SkillAgentSearch skills...

karni

Karni is an open-source Django + React framework for building with AI coding agents. Your agent writes only your domain. The API, admin, permissions and React screens come from a package you upgrade, not code it rewrites in each project. One rule set covers every coding agent.

Install / Use

claude mcp add cbsshekhawat18-lab -- npx -y github:cbsshekhawat18-lab/karni

If the server publishes to npm under a different name, use that package instead β€” check the repo README.

About this skill
πŸ”Œ

MCP Server

Model Context Protocol server

Quality Score

84/100

Supported Platforms

Claude Code
Claude Desktop

Our assessment of karni

karni scores 84/100 on our quality scale, 470th of 608 Data & Analytics skills we index.

Its MCP Server is 22 KB long, well organised into 13 sections with 6 code examples: a thorough specification that gives an agent plenty to work with.

It has 10 GitHub stars, so there is little community track record yet; judge it on its content.

Substance
30/30
Structure
20/20
Description
15/15
Adoption
4/20
Freshness
15/15

Maintenance, license and trust

  • The repository was last updated 2 days ago, so karni is actively maintained.
  • It is released under the Apache-2.0 license, a permissive license that allows use, modification and commercial use with attribution.
  • Its trust signals score 97/100, with no cautions. These come from repository metadata, not a code audit β€” read the skill file before letting an agent act on it.

karni compared with similar skills

All 4 of these similar skills score higher than karni; compare them before choosing.

SkillScoreStarsUpdatedFormat
karni (this skill)by cbsshekhawat18-lab84102d agoMCP Server
claude-memby thedotmack100117.9ktodayCLAUDE.md
Agent-Reachby Panniantong100112.7k9d agoCLAUDE.md
rufloby ruvnet10082.0ktodayCLAUDE.md
headroomby headroomlabs-ai10079.9ktodayCLAUDE.md

Frequently asked questions

How do I install karni?
Run claude mcp add cbsshekhawat18-lab -- npx -y github:cbsshekhawat18-lab/karni. The install tabs above show the steps for each supported agent.
Which AI agents does karni work with?
It is written for Claude Code and Claude Desktop, as a MCP Server file. Other agents that read the same format can often use it too.
Is karni safe to use?
It is Apache-2.0-licensed and scores 97/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
Is karni still maintained?
The repository was last updated 2 days ago, so karni is actively maintained.
<p align="center"> <picture> <source media="(prefers-color-scheme: dark)" srcset="https://raw.githubusercontent.com/cbsshekhawat18-lab/karni/main/assets/brand/karni-horizontal-dark.svg"> <img alt="Karni" src="https://raw.githubusercontent.com/cbsshekhawat18-lab/karni/main/assets/brand/karni-horizontal.svg" width="280"> </picture> </p> <h3 align="center">The Django + React framework where your coding agent writes only your app</h3> <p align="center"> Karni is an open-source Django + React framework for building with AI coding agents. Your agent writes only your domain. The API, admin, permissions and React screens come from a package you upgrade, not code it rewrites in each project. One rule set covers every coding agent. The build fails when an agent hand-writes what Karni ships. </p> <p align="center"> <a href="https://pypi.org/project/karni/"><img alt="PyPI" src="https://img.shields.io/pypi/v/karni"></a> <a href="https://www.npmjs.com/package/@karni/ui"><img alt="npm" src="https://img.shields.io/npm/v/@karni/ui"></a> <a href="https://github.com/cbsshekhawat18-lab/karni/actions/workflows/ci.yml"><img alt="CI" src="https://github.com/cbsshekhawat18-lab/karni/actions/workflows/ci.yml/badge.svg"></a> <a href="https://cbsshekhawat18-lab.github.io/karni/"><img alt="Docs" src="https://img.shields.io/badge/docs-online-FB5D08"></a> <a href="https://github.com/cbsshekhawat18-lab/karni/blob/main/LICENSE"><img alt="License: Apache-2.0" src="https://img.shields.io/badge/license-Apache--2.0-blue"></a> <a href="https://pypi.org/project/karni/"><img alt="Python versions" src="https://img.shields.io/pypi/pyversions/karni"></a> <a href="docs/faq.md#which-python-django-and-react-versions-does-karni-support"><img alt="Node 20.19 or later" src="https://img.shields.io/badge/node-20.19%2B-339933"></a> </p> <p align="center"> <b><a href="#quickstart">Quickstart</a></b> &nbsp;Β·&nbsp; <b><a href="https://cbsshekhawat18-lab.github.io/karni/">Docs</a></b> &nbsp;Β·&nbsp; <b><a href="docs/recipes/index.md">Recipes</a></b> </p> <p align="center"> <b><a href="https://karni-shop-demo.onrender.com/">Live demo</a></b> (sign in as <code>demo@example.com</code> / <code>karni-password</code>, the first page can take a minute to wake) &nbsp;Β·&nbsp; <a href="https://render.com/deploy?repo=https://github.com/cbsshekhawat18-lab/karni"><img alt="Deploy to Render" src="https://render.com/images/deploy-to-render-button.svg"></a> <br> The <a href="docs/demo.md">shop demo</a>: the shop recipe as one service, with a public demo login and a daily reset. </p>

