Claude Code knows how to code.It doesn't know what to build.
972 agents, skills and commands — plus a 9-source data engine that measures real demand before a line is written. And it starts at 2,900 tokens of context, not 51,543.
One payment · lifetime updates · 14-day money back · no subscription
| keyword | demand | comp | opp | strongest incumbent |
|---|---|---|---|---|
| habit tracker: kabi | 55 | 5 | 52.9 | Habit Tracker: Kabi (1 ratings) |
| habit tracker: coolio habits | 58 | 40 | 40.2 | Habit Tracker: Coolio Habits (4 ratings) |
| habit tracker free | 94 | 100 | 21.7 | Habit Tracker (144,806 ratings) |
| habit tracker widget | 88 | 100 | 20.3 | Habit Tracker (144,806 ratings) |
| habit tracker with friends | 82 | 100 | 18.9 | Habit Tracker (144,806 ratings) |
| habit tracker plant | 76 | 100 | 17.5 | Avocation (954 ratings) |
One real result
This is what “measured demand” looks like on the page.
Not a testimonial about how good the kit is — the actual JSON the engine wrote, evidence link included.
A PDF edit / fill / sign assistant — built for iOS users searching “pdf editor”
The engine did not invent this. It found the gap on Google Play, where pdfFiller ranks #9 for “pdf editor” while the iOS side stays weak — then attached the link so you can check it yourself:
play.google.com/store/apps/details?id=com.pdffiller ↗iOS side is weak
still rising
few complaints found
willingness unproven
And it says the uncomfortable part out loud — “willingness to pay is still weak (money score 50)” — instead of selling you the idea. That sentence is the whole point: a scorecard that can say no.
The problem
Your agent is fast. That was never the bottleneck.
You still have to guess what to build
Your agent will happily build whatever you name. It has no idea whether anyone wants it. Most projects die here — not from bad code, from building something nobody was searching for.
Prompt packs eat the window they promise to save
A kit that loads its whole catalogue costs you tens of thousands of tokens on every session, whether you use one agent or none. The thing sold as the cure for context spirals causes them.
A pile of agents is not a process
A hundred specialists with no order between them means you decide, every session, which to call and when. That is the work you were trying to hand over.
Measured, not claimed
What a big kit really costs your context window.
We measured every file we ship. Loading all 972 items at once costs 51,543 tokens — 26% of a 200K window before you type a word. That is the context spiral, caused by the kit meant to fix it.
So the catalogue ships as 13 opt-in packs. You enable the domains you are working in; an unused pack costs nothing, because it is never loaded.
| Pack | What's in it | Items | Context |
|---|---|---|---|
| appfactory | The factory line + the 9-source data engine | 57 | ~2,900 |
| mobile | iOS, Android, Flutter, store submission | 7 | ~470 |
| database | Postgres, Mongo, Redis, schema, query tuning | 25 | ~1,160 |
| product-growth | SEO, pricing, onboarding, retention | 27 | ~1,540 |
| security | Auth, OWASP, secrets, threat modelling | 58 | ~3,280 |
| docs-workflow | Docs, git, planning, architecture | 61 | ~3,190 |
| backend-api | REST, GraphQL, gRPC, microservices | 61 | ~3,620 |
| quality-test | Test strategy, debugging, review, refactor | 84 | ~4,420 |
| web-frontend | React, Vue, Svelte, Tailwind, a11y | 88 | ~5,420 |
| data-ai | Pipelines, RAG, embeddings, MLOps | 92 | ~5,480 |
| languages | Python, TS, Go, Rust, Swift + 30 more | 100 | ~5,430 |
| devops-cloud | Docker, K8s, Terraform, AWS/GCP, CI/CD | 128 | ~7,420 |
| general | The specialists that fit no single domain | 184 | ~7,210 |
Numbers come from the shipped files, and the script that produces them is in the package. Run it yourself and check us.
Curated, not dumped
2,810 files scanned. 1,880 rejected.
Much of this catalogue starts as open-source work — MIT and Apache-2.0, credited per file. We are telling you that on the sales page, because you would find out anyway. What you are paying for is the part nobody wants to do:
- 897could not load — no usable frontmatter
- 642documentation, not an agent
- 188exact duplicates across collections
- 56near-duplicates — python-pro, python-expert and python-engineer are one agent
- 40stubs under 800 bytes
- 27not written in English
- 12padding — an agent re-listed as a slash command to inflate the count
- 18would have shadowed a factory-line agent
What survives is what you would have kept anyway — minus the four afternoons of finding out which ones those were.
The difference
Every other kit is prompts. This one has an engine.
Underneath the agents runs a dependency-free Python engine that reads 9 live sources and turns them into a demand and competition picture. Not a prompt asking a model to imagine market data — actual HTTP calls, actual numbers, reproducible.
Measured keyword demand
App Store autocomplete mined for what people actually search, with competition counts beside it.
Saturated markets killed
An idea facing entrenched competitors is eliminated automatically — before it becomes your next three weeks.
iOS gap analysis
Where demand exists and the shipped apps are bad, measured from low-star review mining.
