atai-label-free-anomaly-discovery-agent-data-prep
Installation
SKILL.md
Label-Free AD Data Prep: raw data → files od can read
The Label-Free Anomaly Discovery agent (od) reads one CSV per batch session and compares
each window of it with the file's other windows. So the prep has two jobs:
- Make files od reads correctly: a timestamp column and numeric channels on an exact grid, with gaps left as gaps.
- Make files od can work on: long enough for LOF to mean something, with the anomalies you care about a small share of each file, and a window that suits the machine.
There are no roles and no labels here: no library, no validation or test split, no normalisation (od scales what it needs, on the file it scores). What matters is where each file starts and ends.