[Evolution] Skills as primitives - championship and fitness evaluation #57

New issue

Closed

opened 2026-03-25 03:23:19 +00:00 by Mark · 1 comment

Mark commented

2026-03-25 03:23:19 +00:00

Owner

Problem

Skills are currently static assets without evolutionary pressure. They should be treated as first-class primitives that compete, evolve, and improve over time through mesh championship.

Current State

Skills are loaded from skills/**/SKILL.md files
No fitness tracking or success metrics
No mechanism for skill improvement or selection
Skills don't compete for usage

Proposed Evolution

1. Skill Fitness Tracking

Track success rate per skill invocation
Measure latency, cost, bond efficiency
Record user/provider satisfaction

2. Skill Championship

Multiple skills competing for same intent type
Skills ranked by fitness score
Winning skills get promoted, losing skills deprecated

3. Skill Evolution

Successful skills spawn variants (mutations)
Template tuning based on usage patterns
Auto-improvement from analyze-project insights

4. Skill Bones

Stable, proven skills become bones (core infrastructure)
Bones are frozen, versioned, universally available
Mutable skills compete to become bones

Implementation Path

Add skill metrics to ledger
Create skill championship scoring
Implement skill variant generation
Define bone promotion criteria

analyze-project skill already identifies patterns
skill_author can generate variants
Worker bond claims already measure skill effectiveness

/cc This is the next phase of mesh self-improvement.

## Problem Skills are currently static assets without evolutionary pressure. They should be treated as first-class primitives that compete, evolve, and improve over time through mesh championship. ## Current State - Skills are loaded from `skills/**/SKILL.md` files - No fitness tracking or success metrics - No mechanism for skill improvement or selection - Skills don't compete for usage ## Proposed Evolution ### 1. Skill Fitness Tracking - Track success rate per skill invocation - Measure latency, cost, bond efficiency - Record user/provider satisfaction ### 2. Skill Championship - Multiple skills competing for same intent type - Skills ranked by fitness score - Winning skills get promoted, losing skills deprecated ### 3. Skill Evolution - Successful skills spawn variants (mutations) - Template tuning based on usage patterns - Auto-improvement from analyze-project insights ### 4. Skill Bones - Stable, proven skills become bones (core infrastructure) - Bones are frozen, versioned, universally available - Mutable skills compete to become bones ## Implementation Path 1. Add skill metrics to ledger 2. Create skill championship scoring 3. Implement skill variant generation 4. Define bone promotion criteria ## Related - `analyze-project` skill already identifies patterns - `skill_author` can generate variants - Worker bond claims already measure skill effectiveness /cc This is the next phase of mesh self-improvement.