ci(deps): Bump softprops/action-gh-release from 2 to 3

.claude/4-MODEL-SYSTEM.md

@@ -0,0 +1,312 @@
+# 4-Model Code Review System with Gemini
+
+## Architecture
+
+```
+┌─────────────────────────────────────────────────────────────┐
+│                    ISSUE DETECTION                          │
+│  (Shell, Python, Java scans via existing scripts)          │
+└────────────────────┬────────────────────────────────────────┘
+                     │
+                     ▼
+┌─────────────────────────────────────────────────────────────┐
+│              4 WORKER AIs GENERATE FIXES                    │
+│                                                             │
+│  ┌──────────┐  ┌──────────┐  ┌──────────┐  ┌──────────┐  │
+│  │  OPUS    │  │ SONNET   │  │  HAIKU   │  │ GEMINI   │  │
+│  │  Fix A   │  │  Fix B   │  │  Fix C   │  │  Fix D   │  │
+│  └──────────┘  └──────────┘  └──────────┘  └──────────┘  │
+│                                                             │
+└────────────────────┬────────────────────────────────────────┘
+                     │
+                     ▼
+┌─────────────────────────────────────────────────────────────┐
+│           3 ARBITER AIs VOTE ON BEST FIX                    │
+│                                                             │
+│  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐    │
+│  │ Opus Arbiter │  │Sonnet Arbiter│  │Gemini Arbiter│    │
+│  │ Chooses: B   │  │ Chooses: B   │  │ Chooses: A   │    │
+│  └──────────────┘  └──────────────┘  └──────────────┘    │
+│                                                             │
+│         MAJORITY VOTE → Fix B (Sonnet) WINS               │
+└────────────────────┬────────────────────────────────────────┘
+                     │
+                     ▼
+┌─────────────────────────────────────────────────────────────┐
+│              GITHUB ISSUE CREATED                           │
+│                                                             │
+│  - Shows all 4 proposed fixes                              │
+│  - Shows 3 arbiter opinions                                │
+│  - Shows vote breakdown                                    │
+│  - Shows final selection with reasoning                    │
+└─────────────────────────────────────────────────────────────┘
+```
+
+## Components
+
+### 1. Worker AIs (4 models)
+
+Each issue receives **4 independent fix proposals**:
+
+- **Opus** (Claude): Highest capability, complex reasoning
+- **Sonnet** (Claude): Balanced quality/speed
+- **Haiku** (Claude): Fast, efficient
+- **Gemini** (Google): Different training, different perspective
+
+Each worker provides:
+
+- Fix approach
+- Complete code changes
+- Rationale for approach
+- Confidence level (0-100%)
+- Potential risks
+
+### 2. Arbiter Panel (3 models)
+
+Three independent arbiters **vote** on best fix:
+
+- **Opus Arbiter**: Evaluates all 4 fixes, selects best
+- **Sonnet Arbiter**: Independent evaluation, selects best
+- **Gemini Arbiter**: External perspective, selects best
+
+**Decision Method**: Majority vote
+
+- If 2+ arbiters agree → that fix wins
+- If 3-way split → highest confidence fix wins
+- Ties broken by confidence scores
+
+### 3. Gemini Integration
+
+**File**: `.claude/gemini_client.py`
+
+Python client that calls Google Gemini API:
+
+```python
+call_gemini(prompt, schema=None)
+```
+
+**Environment Variable Required**:
+
+```bash
+export GEMINI_API_KEY="your-api-key"
+```
+
+**Model Used**: `gemini-1.5-pro-latest` (configurable via `GEMINI_MODEL`)
+
+## GitHub Issue Format
+
+Each issue shows complete transparency:
+
+```markdown
+## Issue
+[Problem description]
+
+## 4-Model Fix Proposals
+
+### OPUS
+- Approach: [description]
+- Confidence: 85%
+
+### SONNET  
+- Approach: [description]
+- Confidence: 90%
+
+### HAIKU
+- Approach: [description]
+- Confidence: 75%
+
+### GEMINI
+- Approach: [description]
+- Confidence: 88%
+
+## Multi-Model Arbiter Decision
+
+**Final Selection**: SONNET
+
+**Arbiter Vote Breakdown**:
+- sonnet: 2 votes
+- opus: 1 vote
+
+### Arbiter Opinions:
+
+**Opus Arbiter**: Selected sonnet
+> Sonnet's approach is more robust and handles edge cases better
+
+**Sonnet Arbiter**: Selected sonnet  
+> My own solution provides the best balance of safety and simplicity
+
+**Gemini Arbiter**: Selected opus
+> Opus solution is more comprehensive, though Sonnet's is safer
+
+## ✅ ACCEPTED FIX: SONNET
+
+**Code Changes**:
+[Complete fix code]
+
+**Rationale**: [Why this approach]
+**Confidence**: 90%
+**Risks**: [List of risks]
+
+---
+Models Used: Opus, Sonnet, Haiku, Gemini (4 workers)
+Arbiters: Opus, Sonnet, Gemini (3 arbiters)  
+Decision Method: Majority vote
+```
+
+## Setup Requirements
+
+### 1. Gemini API Key
+
+```bash
+# Add to your shell profile
+export GEMINI_API_KEY="AIza..."
