Optimize Classifier

# Classifier Optimization Report

## Your Usage Summary

- Total recommendations: N
- Overrides: M (X.X%)
- Ready for optimization: [Yes/No - need 5+ overrides]

## Patterns Found

### You prefer Opus for:

**Keyword: "debug"**
- Frequency: 5 overrides
- Current behavior: Recommends Sonnet
- Proposed: Add "debug" to opus_keywords
- Impact: 5 past overrides would become correct recommendations

**Keyword: "investigate"**
- Frequency: 3 overrides
- Current behavior: Recommends Sonnet
- Proposed: Add "investigate" to opus_keywords
- Impact: 3 past overrides would become correct recommendations

[Repeat for other keywords]

### You prefer Sonnet over Opus for:

[Same structure]

### You prefer Haiku for:

[Same structure]

## Proposed Changes to ~/.cursor/hooks/model-advisor.sh

**Add to opus_keywords (line 46):**
```python
opus_keywords = [
    "architect", "architecture", "evaluate", "tradeoff", "trade-off",
    "strategy", "strategic", "compare approaches", "why does", "deep dive",
    "redesign", "across the codebase", "investor", "multi-system",
    "complex refactor", "analyze", "analysis", "plan mode", "rethink",
    "high-stakes", "critical decision",
    # NEW - Added based on your override patterns:
    "debug", "investigate"  # 8 overrides support this
]

### Step 7: Apply Changes (If User Approves)

If user says "yes" or approves specific keywords:

1. Read `~/.cursor/hooks/model-advisor.sh`
2. Locate the appropriate keyword list (opus_keywords, sonnet_patterns, or haiku_patterns)
3. Add the new keywords to the list with a comment explaining they were auto-added
4. Write the file back using the StrReplace tool
5. Confirm: "Updated! Your classifier now recommends [model] for prompts containing [keywords]. Changes take effect in your next Cursor session."

### Important Notes

- Only suggest keywords with 3+ override occurrences (confidence threshold)
- Don't add generic words that could cause false positives ("the", "make", "update")
- Show the user exactly what will change before changing it
- If the user's overrides are inconsistent (sometimes want Opus, sometimes want Sonnet for the same keyword), note that and ask for clarification

---

## How to Use This Skill

As a user, invoke this skill by saying something like:

- "Optimize my Model Matchmaker classifier"
- "Tune my classifier based on my overrides"
- "Analyze my Model Matchmaker usage and improve it"
- "Update my classifier to match my preferences"

The AI will analyze your logs, find patterns, and propose keyword additions. You review and approve before any changes are made.

## When to Run This

- After your first 50-100 prompts (to establish baseline patterns)
- Monthly or quarterly as your workflow evolves
- After a major project shift (switching from backend to frontend work, for example)
- Whenever you notice you're overriding the advisor frequently