yzhao062/pyod — reverse-engineered prompt

Build me a Python package for finding unusual or suspicious records in data. I want it to work for normal tables first, but also have a clear path for time series, graphs, text, and images. The main experience should be simple, users should be able to install it, pick a detector, train it on their data, and get anomaly scores or predictions with a familiar fit and predict style API.

Also add a smarter mode where the library can inspect a dataset, suggest good anomaly detection methods, run a few options, compare the results, and explain what it found in plain language. It should include lots of detector choices, examples, notebooks, docs, and basic tests so people can trust it.

Please also make it friendly for AI coding agents. Add a command line setup flow for an expert skill, plus an MCP server with tools for listing detectors, explaining them, planning a detection run, running it, and summarizing findings. Look up current docs online if you need to.