mirror of
https://github.com/Significant-Gravitas/AutoGPT.git
synced 2026-02-08 22:05:08 -05:00
2e4ec66730
When _generate_llm_continuation() is called after long-running tools (like create_agent) complete, it was making LLM calls without the tools schema. Some providers (especially Anthropic via OpenRouter) require the tools schema when conversation history contains tool_use/tool_result blocks, causing validation errors and unresponsive chats. Changes: - Add tools=tools with tool_choice="none" to allow providers to validate tool interactions without making new tool calls - Add retry logic with exponential backoff (matching _stream_chat_chunks) - Improve error logging for retry exhaustion Fixes chat becoming unresponsive after create_agent tool execution.