interfluence: Compare Voice Match

You are comparing a piece of text against the user's voice profiles to assess how well it matches.

Process

1. Load available voices using profile_list

2. Load the base profile using profile_get

3. If multiple voices exist, load each delta and merge with base (same H2-section merge as the apply skill)

4. Read the target text (file path or inline)

5. If a file path is provided, check which voice would auto-resolve from config using config_get

6. Analyze the match against each voice:

   • Sentence structure: does it match the voice's patterns?
   • Vocabulary: right register, right word choices?
   • Tone: appropriate relationship to reader?
   • Structure: organized the way the author would in this context?
   • References: right style of cultural/intellectual framing?
   • Anti-patterns: does it violate any "never do this" rules?

7. Present results with scores for all voices:

   Voice match scores:
   - blog: 85% — sentence length and informality align well
   - docs: 60% — too casual for documentation voice
   - base: 72% — core vocabulary matches, structure diverges
   
   Auto-resolved voice for this file path: blog (via pattern 'posts/**')
   

8. Show section-by-section breakdown for the best-matching voice:

   • Highlight the 3 biggest mismatches with suggested fixes
   • Quote specific phrases that feel right and wrong

9. Offer to apply fixes if the user wants (hand off to the apply skill)

Important

• Be specific — don't just say "the tone is off", say exactly which phrases don't match and why
• Acknowledge what already works well, not just what's wrong
• This is diagnostic, not prescriptive — the user decides what to change
• If only base exists (no deltas), compare against base only and skip the multi-voice scoring