manuelschipper

manuelschipper / nah

Public

Context-aware safety guard for Claude Code. A permission system you control.

273
11
100% credibility
Found Mar 12, 2026 at 244 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

nah is a safety guard that monitors and controls tool actions in AI coding environments like Claude Code to prevent harmful operations such as data deletion or exfiltration.

How It Works

1
🔍 Hear about safer AI coding

You discover nah while using your AI coding helper and worry about risky actions like deleting files or accessing secrets.

2
📥 Get the safety guard

Download and set it up in moments with a simple command, no hassle involved.

3
🛡️ Turn on protection

Activate the guard with one click, and it starts watching every action your AI tries to take.

4
🤖 Let AI code freely

Your AI helper runs commands like reading files or running tests, feeling secure knowing the guard is alert.

5
Guard checks the action
✅
Safe action

It lets safe things like testing code go right through without interrupting.

❓
Risky action

It pauses dangerous moves like deleting important files and asks for your confirmation.

6
🔧 Tweak if you want

Easily adjust rules for your projects, like trusting certain folders, to fit your needs perfectly.

🎉 Code worry-free

Enjoy powerful AI help that stays safe, protecting your files and secrets while boosting your productivity.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 244 to 273 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is nah?

nah is a Python safety guard for Claude Code that intercepts every tool call—Bash, Read, Write, Edit, Glob, Grep, and MCP tools—before execution, using context-aware classification to decide allow, ask, or block. It solves Claude's coarse allow/deny permissions by analyzing commands structurally: git push is fine, but --force asks; rm __pycache__ ok, but ~/.bashrc nah; plus content scans for secrets or malware in writes. Users get a controlled permission system with full logging, CLI testing like nah test "rm -rf /", and YAML config for custom rules.

Why is it gaining traction?

Zero-config start with profiles (full, minimal, none), a 5-minute /nah-demo of 25 threat cases, and optional LLM escalation via Ollama or OpenAI for ambiguous calls. CLI shines for dry-runs, rule tweaks (nah allow git_history_rewrite), and logs (nah log --blocks). Project .nah.yaml tightens global rules without supply-chain risks, perfect for context-aware safety in LLM coding agents.

Who should use this?

Claude Code users crafting complex prompts with Bash tools. Teams sharing AI-assisted repos needing exfil/data-loss guards. Devs benchmarking context-aware safety for large language models in agentic workflows.

Verdict

Essential for Claude Code heavyweights—installs clean, docs comprehensive, CLI intuitive. 192 stars and 1.0% credibility mark it alpha; run the demo and nah log before trusting in high-stakes setups.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.