adapter-factory

Integrate new AI models using the adapter-factory's factory pattern with BaseCLIAdapter for command-line tools or BaseHTTPAdapter for HTTP APIs. Register adapters via YAML config, handle request/response parsing, manage retries and timeouts, and substitute environment variables for secrets to enable seamless model participation in deliberations.

CLI adapters handle subprocess execution for local command-line AI tools, managing output parsing and tool-specific results. HTTP adapters integrate cloud APIs, managing HTTP requests, retries, response parsing, and timeout handling. Choose CLI for local models like Ollama; use HTTP adapters for proprietary cloud services.

Yes. The adapter-factory uses config-driven integration where new adapters are registered via YAML without modifying core deliberation code. Environment variable substitution supports secrets, context-aware argument adjustments handle model-specific requirements, and backward compatibility is preserved across all existing adapters.

Write unit and integration tests guided by test-driven development practices included in the skill. For HTTP adapters, use VCR to record and replay HTTP responses, enabling reproducible testing without external API calls. Validation includes prompt-length checks and response parsing verification.

The adapter-factory implements built-in retry and timeout handling in BaseHTTPAdapter, managing transient failures gracefully. Context-aware argument adjustments and proper error handling preserve deliberation flow while logging failures for debugging and maintaining system stability.

No. The skill provides step-by-step guidance and pre-built BaseCLIAdapter and BaseHTTPAdapter base classes that handle boilerplate concerns like retries, timeouts, and environment variable substitution, so you focus on tool-specific output parsing and request construction.