continuous-learning-v2

name	continuous-learning-v2
description	Instinct-based learning system that observes sessions via hooks, creates atomic instincts with confidence scoring, and evolves them into skills/commands/agents.

Continuous Learning v2 — Instinct-Based Architecture

An advanced learning system that turns coding sessions into reusable knowledge through atomic "instincts" — small learned behaviors with confidence scoring.

What Changed from v1

Feature	v1	v2
Observation	sessionEnd hook (session end)	preToolUse/postToolUse (continuous)
Analysis	Main context	Background analysis
Granularity	Full skills	Atomic "instincts"
Confidence	None	0.3–0.9 weighted
Evolution	Direct to skill	Instincts cluster into skills/commands/agents

The Instinct Model

An instinct is a small learned behavior:

---
id: prefer-immutable-patterns
trigger: "when writing new C# DTOs or models"
confidence: 0.7
domain: "code-style"
source: "session-observation"
---

# Prefer Immutable Patterns

## Action
Use records or init-only setters instead of mutable properties.

## Evidence
- Observed 5 instances of immutable pattern preference
- User corrected mutable DTO to record on 2025-01-15

Properties:

• Atomic — one trigger, one action
• Confidence-weighted — 0.3 = tentative, 0.9 = near certain
• Domain-tagged — code-style, testing, git, debugging, workflow, etc.
• Evidence-backed — tracks what observations created it

How It Works

Session Activity
      │
      │ Hooks capture prompts + tool use
      ▼
┌─────────────────────────────────────────┐
│         observations.jsonl              │
│   (prompts, tool calls, outcomes)       │
└─────────────────────────────────────────┘
      │
      │ Observer reads (background)
      ▼
┌─────────────────────────────────────────┐
│          PATTERN DETECTION              │
│   - User corrections -> instinct        │
│   - Error resolutions -> instinct       │
│   - Repeated workflows -> instinct      │
└─────────────────────────────────────────┘
      │
      │ Creates/updates
      ▼
┌─────────────────────────────────────────┐
│         instincts/personal/             │
│   - prefer-immutable.md (0.7)           │
│   - always-test-first.md (0.9)          │
│   - use-data-annotations.md (0.6)       │
└─────────────────────────────────────────┘
      │
      │ Clustering and evolution
      ▼
┌─────────────────────────────────────────┐
│              evolved/                   │
│   - commands/new-feature.md             │
│   - skills/testing-workflow.md          │
│   - agents/refactor-specialist.md       │
└─────────────────────────────────────────┘

Hook Setup

Add to your .github/hooks/default.json:

{
  "version": 1,
  "hooks": {
    "preToolUse": [{
      "type": "command",
      "bash": "scripts/hooks/observe.sh pre"
    }],
    "postToolUse": [{
      "type": "command",
      "bash": "scripts/hooks/observe.sh post"
    }]
  }
}

Directory Structure

~/.copilot/homunculus/
├── observations.jsonl      # Current session observations
├── observations.archive/   # Processed observations
├── instincts/
│   ├── personal/           # Auto-learned instincts
│   └── inherited/          # Imported from others
└── evolved/
    ├── skills/             # Generated skills
    ├── commands/           # Generated commands
    └── agents/             # Generated specialist agents

Observer Script

The start-observer.sh script in this skill's directory runs as a background process analyzing observations.jsonl for patterns:

# Run once
./start-observer.sh --once

# Run continuously (checks every 5 minutes)
./start-observer.sh

It detects:

• Repeated identical tool calls (token waste)
• Sequential reads that could be parallelized
• Verbose commands without output limiting
• Failed operations to learn from

Commands

Command	Description
instinct-status	Show all learned instincts with confidence
evolve	Cluster related instincts into skills/commands
instinct-export	Export instincts for sharing
instinct-import <file>	Import instincts from others

Confidence Scoring

Score	Meaning	Behavior
0.3	Tentative	Suggested but not enforced
0.5	Moderate	Applied when relevant
0.7	Strong	Auto-approved for application
0.9	Near-certain	Core behavior

Increases when: pattern repeatedly observed, user doesn't correct, other sources agree. Decreases when: user explicitly corrects, pattern not seen for extended periods, contradicting evidence appears.

Privacy

• Observations stay local on your machine
• Only instincts (patterns) can be exported — no code or conversation content
• You control what gets exported