An agent skill to start new Django projects or extend existing ones. One sentence in, a running project out: packages wired together, dev/prod settings split, CI included.
/seedkit SaaS landing + waitlist, GDPR-friendly stack (mail, analytics, error reporting), VPS deploy
/seedkit add proper auth — magic link, lockout on brute force, optional 2FA
/seedkit look at our repo and tell us what's worth adding next
LLMs write Django from memory, and that memory is a year or two old: deprecated auth settings, last version's Stripe webhooks, database ports open to the local network. seedkit keeps the knowledge in reference files instead — built from package docs, tested end-to-end, fixed after every failure. The model just types.
A boilerplate generator only offers a fixed menu, frozen at the template's last update; a raw LLM writes any stack you ask for, but from year-old memory. seedkit gives you your exact stack with each package's current API looked up at generation, and the output checked against 9 end-to-end scenarios and audited by a second LLM.
The whole test suite runs on Claude Sonnet: a mid-tier model does the scaffolding, and the frontier-model hours from your subscription go to the code only you can write.
Helps you with: Python deps & venvs, settings for dev vs prod, custom user model, social & password login, passwordless magic-link login, brute-force protection, background jobs, async views, WebSockets, Redis caching, S3 for static & media, outbound email, Tailwind without Node, GDPR-safe analytics, security headers, CSP headers, production error tracking, structured logs, N+1 query detection, safe migrations, linting & formatting, type checking, scheduled DB backups, Docker for local dev, auto-HTTPS reverse proxy, CI pipeline — and more.
/plugin marketplace add viewflow/seedkit
/plugin install seedkit@viewflowVia the skills CLI — installs into whichever agent dirs it detects:
npx skills add viewflow/seedkit # project scope
npx skills add viewflow/seedkit -g # global (all your projects)
npx skills add viewflow/seedkit -a cursor # pin to one agentThen, in whatever empty directory you'd like to populate:
/seedkit
Two skills ship in this repo.
/seedkit started from the actual package docs — the reference files have exact package versions, config snippets, and known pitfalls.
/seedkit-slim relies only on what the model already knows — no reference files.
Both go through the same test cycle: generate code, boot it, check it. Failures get fixed in the skill text.
The skill targets the problems that show up in LLM-generated Django code that doesn't run out of the box: stale package versions, outdated patterns, wrong async defaults, missing production wiring. The testcase loop's job is to find those failures and fix them in the instructions, so the next generation doesn't repeat them.
It also saves you the prompt-writing: the instructions that would otherwise need a paragraph per project are baked in.
- A Django project that boots, smoke-checked before hand-off.
- Real alternatives at every step: Celery or RQ, allauth or magic links, VPS or Fly.
/seedkit add [feature]extends an existing project without rewriting it.- Only the code for options you picked — no dead boilerplate.
This is a fresh project under active development. While the skill is verified against the nine core scenarios in seedkit-examples, we are still mapping out how the agent behaves outside that set.
The testcase harness currently runs only against Claude Sonnet. Other models (Opus, Haiku, GPT, Gemini) are not yet covered — they may work, but skill quality on those models is not verified.
Production deployment scenarios (VPS, Fly, GitHub-SSH) still need verification — they are wired up in the skill but not yet exercised end-to-end against real targets.
If you run into issues, strange behavior, or have ideas for new integrations, please open an issue. Feedback is welcome.
This is AI-generated code, and any human attention is valuable — a person reading it catches what the harness can't.
- Hit a bug or something odd? Open an issue — even a one-line "this broke" helps.
- Run it on another model. We only verify on Claude Sonnet. Point
train/run-tests.shat Opus, Haiku, GPT, or Gemini and share the logs; cross-model coverage is what we need most. - Read the output before you trust it. It boots and passes smoke checks, but hasn't seen production — your review is part of the loop.
For anything bigger, open an issue first so we can talk it through. Full test cycles take a couple hours, so it's worth saving each other the wasted run.
MIT — © 2026 Mikhail Podgurskiy.
$ Sorry, you're right. I probably shouldn't have deleted the production database.
Want me to at least write the restore script?