+
+# Or add to .claude/settings.json env vars
+```
+
+### 2. Python Dependencies
+
+```bash
+pip install requests
+```
+
+### 3. Workflow File
+
+**Location**: `VirtOS/.claude/workflows/multi-model-with-gemini.js`
+
+## Usage
+
+### Run 4-Model Review
+
+```javascript
+// Via Workflow tool
+Workflow({
+  scriptPath: '/home/sfloess/Development/github/FlossWare/VirtOS/.claude/workflows/multi-model-with-gemini.js'
+})
+```
+
+### Test Gemini Client
+
+```bash
+cd /home/sfloess/Development/github/FlossWare/VirtOS
+
+# Simple prompt
+python3 .claude/gemini_client.py "Rate this code quality 0-10"
+
+# With JSON schema
+python3 .claude/gemini_client.py \
+  "Find security issues" \
+  schema.json
+```
+
+## Benefits of 4-Model System
+
+### Diversity
+
+- **Claude models**: Different capability tiers
+- **Gemini**: Different training data, architecture, perspective
+- **Result**: More comprehensive coverage, fewer blind spots
+
+### Robustness
+
+- Single model bias eliminated
+- Consensus-based decisions
+- Multiple independent evaluations
+
+### Transparency
+
+- All proposals documented
+- All arbiter opinions shown
+- Vote breakdown visible
+- User can verify decision quality
+
+### Quality Control
+
+- 3 arbiters provide checks and balances
+- Majority vote prevents outlier selection
+- Confidence scores influence ties
+
+## Example Arbiter Scenarios
+
+### Scenario 1: Clear Consensus
+
+```
+Opus Arbiter → selects Sonnet
+Sonnet Arbiter → selects Sonnet
+Gemini Arbiter → selects Sonnet
+Result: SONNET (3/3 unanimous)
+```
+
+### Scenario 2: Split Decision
+
+```
+Opus Arbiter → selects Opus (90% confidence)
+Sonnet Arbiter → selects Gemini (85% confidence)
+Gemini Arbiter → selects Gemini (88% confidence)
+Result: GEMINI (2/3 majority)
+```
+
+### Scenario 3: Three-Way Split
+
+```
+Opus Arbiter → selects Opus (92% confidence)
+Sonnet Arbiter → selects Sonnet (88% confidence)
+Gemini Arbiter → selects Gemini (85% confidence)
+Result: OPUS (highest confidence wins tie)
+```
+
+## Cost Considerations
+
+**Claude API**: Charged per token
+
+- Opus: Most expensive, best quality
+- Sonnet: Mid-tier pricing
+- Haiku: Cheapest, fastest
+
+**Gemini API**:
+
+- Free tier: 60 requests/minute
+- Paid tier: Variable pricing
+- Generally cheaper than Claude Opus
+
+**Budget Control**: Use workflow `budget` parameter to limit spending
+
+## Continuous Review Integration
+
+The 4-model system integrates with existing continuous review:
+
+1. **Cron triggers** every 10 minutes
+2. **Scans detect** issues (Shell, Python, Java)
+3. **4 workers generate** fixes in parallel
+4. **3 arbiters vote** on best fix
+5. **GitHub issue created** with full transparency
+6. **Auto-fix safe issues** (formatting, imports)
+7. **Auto-commit and push** with co-author signature
+
+## Monitoring
+
+Check workflow progress:
+
+```
+/workflows
+```
+
+View logs:
+
+```bash
+tail -f /tmp/virtos-*.log
+```
+
+Check GitHub issues:
+
+```bash
+gh issue list --label "multi-model-review"
+```
+
+---
+
+**Created**: 2026-06-03
+**Status**: ✅ Active
+**Worker AIs**: 4 (Opus, Sonnet, Haiku, Gemini)
+**Arbiters**: 3 (Opus, Sonnet, Gemini)
+**Decision Method**: Majority vote