Guardrails & review

jev-review

@devagrawal09★ 367TypeScriptMITupdated 2026-09-17▶ runnable

A staged code-review workflow and local dashboard built with TypeSafe Jev.

devagrawal09/jev-review ↗

Where it calls Jev

import { choice, noul, score, TypeSafeClient } from "@typesafe-ai/sdk";

src/review/codebase-judgments.ts:3

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 src/review/codebase-judgments.ts:52

  1. correctnesstrue/false

    Does file.content directly support that this code contains incorrect runtime behavior?

  2. securitytrue/false

    Does file.content directly support that this code weakens a security boundary?

  3. reliabilitytrue/false

    Does file.content directly support that this code can crash, race, leak, deadlock, or recover poorly?

  4. compatibilitytrue/false

    Does file.content directly support an internal inconsistency that can break a caller, format, protocol, or documented behavior?

  5. testGaptrue/false

    Does file.content contain important behavior without adequate targeted evidence in relatedTests?

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: {
    correctness: { type: 'boolean', instructions: 'Does file.content directly support that this code contains incorrect runtime behavior?' },
    security: { type: 'boolean', instructions: 'Does file.content directly support that this code weakens a security boundary?' },
    reliability: { type: 'boolean', instructions: 'Does file.content directly support that this code can crash, race, leak, deadlock, or recover poorly?' },
    compatibility: { type: 'boolean', instructions: 'Does file.content directly support an internal inconsistency that can break a caller, format, protocol, or documented behavior?' },
    testGap: { type: 'boolean', instructions: 'Does file.content contain important behavior without adequate targeted evidence in relatedTests?' },
  },
});

from src/review/judgments.ts:40

  1. correctnesstrue/false

    Does file.patch directly support that this change likely introduces incorrect runtime behavior?

  2. securitytrue/false

    Does file.patch directly support that this change introduces or weakens a security boundary?

  3. reliabilitytrue/false

    Does file.patch directly support that this change can crash, race, leak, deadlock, or recover poorly?

  4. compatibilitytrue/false

    Does file.patch directly support that this change can break an existing caller, format, protocol, or public behavior?

  5. testGaptrue/false

    Does file.patch change important behavior without adequate targeted evidence in changedTests?

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: {
    correctness: { type: 'boolean', instructions: 'Does file.patch directly support that this change likely introduces incorrect runtime behavior?' },
    security: { type: 'boolean', instructions: 'Does file.patch directly support that this change introduces or weakens a security boundary?' },
    reliability: { type: 'boolean', instructions: 'Does file.patch directly support that this change can crash, race, leak, deadlock, or recover poorly?' },
    compatibility: { type: 'boolean', instructions: 'Does file.patch directly support that this change can break an existing caller, format, protocol, or public behavior?' },
    testGap: { type: 'boolean', instructions: 'Does file.patch change important behavior without adequate targeted evidence in changedTests?' },
  },
});

Other projects in this category