Why Karni exists

Every web project starts with the same work: sign-in, a user model, the admin, permissions, CRUD endpoints, tables, forms, Docker and Celery. Only after that does the work on the product begin.

Coding agents write that setup faster, but they write it again in every project. Each copy is a little different, with the hard parts wrong in different places, and every copy has to be maintained. Karni grew out of the 34 projects we built before it.

What is Karni

The problem. Every Django + React project needs the same code before it does anything of its own: serializers, permission classes, admin files, sign-in pages, tables and forms. A boilerplate copies it once. A coding agent writes it again in every project. Either way, a fix made upstream never reaches the copies.

How Karni solves it. That code ships as versioned packages: karni (Python), @karni/ui (React) and the karni command. karni upgrade moves a project to each new release, so a fix reaches every project that depends on Karni.

What you get. You declare a model and who may access it. Karni gives you the REST API, the admin, role-based access, the React screens, the tests and the demo data.

What you still write. Your domain: your models, the rules for who may do what, and the logic and pages your product needs. AGENTS.md tells the agent the same, and karni check fails the build when it hand-writes what Karni ships.

Build an app from one line

  1. Create a project with karni new "Yoga Studio" and open it in your coding agent.
  2. Describe the app in one line after /new-app. AGENTS.md tells the agent what that asks for.
  3. The agent writes karni.yaml from the closest recipe, and asks you only the questions karni plan --questions lists.
  4. karni build makes the app: it adds the modules, writes Karni's code, migrates, seeds the demo data the spec asks for, generates the TypeScript types, and runs the tests and karni check.
  5. The agent writes only the domain logic that is left, in the project's own files. It is done when karni check and karni test pass.
/new-app a booking app for a yoga studio, with online payments

What the agent runs, from a real session:

$ karni recipes list --match "a booking app for a yoga studio, with online payments"
booking   4  Booking: Resources with time slots that members book and pay for, never over capacity. (matched: booking)
shop      3  Online shop: Products in categories, a cart, orders and checkout with online payments. (matched: online, payment)

$ karni plan --questions
No questions: karni.yaml settles everything Karni asks before a build.

$ karni build
plan        22 changes, 0 risky: modules 1, features 1, roles 3, apps 1, entities 3, access 3, pages 3, migrations 1, seed 3, tests 3 (--verbose lists each)
modules     karni add payments
files       27 written: 17 Karni's, 10 the project's (pyproject.toml, frontend/package.json, ...)
migrations  bookings/migrations/0001_initial.py
migrate     applied 84 on default
seed        Resource 4, Slot 30, Booking 60 (demo login demo@example.com / karni-password)
types       frontend/src/api/types.ts and hooks.ts
test        7 passed in 2.88s, tests/openapi.snapshot.json updated (the API changed: commit it)
check       ok
Built: the project matches karni.yaml.

On a developer laptop, karni new --recipe shop builds the shop from scratch on SQLite in 14.1 seconds, its tests and karni check included and the frontend install left out (how fast is it). The flow works without an agent too: every step is a command you can run yourself. The recording below builds a project command by command, then shows its welcome page, its admin and its React screens.

<p align="center"> <img alt="A terminal runs karni new, karni make model, migrate, seed and serve. Then the browser shows the welcome page, the admin listing the seeded events, and the React app's events table and detail view." src="https://raw.githubusercontent.com/cbsshekhawat18-lab/karni/main/docs/assets/karni-demo.gif" width="880"> </p>

Coding-agent benchmark

A coding agent built three small apps (a booking app with payments, a CRM with three roles and import, and a shop with orders and refunds) on plain Django + React and on Karni, with the same model, agent version, prompt and API contract. The harness's own acceptance tests decide success.

