wa:evolve

What problem does it solve?

It transforms raw behavioral observations into reusable heuristic rules by clustering similar patterns, scoring their confidence, and promoting user-approved candidates into active guidance.

Core Features & Use Cases

• Observation clustering: Groups heuristic-capture events by tool and normalized context (minimum cluster size with cross-session boosts).
• Confidence scoring & filtering: Computes confidence from occurrences and applies caps plus boosts for cross-session evidence and correction signals.
• Human-in-the-loop promotion: Presents deduplicated candidates for user approval, optional edits, and appends them to the heuristics store.
• Lifecycle hygiene: Prunes promoted observations to prevent reprocessing and regenerates top-N active guidance for the agent.

Quick Start

Run /evolve to analyze collected observations, show heuristic candidates, and promote approved patterns into active guidance.