What problem does it solve? Evaluating LLMs with the karenina framework spans many workflows—benchmarks, rubrics, templates, verification, scenarios—and users often don't know which API or entry point fits their task. This Skill acts as a dispatcher: it interprets the user's request, supplies cross-cutting karenina knowledge (imports, API tables, common pitfalls), and routes to the correct specialized leaf skill. ## Core Features & Use Cases - Intent-based routing: A decision tree maps requests to nine leaf skills (karenina-qa, karenina-task-eval, karenina-scenarios, karenina-template-authoring, karenina-rubric-authoring, karenina-verification, karenina-cli, karenina-manual, karenina-results), with clarification questions for ambiguous cases. - Cross-cutting reference context: Auto-generated Benchmark API and VerificationResult accessor tables, correct import paths, and gotchas for templates, rubrics, configuration, checkpoints, and pipeline guards. - Synced documentation: A references/ directory with 113 files mirrored from the karenina docs, indexed by topic for answering conceptual questions directly. - Use Case: A user says "I have chatbot logs and want to score them for safety." The Skill recognizes pre-recorded outputs, routes to karenina-task-eval, and supplies the parsing_only=True requirement and TaskEval import path so work starts correctly. ## Quick Start Ask the agent to evaluate an LLM's outputs or build a benchmark with karenina, describing whether you have pre-recorded outputs, need multi-turn conversations, or want single-turn Q&A verification.
