Jevotron: Multiple Jev integrations from the command line

Jevotron launched a command-line tool that scans CSV, YAML, JSON, TOML, text, and OBO files for field-level anomalies using the TypeSafe API, caching each successful assessment in SQLite so unchanged entries are not reassessed. In a pilot on 24 public agent traces, Jevotron matched 130 of 163 step-quality labels (79.8%), with 89.7% precision and 70.3% recall on harmful-step detection. The tool requires a TYPESAFE_API_KEY for scans, while its preview command runs without a key.

One-shot anomaly detection on the command line ¶ one-shot-anomaly-detection-on-the-command-line Give jevotron a file and a little guidance. Get field-level anomaly scores, a focused review queue, and a cache that makes the next run cheaper. Run your first scan quickstart/ Explore the commands reference/cli/ Install agent skills guides/agents/ bash $ jevotron scan airports.csv --guidance "Check airport locations." First, set your API key Get a key from the TypeSafe dashboard https://console.typesafe.ai/ and set it in your shell: export TYPESAFE API KEY="your-api-key" Already set? You're ready to scan. preview needs no key. API key setup → quickstart/ set-your-api-key · See the actual airport results → examples/airports/ actual-output 01 / PREVIEW See what goes in ¶ see-what-goes-in Inspect chunks, field paths, and exact model requests before making an API call. 02 / SCAN Assess every entry ¶ assess-every-entry Jev scores selected fields together. Each entry gets the same guidance and optional examples. 03 / REVIEW Start with the warnings ¶ start-with-the-warnings Sort suspicious entries, export CSV, or pipe JSONL into your existing shell workflow. Small setup. Useful defaults. ¶ small-setup-useful-defaults CSV, YAML, JSON, TOML, text, OBO, and more work out of the box, including gzip files. The format reference reference/formats/ covers defaults and format-specific options. Select fields with --field , add a sentence with --guidance , and run. Longer instructions can come from --guidance-file . A local Python config is available when a project needs custom parsing or reusable settings. Unchanged input reuses its assessment. SQLite saves each successful result as it arrives. Reorder a file, change a reporting threshold, or resume a failed run without reassessing unchanged entries. A review aid with visible evidence. Reports retain each field's probabilities, the entry score, source location, and assessment date. The warning score is the highest field anomaly probability; you choose the threshold. Try a complete example ¶ try-a-complete-example | Example | What you'll do | |---|---| | Airports / CSV examples/airports/ | Find two injected country errors in public data, then compare versions. | | Inventory / YAML examples/inventory/ | Apply written rules and a chosen exemplar to stock records. | | Measurement units / OBO examples/obo/ | Score definitions and repeated synonyms within independent stanzas. | | Agent traces / JSONL examples/agent-traces/ | Classify public agent traces and individual steps, then compare with published labels. | Agent traces: a measured pilot ¶ agent-traces-a-measured-pilot On a small, length-filtered sample of 24 public traces , jt matched 130 of 163 step-quality labels 79.8% . Harmful-step precision was 89.7% , with 70.3% recall . The example uses original messages and tool definitions, with human labels withheld from the model. Try the trace example → examples/agent-traces/ · Read the full analysis and limitations → analysis/agent-traces/