sensei

Sensei

Code review that teaches engineering judgment, not just what to change.

Sensei is a senior engineering coach for developers who want better review, debugging, planning, refactoring, and test feedback. It helps you understand what changed, why it matters, what could break, and what evidence proves the work is safe to ship.

It is useful for senior engineers coaching juniors, solo developers who want a sharper second reviewer, students learning how real code gets reviewed, and anyone using AI to get code working who now needs to make it cleaner, safer, and easier to defend.

The developer still owns the code. Sensei asks the questions that make the code better.

---

Works best inside your repo

Install Sensei in the repo you want it to review.

A repo is the project folder that holds your code. When Sensei runs inside that folder, it can use the context that actually matters:

• Your existing components, helpers, and patterns
• Your architecture docs, AGENTS.md, CLAUDE.md, Cursor rules, or other project instructions
• Your tests, package files, and folder structure
• Your Git state: changed files, open PRs, recent commits, and branch history
• Any other skills you have installed or pointed your agent at

That is the difference between generic code advice and review that knows how your project is supposed to work.

---

Who it is for

• Senior engineers who want juniors to get guidance without being spoon-fed answers
• Solo developers who want a second reviewer before they push code
• Junior developers, students, and bootcamp grads learning how to think like maintainers
• AI-assisted builders who got something working and now need to know if it is fragile, bloated, risky, or ready
• Any developer looking for alternative code review, quality checks, debugging help, or refactoring guidance

---

What Sensei helps with

• Reviews diffs and asks you to explain your reasoning before it judges the code
• Finds code bloat, duplicated logic, unclear responsibility, and over-engineered abstractions
• Checks whether the change follows existing project patterns
• Reviews implementation plans before you build the wrong thing
• Helps debug by turning guesses into hypotheses and experiments
• Explains what your tests prove and what they still miss
• Helps write PR descriptions with intent, risks, and verification
• Checks security-sensitive surfaces like auth, permissions, secrets, user data, and outside input
• Turns finished work into learning, so the next change is better

---

Install

Recommended: install from skills.sh

Run this from the repo where you want to use Sensei:

npx skills add onehorizonai/sensei

Browse Sensei on the registry: skills.sh/onehorizonai/sensei

Claude Code plugin

/plugin marketplace add onehorizonai/sensei
/plugin install sensei
/reload-plugins

Manual install

Project install, best for code review:

git clone --depth 1 https://github.com/onehorizonai/sensei .claude/skills/sensei

User install, useful when you want Sensei available everywhere:

git clone --depth 1 https://github.com/onehorizonai/sensei ~/.claude/skills/sensei

After installing, restart your agent tool and try /sensei-assess.

---

Usage Guide

Quick start: install it, then use it manually

Use this if you just want better review today.

1. Open your project repo in Claude Code, Codex, Cursor, or another agent tool that can use local skills.
2. Install Sensei in that repo.
3. Ask for the kind of help you need.

Good first prompts:

/sensei-assess
/sensei-spar
/sensei-trace
/sensei-gameplan
/sensei-prove-it

Plain-English prompts work too:

Use Sensei to review my uncommitted changes.
Use Sensei to help me debug this bug.
Use Sensei to check if this plan fits the existing codebase.
Use Sensei to tell me what my tests actually prove.
Use Sensei to help me write the PR description.

If you do not know where to start, use /sensei-assess. It reads the repo state and routes you to the right Sensei skill.

More advanced: automate it locally

Use this when you want Sensei to show up at the right moments without remembering every command.

Local automation means your coding tool nudges or runs Sensei on your machine while you work. Claude Code hooks, Cursor hooks/rules, Codex local workflows, shell aliases, and editor tasks can all do this.

Useful local triggers:

Moment	Run
You start a session with changed files	/sensei-assess
You are about to approve an AI coding plan	/sensei-gameplan
You wrote or changed tests	/sensei-prove-it
You are about to open a PR	/sensei-scroll
You just merged or finished a session	/sensei-reflect

Example idea: