GoKawiilJevotron launches CLI tool for anomaly detection across file formats
Jevotron is a new command-line tool that scans files such as CSV, YAML, JSON, TOML and OBO for anomalies, producing field-level probability scores and a review queue. The tool caches results in SQLite so unchanged entries aren't reassessed on subsequent runs, and users can guide scans with text instructions or a local Python config. A pilot test on 24 public agent traces reported the tool matched 130 of 163 step-quality labels (79.8% accuracy), with 89.7% precision and 70.3% recall for identifying harmful steps.
GoKawiil's interpretation of the reporting above, not reported fact.
The caching and incremental-scan design suggests Jevotron is aimed at practical, repeated-use workflows rather than one-off audits, which could make it more attractive for ongoing data-quality monitoring. The agent-trace pilot numbers, while limited to a small sample, offer an early data point on how well automated anomaly detection can substitute for human review of AI agent behavior.
- Jevotron supports multiple file formats including CSV, YAML, JSON, TOML, and OBO, with gzip support
- A local caching system avoids reassessing unchanged file entries across runs
- A small pilot on 24 agent traces showed 79.8% label-matching accuracy and 89.7% precision for harmful-step detection
Source: cmungall.github.io, 2026-10-01
Published there as: “Jevotron: Multiple Jev integrations from the command line”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.