you're OK sharing the contents of the conversation below.

{
  "error": {
    "timestamp": "2026-05-01T23:19:25.653Z",
    "version": "3.53.0",
    "provider": "vscode-lm",
    "model": "claude-code/claude-sonnet-4.6",
    "details": "Unexpected API Response: The language model did not provide any assistant messages. This may indicate an issue with the API or the model's output."
  },
  "history": [
    {
      "role": "user",
      "content": [
        {
          "type": "text",
          "text": "<user_message>\nThis is a simple test message. Please respond with \"Hello\"\n</user_message>"
        },
        {
          "type": "text",
          "text": "<environment_details>\n# VSCode Visible Files\n.devcontainer\\docker-compose.yml,X:\\Microsoft VS Code\\extension-output-ms-dotnettools.vscode-dotnet-runtime-#1-.NET Install Tool\n\n# VSCode Open Tabs\n.devcontainer/devcontainer.json,.devcontainer/docker-compose.yml\n\n# Current Time\nCurrent time in ISO 8601 UTC format: 2026-05-01T23:12:18.610Z\nUser time zone: America/Denver, UTC-6:00\n\n# Git Status\n## main...origin/main [behind 9]\n M .devcontainer/devcontainer-lock.json\n M .devcontainer/devcontainer.json\n?? .devcontainer/Dockerfile\n?? .devcontainer/docker-compose.yml\n\n# Current Cost\n$0.00\n\n# Current Mode\n<slug>ask</slug>\n<name>❓ Ask</name>\n<model>claude-sonnet-4.6</model>\nYou have not created a todo list yet. Create one with `update_todo_list` if your task is complicated or involves multiple steps.\n</environment_details>"
        }
      ],
      "ts": 1777677138942
    }
  ]
}

Errors in sequence after RooCode automatically attempts a retry:

// Please share this file with Roo Code Support (support@roocode.com) to diagnose the issue faster
// Just make sure you're OK sharing the contents of the conversation below.

{
  "error": {
    "timestamp": "2026-05-01T23:20:45.895Z",
    "version": "3.53.0",
    "provider": "vscode-lm",
    "model": "claude-code/claude-sonnet-4.6",
    "details": "The language model returned an empty response without any text or tool calls. This can happen due to API issues, rate limiting, or model-specific problems. The model has been automatically prompted to retry."
  },
  "history": [
    {
      "role": "user",
      "content": [
        {
          "type": "text",
          "text": "<user_message>\nThis is a simple test message. Please respond with \"Hello\"\n</user_message>"
        },
        {
          "type": "text",
          "text": "<environment_details>\n# VSCode Visible Files\n.devcontainer\\docker-compose.yml,X:\\Microsoft VS Code\\extension-output-ms-dotnettools.vscode-dotnet-runtime-#1-.NET Install Tool\n\n# VSCode Open Tabs\n.devcontainer/devcontainer.json,.devcontainer/docker-compose.yml\n\n# Current Time\nCurrent time in ISO 8601 UTC format: 2026-05-01T23:12:18.610Z\nUser time zone: America/Denver, UTC-6:00\n\n# Git Status\n## main...origin/main [behind 9]\n M .devcontainer/devcontainer-lock.json\n M .devcontainer/devcontainer.json\n?? .devcontainer/Dockerfile\n?? .devcontainer/docker-compose.yml\n\n# Current Cost\n$0.00\n\n# Current Mode\n<slug>ask</slug>\n<name>❓ Ask</name>\n<model>claude-sonnet-4.6</model>\nYou have not created a todo list yet. Create one with `update_todo_list` if your task is complicated or involves multiple steps.\n</environment_details>"
        }
      ],
      "ts": 1777677138942
    }
  ]
}

// Please share this file with Roo Code Support (support@roocode.com) to diagnose the issue faster
// Just make sure you're OK sharing the contents of the conversation below.

