talas9

Anti Hall — Git skill for Claude Code

Git community

Anti-hallucination guardrails for Claude Code and Codex: verify-first hooks that block unverified claims, risky git/shell commands, and scope drift.

How to install Anti Hall

This entry records only its repository, not the path inside it, so there is no exact command to give. Open talas9/anti-hall and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Anti Hall does

Anti-hallucination guardrails for Claude Code and Codex: verify-first hooks that block unverified claims, risky git/shell commands, and scope drift.

Alternatives in Git

Auteur — The Claude Code skill that directs a website like a film 1k ★
Destructive Command Guard — The Destructive Command Guard (dcg) is for blocking dangerous git and shell commands from being executed by ag 869 ★
Husky — by evmts - Sets up and manages Husky Git hooks by configuring pre-commit hooks, establishing commit message st 440 ★

README

🛡️ anti-hall

Make Claude Code and Codex verify before they claim — with platform-native guardrails and workflow skills.

[![tests](https://github.com/talas9/anti-hall/actions/workflows/test.yml/badge.svg)](https://github.com/talas9/anti-hall/actions/workflows/test.yml) [![version](https://img.shields.io/github/v/tag/talas9/anti-hall?label=version)](https://github.com/talas9/anti-hall/releases) [![license](https://img.shields.io/github/license/talas9/anti-hall)](LICENSE) ![node](https://img.shields.io/badge/node-%E2%89%A522-brightgreen) ![Claude Code plugin](https://img.shields.io/badge/Claude%20Code-plugin-8A2BE2) ![Codex port](https://img.shields.io/badge/Codex-port-111827)

Terminal demo: the doctor reports anti-hall active, then git-guard blocks a force-push and an AI self-credit trailer (exit 2), command-guard sends a build to a subagent, and the two-line statusline renders.

What it does

Stops your coding assistant from stating things it has not checked: invented library functions, "it works" without running anything, "done" with tasks still open.
Blocks risky git actions outright: force-pushes, and AI credit lines in commit messages and GitHub PR/issue/release text.
Keeps the main conversation responsive by pushing heavy commands and file edits to helper agents, and caps runaway agent spawning.
Adds skills you can call by name for debugging to a proven root cause, reviewing risky changes, and writing a session handover.
Works with Claude Code (plugin) and Codex (separate port). Pure Node, nothing else to install.

**Before and after.** An assistant, mid-task, runs a force-push to `main`. Without anti-hall, the published history is rewritten. With it, the command never runs and the assistant is told: