hermes - 💡(How to fix) Fix [Feature Request] Vision tool bypass in WSL — request for native solution

Problem

In WSL Ubuntu, the built-in vision_analyze tool routes through the auxiliary model pipeline (DeepSeek vision), which performs OCR-level analysis and fails on complex financial screenshots.

The recommended workaround is to use mmx vision describe via the MiniMax CLI. However, this requires manual routing — the model does not default to mmx for image analysis.

Current workaround in AGENTS.md A rule has been added to AGENTS.md to force mmx vision describe for all image tasks, but this is a user-level config hack, not a native solution.

Request Consider adding a native bypass in WSL environments that routes vision_analyze calls directly to mmx vision describe without requiring user configuration. This could be detected automatically via environment variables (WSL_DISTRO_NAME, etc.) or added as a configurable option in config.yaml.

Why this matters

• vision_analyze fails on complex UI screenshots (financial data, tables)
• mmx vision describe produces significantly better analysis for these use cases
• Users on WSL should not need to manually configure AGENTS.md rules to get correct behavior

Environment

• WSL2 Ubuntu
• Hermes Agent latest
• MiniMax CLI installed and authenticated