Jevotron is a command-line interface (CLI) designed for field-level anomaly detection within structured data, including CSV, YAML, JSON, TOML, SQLite, and DuckDB files. Powered by Jev, the tool allows users to identify suspicious entries by applying specific guidance to selected fields.
Jevotron: A CLI tool for field-level anomaly detection in structured data
The tool supports a variety of workflows, such as scanning files with custom instructions, exporting results to review queues, and integrating with shell environments. It features a caching mechanism using SQLite, which reuses previous assessments for unchanged entries to reduce computational costs and API usage. For large guidance documents, users can opt into token-aware batching to manage token consumption.
In a pilot study involving 24 public agent traces, Jevotron matched 79.8% of published step-quality labels. The tool also offers "jt-triage" skills that can be added to AI agents, enabling them to preview records, run pilot scans, and manage review queues to identify cases requiring human reasoning.
Sources
- Jevotron: Multiple Jev integrations from the command line （Hacker News Frontpage, 2026-10-01）
- cmungall/jevotron