What the agent built

One kept run per app and side, each passing every acceptance test, on the Karni 0.5.0 release. Both sides are measured the same way, by the project's own Django and its files (scripts/benchmarks/tokens/inventory.py):

<!-- bench:inventory:start -->

| In the finished app | Plain Django + React | Karni | |---|---|---| | API operations (route and method) | 22 to 25 | 71 to 135 | | Sign-in and account endpoints, outside the admin | 1 to 2 | 16 | | Models in the admin | 1 to 2 | 26 to 30 | | React screens | none | 9 to 12 | | Tests the project ships with | none | 7 to 15 | | OpenAPI schema served | no | yes | | Typed API client in the frontend | no | yes | | Production Docker stack | no | yes | | Rules for the coding agent | no | yes |

<!-- bench:inventory:end -->

Why a Karni run costs more. These tasks check only a small API contract. Plain Django meets it and stops there: no screens, no tests, and two sign-in endpoints at most. With Karni, the same request ends in the whole app above, and the agent also reads Karni's rules and runs its build and checks. So a Karni run costs more and takes longer (below), and for that the app has about 3 to 5 times the API operations, a full sign-in, an admin, React screens and tests.

What it cost

45 runs: five per app on plain Django + React, on the Karni 0.5.0 candidate, and on Karni after the work on its agent flow.

<!-- bench:readme:start -->

| Task | Plain Django + React | Karni 0.5.0 candidate | Karni after the agent-flow work | |---|---|---|---| | Booking with payments | $0.19, 61.76 s, 5 of 5 passed, 5 tool calls | $0.45, 208 s, 3 of 5 passed, 26 tool calls | $0.33, 187 s, 5 of 5 passed, 20 tool calls | | A CRM with three roles and import | $0.21, 70.51 s, 5 of 5 passed, 6 tool calls | $0.56, 278 s, 5 of 5 passed, 31 tool calls | $0.40, 236 s, 5 of 5 passed, 22 tool calls | | A shop with orders and refunds | $0.23, 72.21 s, 5 of 5 passed, 7 tool calls | $0.62, 238 s, 4 of 5 passed, 29 tool calls | $0.44, 255 s, 4 of 5 passed, 24 tool calls | | All three tasks | $0.63, 204 s, 15 of 15 passed, 18 tool calls | $1.63, 724 s, 12 of 15 passed, 86 tool calls | $1.17, 678 s, 14 of 15 passed, 66 tool calls |

Each cell: median cost in US dollars, median wall time, runs that passed every acceptance test, and median tool calls. In the last row the runs add up and the other figures are the sums of the task medians, as the benchmark report gives them.

<!-- bench:readme:end -->
  • Plain Django + React was cheaper and faster on these three small apps. Karni after the agent-flow work cost about 1.8 to 1.9 times as much per app (median) and took about 3 to 3.5 times as long.
  • The agent-flow work cut the median cost by 27 to 30 percent and the context the agent read again by 30 to 43 percent, and raised the runs that passed every test from 12 of 15 to 14 of 15.
  • With Karni the agent wrote fewer lines of its own (median 256 to 431 per app, against 438 to 554), and each Karni project ended with far more code in it (median about 11,500 to 20,000 lines, most of it generated by Karni, against about 440 to 570).

Not counted here: setting up. Both sides started from a ready-made project; the harness, not the agent, made plain's Django and React starting point. Without Karni, that setup is done again for every new project. With Karni, karni new does it, and builds what the first table shows, with no AI tokens. A benchmark that starts from an empty folder and a one-line request is next.

The benchmark report has every figure with its range, the method, the limits, the commands to repeat it and the raw data.

What you get

Every feature, grouped by what you need. The features page lists all of them with their status.

<table> <tr> <td width="50%" valign="top">

Start fast

</td> <td width="50%" valign="top">

Data and access

Truncated for display β€” read the full file on GitHub.

Related Skills

View on GitHub
GitHub Stars10
CategoryData
Updated2d ago
Forks6

Languages

Python

Trust signals

97/100

From repository metadata: license, adoption, age and documentation. Not a code audit β€” see the Safety scan above for what the skill file itself contains.

1 info
karni β€” MCP Server: Install & Safety Check | SkillAgent