-
Notifications
You must be signed in to change notification settings - Fork 4
Open
Labels
enhancementNew feature or requestNew feature or request
Description
Problem
SKILL.md uses a 2-Phase pipeline:
- Phase 1 (5a): structure patterns first
- Phase 2 (5b): sentence/vocabulary patterns
- Phase 3 (5c): self-audit ("does this still sound like AI?")
The MAX mode worker prompt (SKILL-MAX.md:126-154) sends all patterns as a flat list and omits the self-audit step. External models do not apply the phased pipeline or perform quality checks.
Additionally, MAX mode scoring uses subjective evaluation while single mode uses formula-based scoring (core/scoring.md). SKILL-MAX.md acknowledges this divergence but has no reconciliation plan.
Proposed Solution
- Update MAX worker prompt to separate structure-first processing
- Add self-audit instruction to worker prompt
- Document scoring reconciliation plan
Priority
Medium — affects MAX mode output quality
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
enhancementNew feature or requestNew feature or request