specsmith: Skill for Claude Code

VLLM Platform Integration

This skill enables specsmith projects to integrate with VLLM for large language model serving while maintaining governance compliance.

Requirements Covered

• REQ-110: VLLM Platform Integration
• Allows specsmith projects to integrate with VLLM for large language model serving with human approval

Integration Capabilities

Platform Features

• VLLM model serving and deployment
• GPU and CPU resource management
• Model optimization and quantization
• API endpoint configuration for VLLM services
• Performance monitoring and tuning

Governance Requirements

• Human approval required for all VLLM integrations
• Configuration management with governance controls
• Security best practices enforcement
• Resource monitoring and logging

Usage Examples

# Initialize VLLM integration
specsmith skill install vllm-integration

# Configure VLLM platform settings
specsmith vllm configure --platform vllm --model-path /path/to/model --gpu-count 2

# Deploy model via VLLM
specsmith vllm deploy --model my-model --port 8000

Configuration

The skill requires the following configuration in scaffold.yml:

vllm:
  platform: vllm
  model_path: /path/to/models
  gpu_count: 2
  port: 8000
  quantization: "none"

Security Considerations

• All VLLM integrations require human approval
• Model access controls must be configured
• Network security must be enforced
• Resource usage monitoring is required

Compliance Requirements

• Follow security best practices for AI platform integrations
• Maintain audit logs of all VLLM interactions
• Ensure proper resource management and monitoring