- Copilot Chat Extension Version: 0.62.0
- VS Code Version: 1.134.0
- OS Version: Windows 11 Insider Preview Build 28020.2731
- Remote: FedoraLinux-44 (6.18.33.2-microsoft-standard-WSL2)
- Feature: byok
- Selected model: DeepSeek V4 Flash
- Logs:
2026-08-19 20:12:52.336 [info] [VoiceProgress] loop request=request_c033bd51-ed53-44ed-b856-3e4e4c6199de configured=true voice=false subagent=false stream=true
2026-08-19 20:12:52.405 [info] [AutomaticInstructionsCollector] Core vs extension instructions mismatch:
--- diff ---
-
+AGENTS.md
index: -undefined +defined
2026-08-19 20:12:52.744 [info] Request ID for failed request: 5a48b7e9-f495-416e-8b6b-9658c014fc35
2026-08-19 20:12:52.748 [error] Server error: 400 {"error":{"message":"No tool call found for tool output with call_id call_00_fy35KjjSqT8H3PKMTUOM5164.","type":"invalid_request_error","param":null,"code":"invalid_request_error"}}
2026-08-19 20:12:52.748 [error] Request Failed: 400 {"error":{"message":"No tool call found for tool output with call_id call_00_fy35KjjSqT8H3PKMTUOM5164.","type":"invalid_request_error","param":null,"code":"invalid_request_error"}}
2026-08-19 20:12:52.755 [info] ccreq:1af54535.copilotmd | failed | deepseek-v4-flash | 125ms | [copilotLanguageModelWrapper]
2026-08-19 20:12:52.802 [info] [ToolCallingLoop] Stop hook result: shouldContinue=false, reasons=undefinedSteps to Reproduce:
- Add a custom endpoint for DeepSeek V4 Flash in Copilot Chat, select the "Responses" API;
- Create a new agent session and ask it to perform an action requiring a tool call (for example, read a file from workspace).
Related to #330610.
This issue appears after #330610 was fixed. Previously the invalid tool schema error prevented the request from reaching the tool calling stage, so this failure was masked.
LLM invocation without any tool call works correctly now. The failure only happens when using tools with the Responses API.
Workaround:
Still use chat-completions API.
Steps to Reproduce:
Related to #330610.
This issue appears after #330610 was fixed. Previously the invalid tool schema error prevented the request from reaching the tool calling stage, so this failure was masked.
LLM invocation without any tool call works correctly now. The failure only happens when using tools with the Responses API.
Workaround:
Still use chat-completions API.