continuous-learning

name	continuous-learning
description	Instinct-based learning system that observes sessions via hooks, creates atomic instincts with confidence scoring, and evolves them into skills, commands, and agents.
version	0.1.0

Continuous Learning

Turns Claude Code sessions into reusable knowledge through atomic instincts — small learned behaviors with confidence scoring that evolve over time into skills, commands, and agents.

The instinct model

An instinct is one trigger mapped to one action:

---
id: prefer-functional-style
trigger: "when writing new functions"
confidence: 0.7
domain: "code-style"
source: "session-observation"
---

# Prefer Functional Style

## Action
Use functional patterns over classes when appropriate.

## Evidence
- Observed 5 instances of functional pattern preference
- User corrected class-based approach to functional on 2025-01-15

Instincts are atomic, confidence-weighted (0.3–0.9), domain-tagged, and evidence-backed.

How it works

Session activity
  → hooks capture tool use (100% reliable, unlike probabilistic skills)
  → observations.jsonl
  → observer agent detects patterns (background, lightweight model)
  → instincts/personal/   (atomic learned behaviors)
  → /evolve clusters instincts
  → evolved/ commands, skills, agents

Setup

1. Enable observation hooks

Add to ~/.claude/settings.json:

{
  "hooks": {
    "PreToolUse": [{
      "matcher": "*",
      "hooks": [{ "type": "command", "command": "${CLAUDE_PLUGIN_ROOT}/skills/continuous-learning/hooks/observe.sh pre" }]
    }],
    "PostToolUse": [{
      "matcher": "*",
      "hooks": [{ "type": "command", "command": "${CLAUDE_PLUGIN_ROOT}/skills/continuous-learning/hooks/observe.sh post" }]
    }]
  }
}

If installed manually to ~/.claude/skills, replace ${CLAUDE_PLUGIN_ROOT}/skills/continuous-learning with ~/.claude/skills/continuous-learning.

2. Initialize directories

mkdir -p ~/.claude/homunculus/{instincts/{personal,inherited},evolved/{agents,skills,commands}}
touch ~/.claude/homunculus/observations.jsonl

3. Commands

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

Configuration

config.json:

{
  "version": "2.0",
  "observation": {
    "enabled": true,
    "store_path": "~/.claude/homunculus/observations.jsonl",
    "max_file_size_mb": 10,
    "archive_after_days": 7
  },
  "instincts": {
    "personal_path": "~/.claude/homunculus/instincts/personal/",
    "inherited_path": "~/.claude/homunculus/instincts/inherited/",
    "min_confidence": 0.3,
    "auto_approve_threshold": 0.7,
    "confidence_decay_rate": 0.05
  },
  "observer": {
    "enabled": true,
    "model": "haiku",
    "run_interval_minutes": 5,
    "patterns_to_detect": ["user_corrections", "error_resolutions", "repeated_workflows", "tool_preferences"]
  },
  "evolution": {
    "cluster_threshold": 3,
    "evolved_path": "~/.claude/homunculus/evolved/"
  }
}

File structure

~/.claude/homunculus/
├── identity.json
├── observations.jsonl
├── observations.archive/
├── instincts/
│   ├── personal/
│   └── inherited/
└── evolved/
    ├── agents/
    ├── skills/
    └── commands/

Confidence scoring

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

Confidence increases when a pattern is repeatedly observed or goes uncorrected. Confidence decreases when the user corrects the behavior or contradicting evidence appears.

Privacy

Observations stay local. Only instincts (patterns) can be exported — no code or conversation content is included. You control what gets shared.