User repository on GitHub
Utilizing Jev, the RLCD-type model provided by TypeSafe AI, to independently and cheaply judge agentic coding sessions.
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Agents & MCP
Ask Jev — omni-
A PowerShell hook for Codex that intercepts your prompt turn and asks Jev to audit the conversation recorded so far, answering with typed probabilities rather than prose.
- Stars
- 1
- Language
- PowerShell
- License
- MIT
- Published by
- Community
- Author
- omni-
- Last verified
- Oct 2026
This is one of several Jev listings for the same project. The canonical entry is Ask Jev — logicrw.
About this project
An explicit, advisory audit of a local Codex conversation's recorded execution. Windows PowerShell 5.1 is the installation runtime; PowerShell 7.6.5 was also tested. No package installation is needed.
Quoted from the project's README. Read the full README.
Highlights
- Installs one UserPromptSubmit hook command, keeps other hooks and settings intact, and never edits trust hashes or execution policy.
- Each audit batches 13 focused checks plus one mutually exclusive outcome over a window capped at 48 KiB of selected evidence.
- A standalone CLI runs the same extractor offline in preview mode, printing the redacted request and making zero network calls.
- Non-command prompts pass through untouched, with no transcript access and no network call; the help command stays local.
Watch out
Ask-jev sends selected conversation, instructions and tool content to TypeSafe for separately billed Jev inference, and the README says secret removal is not comprehensive: configured keys and obvious credential fields are redacted, but it advises reviewing an offline preview before sending sensitive work. The probabilities are advisory rather than calibrated guarantees, and a low mutation-attribution score does not identify a different author; every check stays visible, with possible issues ordered by probability of yes. Absence of evidence is not evidence: a partial window is labelled, and the extractor reports missing results, compaction, unknown records and budget omissions rather than concluding something never happened. Windows user environment storage is plaintext, Codex's transcript format is not a stable API, and Claude Code and OpenCode are not implemented; other shells, remote or cloud sessions and other operating systems are unverified. The supervisor kills an overlong worker at 28 seconds against a 40-second host timeout. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
No. Installation writes one command into .codex/hooks.json, or $CODEX_HOME/hooks.json, and the README states no conversation is sent during installation. Ordinary prompts pass through silently, without transcript access or network calls, and the help command stays local. Only an audit and an ask transmit selected conversation, instructions and tool content. Matching is exact and case-sensitive, with only a trailing carriage return or newline stripped.
Not proved, and the README is careful about the difference: the percentages estimate whether a statement is true, not the fraction of the conversation checked, and the probabilities are advisory rather than calibrated guarantees. Each audit is 13 focused checks plus one mutually exclusive outcome, every check remains visible, and possible issues are ordered by probability of yes. A low mutation-attribution probability in particular does not name someone else as the author.
Partial, and the README says so in those words. Configured-key matches and obvious credential fields are redacted, and binary payloads, base64 blobs and encrypted reasoning are omitted, but the project explicitly states this is not comprehensive secret removal. The safe habit is the offline preview, which prints the exact redacted request and makes zero network calls, and the JSON output, which omits the input trace from the result.
Not in this version. The stated runtime is Windows PowerShell 5.1, with 7.6.5 also tested, driven by the PowerShell hook runner in the local Codex TUI or desktop app. Claude Code and OpenCode are listed as not implemented, and other session shells, remote or cloud sessions and other operating systems are unverified. The test suite runs offline with synthetic traces and an injected transport, which is not the same as coverage on your own machine.
:jev audits the current conversation, :jev ask with a question returns a typed yes/no judgment plus evidence sufficiency, :jev --details adds the evidence window and omission counts, and :jev help stays local. Ordinary prompts pass through without transcript access.
Running ask-jev.ps1 with -Rollout and -Preview emits the exact redacted request and makes zero network calls. Passing -Json or -Question makes one TypeSafe request and returns a JSON result without the input trace.
It reads a size-bounded snapshot of the exact session, capped at 48 KiB of selected evidence, with the complete request capped at 64 KiB. A partial window is explicitly labeled, and the supervisor kills an overlong worker at 28 seconds.
More Jev resources like this
1GitHub stars
A DeepSeek Harness plugin that exposes System One decisions as agent tools and adds an opt-in auto-guard, with a labeled benchmark it re-runs against the live API.
Jev Agents & MCP#community#jevbest#catalog-import
Community
xienda/dsh-jev-verifyJavaScript
xienda/dsh-jev-verifyJavaScript51GitHub stars
Local LinkedIn recruiting workspace on Browser Use's Jev Ultrafast: Jev screens titles, drives navigation, and assesses profile excerpts against your brief while a live browser stays visible; matches and evidence export as JSON.
Jev Agents & MCP#community#jevbest#catalog-import
Community
skeptrunedev/jev-recruiterPython
skeptrunedev/jev-recruiterPython33GitHub stars
Codex Skill plus stdlib-only Python CLI that recommends which installed skill to read next for a task, allowing no-skill or uncertain answers and never executing candidates.
Jev Agents & MCP#community#jevbest#catalog-import
Community
win4r/jev-skill-suggesterPython
win4r/jev-skill-suggesterPython72GitHub stars
MCP server over stdio that gives coding agents fourteen Jev judgment tools, turning evidence plus an enumerable question into auto, review, or escalate actions from calibrated probabilities.
Jev Agents & MCP#community#jevbest#catalog-import
Community
PyModel/jev-judge-mcpPython
PyModel/jev-judge-mcpPython21GitHub stars
Research runtime that pages a large logical tool and context space through a small bounded Jev decision surface with explicit approvals.
Jev Agents & MCP#community#jevbest#catalog-import
35GitHub stars
Rust terminal tool that asks Jev noul, choice and score questions about piped text or files and returns calibrated probabilities with stable exit codes for shell scripts, CI jobs and agents.
Jev Agents & MCP#community#jevbest#catalog-import
Community
shaharia-lab/jev-cliRust
shaharia-lab/jev-cliRustFrom the Jev community
Posts from builders shipping with Jev right now.