.com availability verified
Checked against the registry, not guessed by a model that will cheerfully invent an available domain.
The line
Four gates, in order, with a real NO at the second one.
Find
The engine reads 9 live sources and returns niches with measured demand — not brainstormed lists.
Decide
A scorecard with a real NO. Saturated markets are killed here, before you write code.
Build
Milestone by milestone, each one followed by an adversarial QA pass that tries to break it.
Ship
Store listing, ASO or SEO, legal pages, submission. The last mile that usually stalls.
Before you buy
What this is not.
Not a guarantee that your app sells
The engine finds measured demand. Demand is not the same as your execution, and nothing here removes that risk.
Not secret code
Much of the catalogue is open source and we name every source. The curation, the pack split and the engine are what you are buying.
Not a replacement for judgement
It presents evidence and a scorecard. You still make the call, and you should sometimes overrule it.
Not a hosted service
It is files on your machine. No account, no licence server, nothing to log into.
Pricing
One kit. One payment. $99.
- ✓All 972 agents, skills and commands
- ✓All 13 packs — enable what you need, when you need it
- ✓The 9-source data engine
- ✓Lifetime updates
- ✓Unlimited commercial and client use
- ✓14-day money back
For comparison, priced the same week: AgentKit Engineer $99 · AgentKit Bundle $149 · MakerKit $299 · Bedrock $499. We are not the cheapest thing on that list and we are not trying to be — we are the only one that publishes what it costs your context.
Questions people actually ask.
Most of this is on GitHub for free. Why pay?
Because we say so on the box: much of the catalogue comes from MIT and Apache-2.0 collections, every file names its source, and ATTRIBUTION.md lists all of them. What you're paying for is the 1,880 files we threw away out of 2,810, the deduplication, the pack split that keeps this out of your context window, and the data engine — which is ours and exists nowhere else. If you'd rather assemble it yourself, the sources are listed. That's a real option and we're not going to hide it.
Won't 972 items destroy my context window?
That is exactly why the catalogue is split. Everything loaded at once costs 51,543 tokens — 26% of a 200K window before you type anything. So it ships as 13 opt-in packs: the factory line plus the engine costs about 2,900 tokens, and a typical setup with two domain packs is around 10,000. An unused pack costs nothing because it is never loaded. Every number here is measured from the shipped files, and the script that measures them is in the repo.
What is the data engine, exactly?
A dependency-free Python engine that reads 9 live sources — App Store autocomplete and low-star reviews, Google Play, Reddit, Hacker News, GitHub, Product Hunt, Stack Exchange, Google Trends — and turns them into a demand and competition picture: measured keyword demand, saturated-market elimination, iOS gap analysis, .com availability. Prompts cannot do this; it is HTTP calls and analysis. It is the reason the kit can tell your agent WHAT to build instead of only how.
How is this different from every other agent kit?
Other kits make the coding agent write better code. That is useful, and this one does it too — 420 agents' worth. The difference is the step before: a kit that hands you 100 prompts still leaves you guessing what to build. This one measures demand first, kills the idea if the market is saturated, then builds and ships it. On raw counts it is 972 against a typical kit's 248; on context cost it starts at 2,900 tokens instead of loading everything at once.
What do I actually get, and how do I install it?
Access to the private GitHub repository — 13 plugin manifests, 420 agents, 323 skills, 229 commands and the Python engine. After paying you connect your GitHub account on the receipt page, accept the invite, then run two commands in Claude Code: /plugin marketplace add 247pro/appfactory and /plugin install appfactory@appfactory. Updates arrive with git pull rather than a fresh download. No licence server phoning home — once installed it runs entirely on your machine.
Does it work with anything other than Claude Code?
The agents, skills and commands are plain markdown with frontmatter, so most of it is portable to any agent that reads that format. The plugin manifests and slash commands are Claude Code's contract specifically. The Python engine runs standalone from a terminal, independent of any agent.
Is it a subscription?
No. One payment, yours permanently, updates included. Nothing to cancel, no seat count.
Can I use it for client work?
Yes — unlimited personal and commercial projects, including billed client work. The one thing you may not do is repackage the catalogue and sell it as a competing kit.
Do the bundled open-source files have proper licences?
Yes, and it matters more than it sounds. Everything bundled is MIT or Apache-2.0 — both permit commercial redistribution — and every file carries a comment naming its source repository, licence and copyright holder. ATTRIBUTION.md lists each project, its licence and how many files came from it. The kits that do not tell you this are the ones to worry about.
What if it isn't for me?
Message us first — most of what feels broken is a five-minute fix and we would rather fix it than take your refund. If it genuinely is not right, you get your money back. No forms.
Who is behind this
Emrah Sinekli
I build and ship my own products with this kit — mobile apps and web tools. The factory line exists because I needed it: choosing what to build was costing me more than building it. Every number on this page is measured from the files I ship, and I answer support myself.
Stop guessing what to build.
972 agents, skills and commands, 9 live data sources, 2,900 tokens to start.
Get the kit — $99