jhaizhou-ops

jhaizhou-ops / karma

Public

Make AI Agents never violate your rules in long tasks — auto-correct violations before they frustrate you. Pure-engineering zero-LLM hook system for Claude Code / Codex CLI / Gemini CLI. Measured violation rate ≈ 0%.

agent-tooling ai-agent claude-code codex-cli coding-assistant
10
0
100% credibility
Found May 15, 2026 at 10 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

Karma is a lightweight system that ensures AI coding assistants consistently follow your personal guidelines during extended interactions by injecting reminders and blocking violations in real time.

How It Works

1
🔍 Discover karma

You hear about karma while using your AI coding buddy and struggling with it forgetting your instructions over long chats.

2
📥 Get it set up

You grab it and set it up in seconds so it works with your AI helper right away.

3
✏️ Tell it your wishes

You describe a few key preferences like 'think long-term' or 'keep things non-blocking' in simple words, and it understands perfectly.

4
🔄 Restart and go

You refresh your AI helper, and now every chat starts with your wishes front and center.

5
🛡️ Watch it stay on track

As you work on big projects, it gently reminds or blocks slip-ups before they happen, keeping everything aligned without you repeating yourself.

🎉 Perfect teamwork

Your AI buddy follows your style flawlessly through marathon sessions, making coding feel effortless and reliable.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 10 to 10 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 karma?

Karma hooks into Claude Code, Codex CLI, and Gemini CLI to pin your 5-10 core rules into every AI agent prompt, blocking violations like sleep commands or quick patches before they run in long tasks. It auto-injects reminders across 8 hook points—pre/post-tool use, session starts, subagents—with <60ms latency and measured ~0% violation rate in real sessions. Built in pure Python with zero LLM dependencies, it even lets you type `/karma ` for the agent to draft and add rules via YAML.

Why is it gaining traction?

Unlike karma coverage github runners or karma jasmine github test tools, this karma tool github stands out by complementing setups like Karpathy's CLAUDE.md—handling personal prefs (no jargon for non-tech users, root-cause fixes) that drift in 60K+ contexts. Developers dig the real-time blocks (e.g., denying sleep without background=True), cross-client install (`karma install-hooks`), and natural-language rule refinement without YAML hassle. No network calls or APIs means instant, reliable enforcement.

Who should use this?

Backend devs chaining long Claude Code sessions for refactors, ML engineers running eval pipelines in Codex CLI tired of data leakage drifts, or full-delegation users in Gemini CLI wanting agents to push without "what's next?" stops. Ideal if you're how to make agents in claude code without repeating rules every turn.

Verdict

Grab it if you're deep in these CLIs—early promise with full tests and bilingual docs outweighs 10 stars and 1.0% credibility score. Still beta; audit false positives via `karma audit` before production. (198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.