Summary

Add MiniMax AI as a third LLM provider in the MimiClaw embedded AI agent, alongside OpenAI and Anthropic.

MiniMax offers an OpenAI-compatible API, so this change introduces a provider_uses_openai_format() helper to share the existing OpenAI message format, Bearer auth, and response parsing path, while routing requests to MiniMax's own endpoint (api.minimax.io/v1/chat/completions).

Changes

• llm_proxy.c: Add provider_is_minimax(), provider_uses_openai_format(), separate endpoint/cert cache for MiniMax, three-way provider routing in endpoint URL, cert management, and HTTP auth
• mimi_config.h: Add MIMI_MINIMAX_API_URL with #ifndef guard
• serial_cli.c: Update CLI help text (anthropic|openai|minimax) and default URL routing
• mimi_secrets.h.example: Document MiniMax as a provider option
• README.md / README_zh.md: Add MiniMax to provider documentation with model names
• tests/test_minimax_provider.py: 24 unit tests (source-level validation) + 3 integration tests (real API calls)

Usage

Available models: (latest, 1M context), (204K context)

Design

MiniMax uses the same OpenAI-compatible format for:

• Message structure (system/user/assistant roles)
• Bearer token authentication
• Response format (choices[].message.content)
• Tool calling (function type)

This means all existing OpenAI format code paths are reused via provider_uses_openai_format(), keeping the diff minimal. Only endpoint routing and TLS cert caching needed three-way provider dispatch.

Test Plan

• 24 unit tests validate C source-level correctness (provider routing, config, CLI, docs)
• 3 integration tests verify real MiniMax API compatibility (chat, tool calling, auth)
• All 27 tests passing