Agent tooling

bouncer

@clownware★ 1TypeScriptupdated 2026-09-20▶ runnable

Jev-powered Judgment layer for Claude Code. Stops paying reasoning prices for if-statements: a PreToolUse hook scores every tool call against a YAML policy you own — allow / deny in ~100 ms, no LLM in the loop. TypeSafe Jev now, local models next. Dry-run by default, calibration table published.

clownware/bouncer ↗

Where it calls Jev

//   POST https://api.typesafe.ai/v1/systemone

src/adapters/jev.ts:7

The link points at the commit we read, so the line number still holds.

What it asks Jev

These question sets are lifted from this project's source exactly as written.

from test/adapters.test.ts:6

  1. destructivetrue/false

    It destroys something.

  2. secretstrue/false

    It exposes a credential.

Run it on your own text

The code
import { experimental_evaluate as evaluate } from 'ai';

const { answers } = await evaluate({
  model: 'typesafe-ai/jev',
  state,
  questions: {
    destructive: { type: 'boolean', instructions: 'It destroys something.' },
    secrets: { type: 'boolean', instructions: 'It exposes a credential.' },
  },
});

from test/reasoning.test.ts:4

  1. on_brandtrue/false

    It sounds like us.

Run it on your own text

The code
import { experimental_evaluate as evaluate } from 'ai';

const { answers } = await evaluate({
  model: 'typesafe-ai/jev',
  state,
  questions: {
    on_brand: { type: 'boolean', instructions: 'It sounds like us.' },
  },
});

Other projects in this category