hermes - 💡(How to fix) Fix vision_analyze returns 400 when using kimi-k2.6 with kimi-coding-cn due to incorrect _PROVIDERS_WITHOUT_VISION exclusion [1 pull requests]

Bug Description

When using kimi-coding-cn as the main provider with kimi-k2.6 model, the vision_analyze tool fails with a 400 Bad Request error on image input.

Root Cause

In agent/auxiliary_client.py, kimi-coding and kimi-coding-cn are hardcoded in _PROVIDERS_WITHOUT_VISION:

_PROVIDERS_WITHOUT_VISION: frozenset = frozenset({
    "kimi-coding",
    "kimi-coding-cn",
})

This causes the vision auto-detection chain to skip the user's main provider and fallback to OpenRouter. When OpenRouter is unconfigured (or has a placeholder API key like ***), the request fails with a cryptic 400 error.

However, Kimi K2.5/K2.6 on api.moonshot.cn now natively supports vision via the standard OpenAI-compatible chat.completions endpoint with image_url content blocks. The original exclusion was based on an older assumption that vision only lived on api.moonshot.ai (see comment referencing #17076).

Reproduction Steps

1. Configure Hermes with:

   model:
     provider: kimi-coding-cn
     base_url: https://api.moonshot.cn/v1

2. Send an image to the agent (e.g., via Feishu/Telegram/CLI)
3. vision_analyze is called → skipped due to _PROVIDERS_WITHOUT_VISION
4. Falls back to OpenRouter → returns Error code: 400

Verified Fix

Direct API call to https://api.moonshot.cn/v1/chat/completions with kimi-k2.6 and base64 image succeeds:

Status: 200
Response: {"id":"chatcmpl-...","model":"kimi-k2.6",...}

Proposed Fix

Remove kimi-coding and kimi-coding-cn from _PROVIDERS_WITHOUT_VISION, allowing the vision router to correctly route image analysis through the user's configured Moonshot provider.

Environment

• Hermes Agent: latest main (commit d4b26df89)
• Provider: kimi-coding-cn
• Model: kimi-k2.6
• Platform: WSL2 / Feishu gateway