What is jev-humanizer?

An anti-slop skill for AI agents that read SKILL.md, such as Claude Code and Codex. It removes bureaucratese, clichés, false contrasts, long dashes and filler that make a reader recognize machine-written text, while leaving facts and numbers in place.

The agent does not have to hunt for slop itself: Jev, the classifier that returns probabilities rather than prose, checks every paragraph against 35 semantic patterns. A linter adds mechanical findings such as dashes and arrows.

What does the model actually edit?

Only the flagged spots. The skill hands the model a findings table and the model revises what was marked. Its favorite phrases go unnoticed by the writer but not by the classifier, so the fixes stay targeted instead of rewriting everything.

What does it cost to run?

Jev takes $0.042 per million input tokens with output free. A 200-word post costs $0.0015 to $0.005 per run and typically takes two runs; that price covers only the audit, not whatever the agent itself costs.

Does it work for English text?

Partly. The reference and linter currently know only Russian, while Jev reads English too. On an example from blader/humanizer it caught a not just X, it is Y contrast, a long dash and an emoji, but missed Let that sink in and a rhetorical question.

Clone the repo to ~/projects/jev-humanizer, symlink its skill directory into ~/.claude/skills/jev-humanizer or ~/.codex/skills/jev-humanizer, and provide the Jev key via the TYPESAFE_API_KEY variable or ~/.config/typesafe.env. Python 3 with no extra packages is enough.

Does my text leave my machine?

Paragraphs are sent to TypeSafe's API in the US. With the --no-jev flag nothing leaves the machine, but then only the local linter checks the text. The terminal audit also supports --sections and --json output with token usage.

Jev directory

CommunityJev Agents & MCP

Jev Humanizer

An anti-slop Agent Skill for Claude Code, Codex and any agent that reads SKILL.md: Jev checks every paragraph against 35 semantic patterns, a local linter adds mechanical tells like dashes, and the model edits only what was flagged.

See highlights Open github.com

Stars: 3
Language: Python
License: MIT
Reading time: 2 min
Published by: Community
Author: snjrusmn
Last verified: Oct 2026

snjrusmn/jev-humanizer

User repository on GitHub

View on GitHub

Антислоп скилл для AI-агентов: Jev от TypeSafe находит признаки нейросети в каждом абзаце, модель правит только найденное. На основе smixs/humanizer-ru и blader/humanizer.

Forks: 0
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

Tagscommunity catalog-import logicrw russian writing anti-slop

Works forJev for Classification Jev for Verification

Used inJev for Media & Entertainment

Languages: Python
Platforms: CLI
Capabilities: Open sourceRequires API key

Highlights

Jev checks each paragraph against 35 semantic patterns and returns probabilities instead of text; a linter adds mechanics such as dashes and arrows.
The model receives a findings table and edits only the flagged spots, so facts and numbers stay in place.
Costs less than a cent per post: Jev charges $0.042 per million input tokens with free output, and a 200-word post runs $0.0015-$0.005 per pass.
The reference and linter are Russian-only, though Jev reads English; on an English example it caught not just X but Y phrasing while missing Let that sink in.
Install symlinks the skill into ~/.claude/skills or ~/.codex/skills; the TypeSafe key comes from TYPESAFE_API_KEY or ~/.config/typesafe.env.

Quickstart

bash

git clone https://github.com/snjrusmn/jev-humanizer.git ~/projects/jev-humanizer

# Claude Code
mkdir -p ~/.claude/skills
ln -s ~/projects/jev-humanizer/skill ~/.claude/skills/jev-humanizer

# Codex
mkdir -p ~/.codex/skills
ln -s ~/projects/jev-humanizer/skill ~/.codex/skills/jev-humanizer

Watch out

Grounded in the truncated repository README (primarily Russian) and GitHub metadata reviewed on 2026-10-15; the README was cut off mid-diagram in the how-it-works section. Paragraph text is sent to TypeSafe's API in the US unless the --no-jev linter-only flag is used.

Questions about this entry

Short answers drawn from the linked project or page.

From the Jev guides

Original write-ups on what this entry is for.

MCP tool routing: Jev in front of the tool call

Oct 16, 2026

Tool menus can outgrow the context that reads them. The catalog's Jev-based routers and gates narrow the list: benchmarked tool search, per-tool scoring, fail-closed approvals, and WebMCP in the browser.

Jev Agents & MCP#mcp#tool-routing#agents

Read article

Approval policies for coding agents: hooks, guards, and gateways

Oct 16, 2026

How the Jev guard family turns coding-agent permissions into auto-approve, deny, or ask: hooks that enforce, guards that fail safe, and gateways that gate generation.

Jev Agents & MCP#coding-agents#guardrails#hooks

Read article

From the Jev AI model community

Posts from builders shipping with Jev right now.

Follow @typesafeai