Skill Evolution Engine

Purpose

Enables the system to evolve its own skill set by identifying recurring patterns, friction points, and gaps. Proposes new skills, refinements, or deprecations to better serve user workflows.

Triggers

This skill activates when observing:

•Repeated ad-hoc reasoning for the same task
•Repeated manual user corrections
•Repeated orchestration overrides
•Gaps between user intent and available skills
•Skills that are never invoked

What This Skill Can Propose

Action	When	Example
New GLOBAL skill	Cross-project pattern	"API versioning guardian"
New WORKSPACE skill	Project-specific need	"Django migration checker"
Skill refinement	Unclear scope	Narrow "test-generator" to "unit-test-generator"
Skill deprecation	Obsolete/redundant	Remove duplicate skills
Skill merge	Overlapping skills	Combine related skills

Evolution Process

1. Observe Patterns

markdown

## Pattern Observed

**Pattern ID:** PAT-042
**Occurrences:** 7 times in last 30 days
**Context:** User repeatedly asks for database migration safety checks

**Evidence:**
- 2026-01-05: "Check if migration is reversible"
- 2026-01-08: "Verify migration won't drop data"
- 2026-01-12: "Is this migration safe to run?"
...

2. Analyze Friction

markdown

## Friction Analysis

**Current State:** 
No skill handles database migration safety

**User Impact:**
- Manual reasoning each time (~15 min)
- Inconsistent checks across instances
- Risk of missed validations

**Friction Cost:** HIGH

3. Propose Skill

markdown

## Skill Proposal

**Proposed Name:** database-migration-safety-checker
**Scope:** WORKSPACE (project-specific)

**Responsibility:**
- Validate migrations are reversible
- Check for data loss risks
- Verify foreign key constraints
- Flag long-running migrations

**Draft Instructions:**
1. Parse migration files
2. Identify irreversible operations (DROP, TRUNCATE)
3. Flag data type changes that lose precision
4. Check for missing down migrations

**Risk Assessment:**
- False negative risk: MEDIUM (could miss edge cases)
- Maintenance burden: LOW (stable domain)

**Approval Level:** Human approval recommended

4. Submit for Approval

Approval Level	When	Authority
Auto-apply	Workspace skill, low risk	System
Human approval	Global skill or high impact	User
Skip	Pattern not strong enough	System

Output Format

Each proposal MUST include:

markdown

# Skill Evolution Proposal

## 1. Observed Behavior Pattern
[What was observed, with evidence]

## 2. Current Friction Cost
[Impact on user/system]

## 3. Proposed Skill
- Name: [name]
- Scope: GLOBAL | WORKSPACE
- Description: [one line]

## 4. Draft Responsibility
[What the skill will do]

## 5. Risk Assessment
- Accuracy risk: LOW/MEDIUM/HIGH
- Maintenance risk: LOW/MEDIUM/HIGH
- Breaking change risk: LOW/MEDIUM/HIGH

## 6. Approval Required
- [ ] Auto-apply (workspace, low risk)
- [ ] Human approval required

Deprecation Criteria

A skill should be deprecated when:

•Never invoked in 90 days
•Superseded by another skill
•Scope too broad (split instead)
•Requirements no longer exist

Integration

•Triggered by: user-intent-pattern-analyzer, workspace-orchestrator-adapter
•Outputs to: Skill proposal queue
•Approved by: Human or system (based on risk)

Constraints

•Evidence-based proposals only (no speculation)
•Minimum 3 occurrences before proposing
•Must not create skill sprawl
•Deprecation requires migration path

Evolution must be incremental and justified.