mirror of
https://github.com/langgenius/dify-official-plugins.git
synced 2026-07-22 01:55:27 -04:00
Error in agent node when using DeepSeek V3.2(volcengine_maas) thinking mode #888
Reference in New Issue
Block a user
Delete Branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @leslie2046 on GitHub (Dec 19, 2025).
Self Checks
Dify version
1.11.1
Plugin version
0.0.37
Cloud or Self Hosted
Self Hosted (Docker)
Steps to reproduce
Add a tool for agent node.eg. current time
✔️ Error log
Run failed: Failed to transform agent message: req_id: 54c9507361 PluginInvokeError: {"args":{},"error_type":"Exception","message":"read llm model failed: request failed: req_id: ce17348edd PluginInvokeError: {"args":{},"error_type":"ArkBadRequestError","message":"Error code: 400 - {'error': {'code': 'InvalidParameter', 'message': \"ThinkingMode: thinking, invalid message without reasoning_content/tool_calls
{'tool_calls': [], 'content': '\\u003cthink\\u003e\\\\\\\\\\\\\\\\n用户询问当前时间。我需要使用current_time工具来获取当前时间。这个工具不需要任何参数。我会调用它。\\\\\\\\\\\\\\\\n\\u003c/think\\u003e', 'role': 'assistant'}after last user message Request id: 0217661505518124fe949375ac26682761f342713cef3321c4e2b\", 'param': '', 'type': 'BadRequest'}}, request_id: 20251219212231of1zHtgdX4Y4pHdDlOPl"}"}@dosubot[bot] commented on GitHub (Dec 19, 2025):
This is a known issue with DeepSeek V3.2 in thinking mode: the plugin sometimes fails to include the required reasoning_content and tool_calls fields in the agent message, which triggers the PluginInvokeError you’re seeing. The root cause is usually a bug in the plugin’s streamed output handling—specifically, using string replace instead of appending new content can cause missing or incomplete fields in the response.
If you’re self-hosted, a common workaround is to patch the plugin so that streamed output appends new content directly (e.g.,
full_text += resp_content[len(full_text):]) instead of using string replace. This prevents character loss and empty output. Other things to try: increase the max_tokens parameter, reduce prompt size, or split large knowledge base chunks. If direct API calls to DeepSeek V3.2 with thinking mode enabled return both reasoning_content and content, the issue is almost certainly in the plugin’s output handling logic, not the model itself. Keep an eye out for plugin updates, as this area is under active development. If you want help locating the code section to patch, let me know your deployment details and I can point you to it.See details and workaround
To reply, just mention @dosu.
How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other