{
  "error": {
    "timestamp": "2026-05-01T23:22:42.572Z",
    "version": "3.53.0",
    "provider": "vscode-lm",
    "model": "claude-code/claude-sonnet-4.6",
    "details": "Unexpected API Response: The language model did not provide any assistant messages. This may indicate an issue with the API or the model's output."
  },
  "history": [
    {
      "role": "user",
      "content": [
        {
          "type": "text",
          "text": "<user_message>\nThis is a simple test message. Please respond with \"Hello\"\n</user_message>"
        },
        {
          "type": "text",
          "text": "<environment_details>\n# VSCode Visible Files\n.devcontainer\\docker-compose.yml,X:\\Microsoft VS Code\\extension-output-ms-dotnettools.vscode-dotnet-runtime-#1-.NET Install Tool\n\n# VSCode Open Tabs\n.devcontainer/devcontainer.json,.devcontainer/docker-compose.yml\n\n# Current Time\nCurrent time in ISO 8601 UTC format: 2026-05-01T23:12:18.610Z\nUser time zone: America/Denver, UTC-6:00\n\n# Git Status\n## main...origin/main [behind 9]\n M .devcontainer/devcontainer-lock.json\n M .devcontainer/devcontainer.json\n?? .devcontainer/Dockerfile\n?? .devcontainer/docker-compose.yml\n\n# Current Cost\n$0.00\n\n# Current Mode\n<slug>ask</slug>\n<name>❓ Ask</name>\n<model>claude-sonnet-4.6</model>\nYou have not created a todo list yet. Create one with `update_todo_list` if your task is complicated or involves multiple steps.\n</environment_details>"
        }
      ],
      "ts": 1777677138942
    }
  ]
}

[roomote-v0]

I've analyzed this issue and here's my proposed implementation plan:

Root Cause Analysis:

The VS Code LM API handler in src/api/providers/vscode-lm.ts (line 382) sends the system prompt as an Assistant message:

const vsCodeLmMessages: vscode.LanguageModelChatMessage[] = [
    vscode.LanguageModelChatMessage.Assistant(systemPrompt),
    ...convertToVsCodeLmMessages(cleanedMessages),
]

The VS Code Language Model API only supports User and Assistant roles (no System role). While sending the system prompt as an Assistant message works for some models (e.g., GitHub Copilot GPT models), the claude-code extension's Claude Sonnet 4.6 model likely interprets this initial Assistant message as a pre-filled response. When the model sees the conversation already starts with its own "response," it may produce an empty reply for the subsequent user message -- leading to the "did not provide any assistant messages" error.

Proposed Fix:

1. Change system prompt delivery in src/api/providers/vscode-lm.ts: Send the system prompt as a User message instead of Assistant, wrapping it with clear system instruction markers so models understand the intent:

   vscode.LanguageModelChatMessage.User(`[SYSTEM INSTRUCTIONS]\n${systemPrompt}\n[END SYSTEM INSTRUCTIONS]`)

2. Add the same fix for completePrompt: Ensure consistency in the completePrompt method as well.

3. Update tests in src/api/providers/__tests__/vscode-lm.spec.ts: Update existing tests to reflect the change from Assistant to User role for the system prompt message, and add a test verifying the system instruction wrapping.

This approach is the least disruptive change and aligns with how other VS Code LM API consumers handle system prompts (since there is no System role in the API). The [SYSTEM INSTRUCTIONS] markers help models distinguish the system prompt from actual user messages.

Please mention @roomote in a comment saying that you approve this plan, and I'll proceed with the implementation.

[bosconian-dynamics]

Update: this may have just come from a misunderstanding, or either Roo or the VS LM API erroneously listing a Claude Code model.

I see now that in addition to claude-code/claude-sonnet-4.6, the LM API provider populates the model list with Copilot-specific provisions:

• copilot/claude-sonnet-4.6
• copilotcli/claude-sonnet-4.6

When I swapped to copilot/claude-sonnet-4.6 it worked without issue.

The OpenRouter model list is also available when the VS LM API provider is selected, which makes some sense to me - I do have the GitHub Copilot extensions configured with my OpenRouter credentials. But I'm not really sure why claude-code/claude-sonnet-4.6 was provided in the model list, as I have never had a Claude Code subscription or configured any tool to use it.

My uninformed suspicion is that claude-code/claude-sonnet-4.6 should not have been in that list. But then again, it sort of appears to me that the <environment details> in the error message might actually contain a response at the end, which would indicate that there was some degree of a successful exchange? I'm not sure - probably just thinking too much for my own good.