jev-harness
TypeScript library adding policy, confidence gates, shadow mode, recipes and an eval CLI around Jev answers.
Overview
jev-harness by antoniocoppe is a small TypeScript library that turns Jev answers into shippable actions. It adds what a single API call lacks in production: a policy mapping answers to actions (notify, suppress, place, skip and so on), a confidence gate that reviews, escalates or suppresses instead of guessing, shadow mode that logs what would happen without changing behaviour, copy-paste recipes for alerts, routing, row filters and agent handoffs, and an offline eval CLI that replays fixtures and asserts on actions rather than text.
The README's example decides what to do with a disk alert using a disposition Choice, a severity Score and a needs-human Noul, with a minimum confidence of 0.55 and review on low confidence. It reports filtering 24 rows with a natural-language predicate in 1.3 seconds with Jev versus 48.9 seconds with the Claude Code CLI on the author's machine. It is not affiliated with TypeSafe and needs Node 20+.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | What should we do with this alert? | notify · queue · suppress |
| Score | How bad is this? | Low · Medium · High · Critical |
| Yes / no | Does a human need to look at this? | — |
How it works
- Define questions and a policy
- harness.run sends state to Jev
- Apply minConfidence gate; low confidence goes to review
- Execute or shadow-log the action; replay fixtures in evals
Strengths
- Shadow mode for safe rollout
- Evals assert on actions
- Reusable recipes
Caveats
- Speed comparison is a single local measurement
- Unofficial
Getting started
`npm install jev-harness` and `export TYPESAFE_API_KEY=...`, then use `new DecisionHarness().run({...})`.At a glance
- Type
- CLI
- Status
- beta
- Language
- TypeScript
- Runtime
- Node 20+ library and CLI
- Has code
- yes
- Has tests
- yes
Use cases and integrations
How curated lists describe it
Unofficial production layer around Jev: policy, confidence gate, shadow mode, recipes, and an eval CLI
AnotiaWang/awesome-decision-modelsDecision harness for TypeSafe Jev — confidence gates, shadow mode, recipes, and evals. Claude CLI 48.9s → Jev 1.3s on the same row-filter job.
hellogumbo/awesome-jevTypeScript library that wraps Jev answers in policies, confidence gates, shadow mode, and reusable recipes.
MrJev/awesome-jev
Frequently asked questions
What does the harness add?
Policies, confidence gates, shadow mode, recipes and offline evals around Jev.
What is shadow mode?
Logging what the policy would do without changing live behaviour.
Is it official?
No, it is not affiliated with TypeSafe.
Related projects
Eval-first MCP server for Jev: prototype questions, map them over many items, and measure accuracy and thresholds on labelled data.
MCP serverbetaTypeScriptTool that lets code and agents ask Jev typed questions via reusable jevels, mapping answers to act, mark or fall_back.
ProjectprototypeTypeScriptDeclarative compiler that turns state, decisions and labeled examples into a measured, frozen JSON program of typed Jev questions.
LibraryprototypePythonAgent skills for placing, building, evaluating and improving decision-model systems, with Jev as the default hosted example.
Agent skillbetaClaude Code plugin and CLI for the Jev funnel: judge items in bulk, settle confident ones in code, send only uncertain ones onward.
PluginbetaPythonUnofficial Claude Code plugin that checks 'done' claims and small judgment calls with Jev, logs receipts, and can warn on unverified stops.
Pluginbeta