@cloudstreamsoftware/claude-tools 1.0.0 → 1.1.0

package/skills/continuous-learning-v2/SKILL.md

@@ -0,0 +1,126 @@
+---
+name: continuous-learning-v2
+description: Instinct-based learning system that observes sessions, creates atomic instincts with confidence scoring, and evolves them into skills/commands/agents
+version: 2.0.0
+status: active
+introduced: 1.3.0
+lastUpdated: 2026-01-27
+---
+
+# Continuous Learning v2 - Instinct-Based Architecture
+
+## Overview
+
+This system transforms Claude Code sessions into reusable knowledge through atomic "instincts" - small learned behaviors with confidence scoring. The v2 architecture upgrades from the original continuous-learning skill with better pattern detection and evolution capabilities.
+
+## Core Concepts
+
+### Instincts
+Minimal, actionable units featuring:
+- **Single trigger and action pairing** - One behavior per instinct
+- **Confidence weighting** - 0.3 to 0.9 range based on observations
+- **Domain classification** - code-style, testing, workflow, git, etc.
+- **Evidence tracking** - How many times the pattern was observed
+
+### Example Instinct
+
+```yaml
+id: prefer-functional-style
+trigger: "when writing new functions"
+action: "Use functional patterns over classes"
+confidence: 0.8
+domain: code-style
+observations: 12
+source: session-observation
+```
+
+## Workflow
+
+1. **Capture** - PreToolUse/PostToolUse hooks observe session activity
+2. **Analyze** - Background process detects patterns (corrections, workflows, errors)
+3. **Generate** - Creates instincts in personal directory with confidence scores
+4. **Evolve** - Clusters related instincts into commands/skills/agents
+
+## Key Improvements Over v1
+
+| Feature | v1 | v2 |
+|---------|----|----|
+| Observation | Stop hook (unreliable) | PreToolUse/PostToolUse (reliable) |
+| Granularity | Full skills | Atomic instincts |
+| Sharing | Manual | Import/Export commands |
+| Evolution | Manual | Automatic clustering |
+
+## Directory Structure
+
+```
+~/.claude/homunculus/
+  observations.jsonl           # Raw session observations
+  instincts/
+    personal/                  # Learned from your sessions
+    inherited/                 # Imported from teammates
+  evolved/
+    commands/                  # Generated command files
+    skills/                    # Generated skill files
+    agents/                    # Generated agent files
+```
+
+## Commands
+
+| Command | Purpose |
+|---------|---------|
+| `/instinct-status` | View all instincts with confidence scores |
+| `/instinct-export` | Export instincts for sharing |
+| `/instinct-import` | Import instincts from teammates |
+| `/skill-create` | Generate skills from git history |
+| `/evolve` | Cluster instincts into higher structures |
+
+## Pattern Detection
+
+The system detects several pattern types:
+
+- **User corrections** - When you fix something Claude did wrong
+- **Repeated workflows** - Same sequence of actions multiple times
+- **Error resolutions** - How errors were fixed
+- **Code style preferences** - Formatting, naming, structure choices
+
+## Confidence Scoring
+
+| Range | Meaning | Action |
+|-------|---------|--------|
+| 0.3 - 0.5 | Observed once or twice | Track but don't apply |
+| 0.5 - 0.7 | Growing pattern | Suggest when relevant |
+| 0.7 - 0.9 | Established pattern | Auto-apply when triggered |
+| 0.9+ | Core workflow | Always apply |
+
+## Privacy & Sharing
+
+### Local only (not exported):
+- Raw observations
+- File paths
+- Session transcripts
+- Code snippets
+
+### Shareable (via export):
+- Trigger patterns
+- Actions
+- Confidence scores
+- Domain classifications
+
+## CloudStream Integration
+
+Works with existing CloudStream workflows:
+
+1. **Client isolation** - Instincts can be filtered by project/client
+2. **Knowledge capture** - Export high-confidence instincts during handoff
+3. **Team learning** - Import team conventions for consistency
+4. **Consultancy patterns** - Evolve client-specific patterns into reusable skills
+
+## Setup
+
+The system activates automatically when continuous-learning-v2 skill is loaded. Instincts are stored in the user's home directory and persist across sessions.
+
+## Related
+
+- [continuous-learning](../continuous-learning/SKILL.md) - Original v1 skill (still available)
+- [consultancy-workflows](../consultancy-workflows/SKILL.md) - Client knowledge capture
+- `/learn` command - Manual pattern extraction

package/skills/continuous-learning-v2/config.json

@@ -0,0 +1,61 @@
+{
+  "version": "2.0",
+  "description": "Configuration for continuous-learning-v2 instinct system",
+
+  "observation": {
+    "enabled": true,
+    "storage": "~/.claude/homunculus/observations.jsonl",
+    "tracked_tools": ["Edit", "Write", "Bash", "Read", "Grep", "Glob"],
+    "excluded_tools": ["TodoWrite"]
+  },
+
+  "instincts": {
+    "personal_dir": "~/.claude/homunculus/instincts/personal/",
+    "inherited_dir": "~/.claude/homunculus/instincts/inherited/",
+    "min_confidence": 0.3,
+    "auto_approve_threshold": 0.7,
+    "max_instincts": 100,
+    "decay": {
+      "enabled": true,
+      "rate": 0.05,
+      "min_observations_to_decay": 30
+    }
+  },
+
+  "observer": {
+    "enabled": false,
+    "model": "haiku",
+    "interval_minutes": 5,
+    "patterns_to_detect": [
+      "user_corrections",
+      "repeated_workflows",
+      "error_resolutions",
+      "code_style_preferences"
+    ]
+  },
+
+  "evolution": {
+    "enabled": true,
+    "auto_evolve": false,
+    "cluster_threshold": 3,
+    "output_dir": "~/.claude/homunculus/evolved/",
+    "types": ["command", "skill", "agent"]
+  },
+
+  "integration": {
+    "skill_creator_api": "https://api.skillcreator.dev",
+    "backward_compatible_with_v1": true
+  },
+
+  "domains": [
+    "code-style",
+    "testing",
+    "workflow",
+    "git",
+    "architecture",
+    "security",
+    "documentation",
+    "zoho",
+    "compliance"
+  ]
+}