hermes - 💡(How to fix) Fix [Feature] Quality-gated fallback: autoCurrent fallback only triggers on HTTP errors (429, timeout). No way to auto-escalate when a cheap model (e.g. deepseek-v4-flash) produces poor responses.-escalate to stronger model when weaker one underperforms

Problem

Current fallback only triggers on HTTP errors (429, timeout). No way to auto-escalate when a cheap model (e.g. deepseek-v4-flash) produces poor responses.

Use Case

Typical cost-saving chain:

MiniMax (free) → flash (cheap) → pro (expensive)

User must manually re-ask with pro when flash fails. Breaks autonomy.

Proposed Solution

Add quality_gated config in fallback_providers:

- provider: deepseek
  model: deepseek-v4-flash
  quality_gated:
    enabled: true
    max_consecutive_issues: 3
    signals:
      - self_reported_failure
      - tool_loop_detected
      - user_sentiment_negative

Simpler Alternative

A /pro slash command to switch to deepseek-v4-pro mid-session.