ask-jev
Windows Codex hook that audits a conversation's recorded execution with Jev and answers yes/no questions like 'Did the tests pass?'.
Overview
ask-jev is an explicit, advisory audit tool for local Codex conversations on Windows PowerShell. Typing :jev audits the current conversation's recorded execution; :jev ask followed by a one-line yes/no question (such as Did the tests pass?) returns a typed judgment plus evidence sufficiency; --details adds the evidence window and omission counts. It uses Codex's UserPromptSubmit blocking reason to show results without starting a Codex model turn.
Audit and ask deliberately send selected conversation, instructions and tool content to TypeSafe for separately billed Jev inference, while ordinary prompts pass through without transcript access or network calls. Results are probabilities, not generated explanations. It installs per repository or per user with a PowerShell script, uses TYPESAFE_API_KEY and an optional TYPESAFE_MODEL (jev-latest by default), and guards against duplicate requests if installed in both scopes.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Did the tests pass? | yes · no |
| Yes / no | Is the recorded evidence sufficient to judge? | yes · no |
How it works
- User types :jev or :jev ask <question>?
- Hook selects relevant recorded conversation and tool content
- Jev returns a typed judgment and evidence sufficiency
- Result shown via Codex's blocking reason, without a model turn
Strengths
- Independent, cheap check of agent claims
- No transcript access for ordinary prompts
- No package installation needed
Caveats
- Windows PowerShell only
- Sends conversation content to TypeSafe when invoked
Getting started
powershell.exe -NoProfile -File .\install.ps1 (add -User for user-level), then trust the hook via /hooks.At a glance
- Type
- Project
- Status
- beta
- Language
- PowerShell
- Runtime
- Codex hook (Windows)
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
A Windows PowerShell tool for auditing recorded Codex execution evidence with :jev.
logicrw/awesome-jev-projectsUtilizing Jev, the RLCD-type model provided by TypeSafe AI, to independently and cheaply judge agentic coding sessions.
hellogumbo/awesome-jev
Frequently asked questions
What can I ask?
A one-line yes/no question about the conversation's recorded execution, ending in a question mark.
Does it read every prompt?
No, ordinary prompts pass through without transcript access or network calls.
Which platform?
Windows PowerShell 5.1 (7.6.5 also tested) with Codex.
Related projects
Open-source Claude Cowork alternative built on OpenCode; per list notes, its eval testkit uses Jev as a verification judge.
ProjectproductionTypeScriptUnofficial Claude Code plugin that checks 'done' claims and small judgment calls with Jev, logs receipts, and can warn on unverified stops.
PluginbetaCLI that grades the coding-agent sessions on your disk (Claude Code, Codex, opencode, Gemini CLI) with Jev for cents.
CLIbetaTypeScriptAgent skill that tells you when you need an eval and how to build it, with an opt-in lane routing closed-label residue to Jev.
BenchmarkbetaPythonAgent observability platform that finds and fixes agent failures, with an optional Jev preclassifier for conversation checks.
ProjectproductionTypeScriptClaude Code plugin and CLI for the Jev funnel: judge items in bulk, settle confident ones in code, send only uncertain ones onward.
PluginbetaPython