Claude Code with DeepSeek Anthropic-compatible API fails during project analysis: Invalid user_id

## Description

I encountered a compatibility issue when using Claude Code with DeepSeek's Anthropic-compatible API endpoint.

Claude Code can start successfully, and it works in a clean empty directory. However, when I run it inside a real project directory and ask it to read or analyse the codebase, it starts by searching/listing files and then fails during a subsequent API request with the following error:

```text
API Error: 400 Invalid 'user_id': string does not match pattern. Expected a string that matches the pattern '^[a-zA-Z0-9_-]+$'.
```

Environment
Tool: Claude Code
API endpoint: https://api.deepseek.com/anthropic
Model: deepseek-v4-pro[1m]
OS: macOS
Usage scenario: Claude Code project-level code analysis

My environment variables are configured as follows:

export ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic
export ANTHROPIC_AUTH_TOKEN=sk-xxxx
export ANTHROPIC_MODEL='deepseek-v4-pro[1m]'
export ANTHROPIC_DEFAULT_OPUS_MODEL='deepseek-v4-pro[1m]'
export ANTHROPIC_DEFAULT_SONNET_MODEL='deepseek-v4-pro[1m]'
export ANTHROPIC_DEFAULT_HAIKU_MODEL=deepseek-v4-flash

I also previously tested with:

export CLAUDE_CODE_SUBAGENT_MODEL=deepseek-v4-flash
export CLAUDE_CODE_EFFORT_LEVEL=max

But the issue still appears after disabling these two variables.

Steps to reproduce
Configure Claude Code to use DeepSeek's Anthropic-compatible endpoint.
Run Claude Code in a clean empty directory.
It starts successfully.
Run Claude Code inside an actual code project.
Ask Claude Code to read or analyse the project code, for example:
Please read this codebase and explain how it works.
Claude Code starts file search/listing operations.
The request then fails with:
API Error: 400 Invalid 'user_id': string does not match pattern. Expected a string that matches the pattern '^[a-zA-Z0-9_-]+$'.
What I have already tried

I have tried the following:

Cleared global Claude Code configuration and cache.
Removed local project-level .claude configuration.
Removed .mcp.json and mcp-server.json.
Disabled Claude Code plugins.
Disabled MCP-related configuration.
Removed Claude Code backups and old project history.
Tested in a clean directory, where Claude Code can start successfully.
Tested again in the real project directory, where the error happens during codebase analysis.

Because the clean directory works but project-level analysis fails, it seems that the issue may be triggered by the request generated during Claude Code's tool/codebase workflow.

Expected behaviour

Claude Code should be able to continue project-level code analysis through DeepSeek's Anthropic-compatible endpoint without the Invalid user_id error.

Actual behaviour

Claude Code starts normally, performs some file search/listing operations, and then the API request fails with a 400 error:

Invalid 'user_id': string does not match pattern.
Possible cause

This may be related to how the DeepSeek Anthropic-compatible endpoint handles Claude Code requests involving tool calls, project context, session metadata, or user identification fields.

The error message mentions user_id, but I could not find any invalid user_id in my local Claude Code configuration. It is possible that some field generated by Claude Code during project analysis is being interpreted or validated as user_id by the compatibility layer.

Related reference

There is a similar Claude Code issue where invalid tool/MCP names caused a pattern validation error:

[https://github.com/anthropics/claude-code/issues/858](https://github.com/anthropics/claude-code/issues/858?utm_source=chatgpt.com)

Although my error is shown as user_id, the behaviour seems similar in that a field generated during Claude Code execution may not match the expected pattern.

Request

Could you please confirm whether DeepSeek's Anthropic-compatible API fully supports the current Claude Code request format, especially during project-level code analysis involving tool calls?

If this is a known compatibility issue, could you provide a recommended workaround, such as:

a specific Claude Code version,
a recommended DeepSeek model setting,
environment variables to disable certain Claude Code features,
or guidance on which request fields need to be sanitised?

Thank you.