Inspect AI — анализ логов оценок моделей

data

read-eval-logs is a Claude Code skill that reads and analyses Inspect AI evaluation log files using the Python API and CLI commands. It covers listing logs filtered by status (success, error, retryable), reading full logs or headers only, memory-efficient streaming of samples, and isolating failed samples via summaries — all against .eval and .json formats. Key CLI operations include inspect log list, log dump, log convert, and log schema, plus the interactive inspect view viewer. The EvalLog and EvalSample structures expose aggregate scores, model usage stats, full message history, and event transcripts. The skill is aimed at ML engineers and researchers who need to debug evaluation runs, inspect per-sample results, or convert log formats without writing boilerplate scripts from scratch.