The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
Self-hosted coding agent CLI with workflow superpowers
10x is a terminal coding agent that breaks work into steps and can route each step to different model tiers. You can use its built-in Superpowers commands or define your own in `.10x/superpowers/` and `.10x/skills/`.
Builders who want a reusable agent workflow for coding, review, debugging, and test generation.
You can turn routine coding tasks into repeatable multi-step agent workflows instead of prompting from scratch each time.
What it does
Multi-step superpowers
Chains multiple steps together so a task can be analyzed, then implemented, then checked with different models.
Smart model routing
Chooses a model tier for each step, so simple steps can use faster models and harder steps can use stronger ones.
Built-in coding commands
Includes commands for review, PR descriptions, refactoring, debugging, explanation, and test generation.
Custom workflows
Lets you define your own superpowers in `.10x/superpowers/` or `~/.config/10x/superpowers/`.
Custom skills
Supports prompt files in `.10x/skills/` or `~/.config/10x/skills/` and runs them with `/<skill-name>`.
Project context file
Uses `10X.md` in the project root to carry project-specific context and conventions.
How to get it
- 1Run
npm install -g 10x-cli 10x
README
10x
Up to 20x faster coding - with Superpowers.
Quick Start
npm install -g 10x-cli
10x
Why 10x?
| Feature | 10x | Claude Code | Cursor | GitHub Copilot |
|---|---|---|---|---|
| Superpowers (Multi-Step Pipelines) | Chain models for complex workflows | No | No | No |
| Smart Model Routing | Auto-picks fastest model per task | Single model | Single model | Single model |
| Speed | Up to 20x faster | 1x | ~1x | ~1x |
| Open Source | MIT Licensed | Closed source | Closed source | Closed source |
| BYOK (Bring Your Own Key) | Full control over costs | Subscription only | Subscription only | Subscription only |
Superpowers
Multi-step AI workflows that chain different models together. Each step can use a different model tier, automatically routing to the fastest model that can handle it.
| Command | Description |
|---|---|
/review <path> | Code review with security, performance, and style analysis |
/pr | Generate PR description from staged/committed changes |
/refactor <file> | Guided refactoring with analysis and implementation |
/debug <issue> | Step-by-step debugging: reproduce, analyze, fix |
/explain <path> | Deep dive explanation of code architecture |
/test <file> | Generate comprehensive test suite |
Custom Superpowers
Define workflows in .10x/superpowers/ or ~/.config/10x/superpowers/:
---
name: debug
trigger: /debug
---
## Step 1: Understand (model: fast)
{{input}} - Find and read the relevant code.
## Step 2: Fix (model: smart)
Based on {{previous}}, implement a fix.
Model Tiers
| Tier | Model | Speed | Best For |
|---|---|---|---|
| ⚡⚡ Superfast | GPT OSS 20B | 20x | Simple queries, explanations |
| ⚡ Fast | Kimi K2 1T | 4x | Code generation, refactoring |
| ◆ Smart | Claude Opus 4 | 1x | Complex reasoning, architecture |
Configuration
Project Context
Create 10X.md in your project root:
# Project: MyApp
Tech: TypeScript, React, PostgreSQL
Conventions: Functional components, named exports
Custom Skills
Create prompts in .10x/skills/ or ~/.config/10x/skills/:
---
name: commit
---
Analyze staged changes and generate a conventional commit message.
Invoke with /<skill-name>.
CLI
10x Start interactive session
10x --byok Use your own OpenRouter API key
10x --model <tier> Set model tier (superfast, fast, smart)
10x --resume <name> Resume a session
10x -x "<prompt>" Execute prompt and exit
License
Files in the repo
- apps
- media
- packages
- .gitignore
- bun.lock
- LICENSE
- package.json
- README.md
- tsconfig.base.json
- turbo.json
Discussion (0)
Ask about usage, or say what you built with itSign in to join the discussion.
No comments yet. Be the first to say what this is good for.
More harnesses
from vibe coding to agentic engineering - practice makes claude perfect
🌊 The original agent meta-harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversational AI systems. Features adaptive memory, self-learning intelligence, RAG integration, and native Claude Code / Codex / Hermes and many more Integrated
Practical patterns, starters & CLI tools for loop engineering with AI coding agents. Design systems that prompt and orchestrate agents (inspired by Addy Osmani and Boris Cherny). Includes loop-audit, loop-init, loop-cost.
Git. Ship. Done - Core

The most RAM efficient harness