Why not just write the rules in project markdown?

Markdown alone is a suggestion the agent can forget. This writes hooks that return block decisions, so failing-test commits, dangerous commands, and protected-path writes are stopped even mid-session.

Which preset should I start from for a new repo?

Minimal covers dangerous, main-branch, and build-output guards, safe adds test and typecheck gates plus secret handling, and strict adds test-first on every edit. The init flag always wins and never calls Jev.

How does the Jev chooser turn my description into blocks?

It asks enable or disable per block plus one strictness question, using fixed thresholds with a middle band that keeps the preset default. Rule text and free-form params never come from the model and locked entries resist weakening.

What is the cost of per-call semantic rule checks?

Each enforced rule adds one Jev question per tool call with block, ask, and allow bands, while lint rules scan the staged diff including about-to-be-staged content. Both layers add to deterministic guards without relaxing them.

What do the minimal, safe and strict presets enable?

minimal covers dangerous-command, main-branch and build-output guards; safe adds test and typecheck gates, lockfile and secret-file guards and lint on save; strict adds test-first TDD on every source edit.

Which agents does oh-my-harness write hooks for?

Claude Code, Codex and Pi at once, all generated and kept in sync from one harness.yaml; omh sync --check fails CI when the markdown and hooks drift apart.

What always-on guards are added to every preset?

Harness self-protection, a --no-verify guard and a force-push guard are added to every preset and every Jev result, so an agent cannot edit its own hooks to switch the guardrails off.

CommunityJev Agents & MCP

Oh My Harness

Single-command harness that turns a plain-language stack description into synced docs plus blocking hooks for Claude Code, Codex, and Pi, with deterministic presets or one Jev chooser call.

See highlights

Open github.com

Stars: 16
Language: TypeScript
License: MIT
Published by: Community
Author: kyu1204
Last verified: Oct 2026

kyu1204/oh-my-harness

User repository on GitHub

View on GitHub

Tame your AI coding agents with natural language. Generate enforced guardrails (CLAUDE.md, hooks, settings) from a single command.

Forks: 2
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.Homepage

About this project

CLAUDE.md is a request. oh-my-harness is enforcement. One command turns "TDD enforced, block dangerous commands" into hooks that actually block your AI coding agent — for Claude Code, Codex and Pi at once. That is the whole setup. Your agent now hits a wall when it tries to: Every decision is logged to .omh/state/events.jsonl — omh stats shows what your agent tried and what got stopped, and omh explain tells you in plain language why the last few calls were blocked and how to allow once or change the rule.

Quoted from the project's README. Read the full README.

Tagscommunity jevbest catalog-import

Works forJev for Classification Jev for Moderation Jev for Tool calling

Used inJev for Software & DevTools

Languages: TypeScript

Highlights

Generates hooks, docs, and settings from one harness file and fails CI when generated files drift.
Blocks risky commits, force pushes, hook tampering, and protected-path writes while auto-linting on save.
Offers offline presets plus an optional sub-cent Jev call that maps descriptions to blocks with confirm-before-apply.
Logs every decision to local JSONL with stats dashboards and plain-language explanations for blocked calls.

Watch out

Preset blocks are templated rather than model-written, other-language descriptions get lower confidence, and semantic guards allow-and-log on API errors. 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.

More Jev resources like this

patdown

Node CLI that lints a file tree against fuzzy natural-language rules in one Markdown file, judging each matched file with a swappable TypeSafe-backed judge.