See it run
Current Django knowledge and best practices for coding agents
Seedkit is a skill for Claude Code, Cursor, and other coding agents. Tell it what you must build. It asks you a few questions, then writes the project.
Measured result
Sonnet with the skill scores above Fable and Opus without it.
The generated projects pass more production checks. To measure that, the same nine projects were generated more than once. Sonnet ran with the skill and without it. Fable and Opus ran without it. A second model then ran the same eight security and configuration checks on every result. With the skill, all nine projects passed all eight — 72 of 72. Without it, a quarter of the checks failed.
Control arms: Fable 54 of 72 on all nine projects, Sonnet 33 of 48 on six, Opus 29 of 32 on four. One run for each project and model, Claude models only. The numbers show that a difference exists. They do not rank the models. Full method →
Comparison
A cookiecutter gives you one fixed set of packages, and you delete the parts that you do not need. Seedkit knows more than 50 Django packages, and most topics offer more than one option. It asks what you must build, then writes only the code for the options that you select.
Background jobs on Celery, RQ, or the database. Static files on S3 or WhiteNoise. A deploy to a VPS, to Fly.io, or over SSH. You receive the code for your answers, and no other code.
Seedkit writes a normal Django project once. It adds no wrapper library, and your project has no dependency at run time.
Seedkit resolves each version when it writes the project. A template gives you the versions from its own release date.
Give the command add [feature] in your repository. Seedkit adds the dependency, the settings, the .env example, and the CI step.
What it adds
Each file has the packages, the settings, and the templates for one topic.
Together they cover every step, from an empty directory to a deploy. They also
cover the small items, such as robots.txt and the sitemap. Seedkit
writes only the files that your answers select.
Install
In Claude Code:
claude plugin marketplace add viewflow/seedkit claude plugin install seedkit@viewflow
In Cursor, Codex, OpenCode, or Gemini CLI, use the skills CLI:
npx skills add viewflow/seedkit
Then, in an empty directory:
/django-seedkit
Why a skill and not a prompt
The model writes Django from memory, and that memory is one or two years old. So it writes an authentication setting that Django deprecated, or last year's Stripe webhook. Or it opens the database port to the local network. You can correct all that in your prompt — if you already know the answers. Seedkit already knows them, and it is that prompt, written and tested.
A reference file keeps only the part of the package documentation that the model does not know. Today's setting names, today's package names. One file for each topic: authentication, payments, storage, deploy.
You do not have to know what to ask for. Which package locks an account after failed logins? Which settings does a production deploy need? The files have the answer, and the skill offers it to you.
More than 100 hours of tests found what the model gets wrong. Each file goes into a generated project. We start it, read the result, then correct or drop what failed. What stays in the file is the part the model does not know. You skip the search.
FAQ
Yes. A cookiecutter gives you one fixed set of packages, and you delete the parts that you do not need. Seedkit asks what you must build, then writes only the code for the options that you select.
Yes. Install it into Cursor, Codex, OpenCode, Gemini CLI, or any agent that the skills CLI supports. Use the command npx skills add viewflow/seedkit.
No. Seedkit writes a normal Django project once. It adds no wrapper library, no dependency at run time, and nothing more to upgrade later.
Any model can run the skill. The knowledge is in the reference files, so the model does not supply it. Sonnet with the skill scored 72 of 72, so a mid-tier model is enough. Keep the expensive models for the code that only you can write. Our tests run on Sonnet, so other models have less coverage. Report a problem in the GitHub issue tracker.