Getting Started with cursor-kenji

August 4, 2026 · View on GitHub

A plain-language guide for non-technical users and Cursor beginners.


What is this?

cursor-kenji is a collection of "skills" for Cursor — the AI-powered code editor.

Think of skills like apps on a phone. You install them once and they're there when you need them. When you type something in Cursor chat — like "audit my security" or "make this page look better" — the matching skill activates and tells the AI exactly how to do that job properly.

Without skills, the AI does its best. With skills, it follows a documented step-by-step workflow.


Step 1: Install Cursor

If you don't have Cursor yet: download it at cursor.com. It's free and it's basically VS Code with AI chat built in.


Step 2: Install cursor-kenji

Open a terminal (on Mac: press Cmd+Space, type "Terminal", press Enter) and paste:

npx skills add kensaurus/cursor-kenji

If you get "command not found", run npm install -g skills first, then try again.

Alternative — no npm:

git clone https://github.com/kensaurus/cursor-kenji.git
cd cursor-kenji
./install.sh

Or with npx directly:

npx @kensaurus/cursor-kenji

Step 3: Restart Cursor

Close and reopen Cursor. That's all — the skills are now active.


Step 4: Use a skill

Open Cursor, open a project, and type in the chat. You don't have to remember skill names. Just describe what you want.

Bundled workflows (do the most work with one phrase)

These chain multiple skills into a single tracked loop:

Type this in chat…What happens
build a feature: user notificationsSpec → TDD → implement → smoke test → PR — the whole loop
fix this Sentry error and ship itTriage → reproduce → fix → verify → PR → resolve issue
is this ready to ship?Adversarial test + security + bundle + perf → go/no-go verdict
prepare the app for launchSEO + PWA + bundle + quality gate + deploy smoke → launch checklist
I'm new to this repo, orient meReads the codebase and produces a 5-minute briefing

Individual skills

Type this in chat…What happens
grill me about this planThe AI interviews you — one question at a time — until you're both sure what to build
pin down our terminologyBuilds a project glossary (CONTEXT.md) so the AI stops using the wrong words
resolve the merge conflictsTraces each conflict back to why the code was written, resolves with intent, re-runs checks
audit my app's securityScans for OWASP vulnerabilities, checks auth, flags secrets in code
make this page look betterImproves layout, spacing, hierarchy — like a designer would
commit my changesWrites a proper conventional commit message for you
audit my database schemaChecks naming, indexes, RLS policies, data types
deploy my npm packageWalks through Changesets → CI → publish, step by step
write a PRCreates the pull request with a proper title and description
red team this appAdversarial sweep — UX, data pipeline, security, performance
make the app feel alive with motionAdds coherent, accessible animations that match your design system
improve this formAccessible labels, real validation, error/success states, multi-step flows
clean up our design systemConsolidates drifted colors/components into one source of truth
set up guardrails so AI doesn't break thingsInstalls pre-commit + CI checks against secrets, bugs, and risky ops
is my app production-ready?Checks timeouts, retries, idempotency — the reliability the happy path skips

The AI picks the right skill automatically based on what you typed.


A typical session, start to finish

The single biggest cause of bad AI output isn't bad code — it's the AI building the wrong thing. This loop prevents that:

1. Get oriented (new repo only)

I'm new to this repo, orient me

The AI reads the codebase and gives you a 5-minute briefing.

2. Get grilled before any code

/grill-me I want to add a referral program

Instead of guessing, the AI interviews you — one question at a time, each with a recommended answer so you can just say "yes" or push back:

"Should a referral reward fire on signup or on first payment? I'd recommend first payment — it prevents signup-farming. Agree?"

It looks up facts in your codebase itself and only asks you the decisions. Nothing is built until you confirm. The session ends with a decision log.

3. Build from the decisions

build the feature from those decisions

The workflow-build-feature loop takes over: spec → failing test → code → smoke test → PR. Because you were grilled first, the spec matches what you actually meant.

4. Hand off when you stop

Long session? Context getting full? Type:

/handoff finish the referral UI tomorrow

You get a compact handoff document (what's done, what's verified, exact next steps, which skills to invoke) saved outside your repo. Tomorrow, paste its path into a fresh chat and continue where you left off — no re-explaining.

The rhythm: orient → grill → build → prove → hand off. Skip the grilling and you're back to the AI guessing.


Updating

To get the latest skills:

npx skills add kensaurus/cursor-kenji

Running the same command again overwrites with the latest version.


Frequently asked questions

Do I need to configure anything? No, for most skills. Some skills use external services (Sentry, Supabase, Langfuse) — those need API keys in your environment. The skills tell you when that's needed.

Will this slow down Cursor? No. Skills are just text files. They're loaded by the AI only when relevant.

Can I delete skills I don't need? Yes — delete any folder from ~/.cursor/skills/. The skill is gone.

Can I add my own skills? Yes. See CONTRIBUTING.md for the template.

Does this work with Claude, GPT, etc.? cursor-kenji is built for Cursor's agent system. Skills are text files, so the format is readable by any model Cursor supports.

Is it free? Yes, MIT licensed.


Planning skills — audit first, fix after you approve

17 plan-* skills run before you change code. They produce burndowns and phased roadmaps. You approve each phase, then run the matching execution skills.

Full guide: PLAN-LOOPS.md — grouped loops (not one mega-chain):

LoopWhen to run
Six-skill loopUI/IA hardening on an inherited codebase
Pre-launch hardeningSecurity spine + dependency provenance
Observability & spendSentry/Langfuse gaps, LLM cost caps
Mobile gateCapacitor native security, then App Store / Play paperwork
Growth gateAnswer-engine (AEO) visibility
Authenticityplan-antislop — prose/visual/code slop pass

Quick start (six-skill loop only):

  1. plan-uiux-unification — UI/UX + design system
  2. plan-stub-checker — dead buttons, fake data, unwired handlers
  3. plan-test-coverage — user stories → test matrix, fake-green gaps
  4. plan-perf-audit + plan-security-audit — performance + security (parallel)
  5. plan-docs-sync — docs match reality (last)

Say in Cursor:

Run the six-skill plan loop — plan only, no fixes until I approve each phase.

For security or launch prep, see the other loop prompts in PLAN-LOOPS.md.


Get help


More from @kensaurus

Other free apps and tools from the same studio — all built with these skills. Full list with descriptions → README § Also by @kensaurus.

Mushi Mushi — know when users hit a bug your monitoring missed

npx mushi-mushi

Shake-to-report widget → AI-classified bug reports → optional AI draft PR fix. Free tier 1,000 reports/month, MIT SDK.