What problem does it solve? Building a new karenina adapter requires understanding how a target SDK maps to karenina's three port interfaces (AgentPort, LLMPort, ParserPort), but SDK capabilities like streaming, structured output, and trace access vary widely and are easy to overlook. This Skill structures Phase 1 of adapter creation by systematically collecting SDK documentation, probing capabilities, and resolving design decisions before any code is written. ## Core Features & Use Cases - SDK Documentation Collection: Fetches README, API references, trace/observability docs, and message type documentation for the target SDK via web retrieval. - Capability Assessment: Walks through a structured capability matrix covering agent loops, async support, MCP integration, tool definitions, usage/token tracking, streaming, structured output, recursion limits, timeout recovery, and filesystem backends. - Design Decision Resolution: Asks targeted clarifying questions about deep-agent vs scaffolded orchestration, built-in tool handling, registry naming, availability checks, fallback behavior, and distribution model (built-in, plugin, or manual). - Use Case: When integrating a new agent framework like CrewAI or AutoGen into karenina, run this Skill first to produce a context document at docs/adapters/<name>-context.md that Phase 2 (adapter design) consumes directly. ## Quick Start Ask the assistant to gather context for a new karenina adapter for your target SDK, providing the SDK name, documentation URL, and PyPI package name.
