instinct-system

What problem does it solve?

The instinct system captures, tracks, and validates recurring project-specific patterns so the assistant stops guessing and starts following real conventions. It prevents one-off observations from becoming incorrect rules and reduces manual configuration by learning conventions through active confirmation.

Core Features & Use Cases

• Confidence-scored hypotheses: Create instincts at low confidence and raise or lower them through observed confirmations and contradictions.
• Observe-hypothesize-confirm lifecycle: Actively seek evidence across related files, adjust scores, flag stale items, and remove dead instincts.
• Promotion to permanent memory: Promote well-supported instincts to MEMORY.md and export/import with decay for cautious sharing between projects.
• Use case: During feature scaffolding, detect that handlers return Result<T> across the codebase, confirm across handlers, then apply the convention automatically once confidence reaches the follow threshold.

Quick Start

Load the instinct system and scan a few related files to form or update instincts and adjust their confidence levels.