OneShot

Your code. You own it.

Coded, built, tested, verified.

Pick the hard part of your product. Paste one prompt into Claude Code or Cursor. Your agent builds it, and a test suite tells you when it’s actually done — potentially skipping weeks of iteration. Then check our receipt: we ran the same build ourselves and published exactly how it went, including the packs that pass 2 of 3.

  • You own the code, forever

    We don’t host anything or run anything. Your agent builds it in your repository, and it stays yours if we disappear tomorrow. No runtime dependency on us, no account, no API key.

  • Know the AI bill before you start

    Every pack publishes what the build actually cost us in tokens and dollars when we ran it. Not an estimate — a measurement, on the receipt, before you buy.

  • One price, not another subscription

    Buy it once. No seats, no monthly minimum, nothing that grows with your usage. The alternative to renting this from a SaaS forever is building it — this is how you do that in an afternoon.

  1. 01

    Buy the pack

    You get a folder: what to build, how it should be built, and the tests that decide when it is done. Every hard decision already made.

  2. 02

    Feed it to your agent

    Paste the pack’s prompt as your first message and leave it running. It works through the build itself, using a cheaper model where it can.

  3. 03

    Verify

    Run one command. Every check prints pass or fail. Green means done — and you can compare it against the receipt we published before you bought.

Catalog

How receipts work

Before a pack is listed, our verification harness executes it the same way you will — a fresh agent, a clean scaffold, the pinned model. The receipt is the machine-generated record of those runs, published in full.

Pass rate, as M of N
Our harness runs each pack from a clean scaffold N times on the pinned model and publishes exactly how many runs passed. Agent builds are stochastic — a pack that passes 4 of 5 is listed as 4 of 5, never as "always one-shots."
Everything pinned
The exact model id and exact dependency versions the runs used are locked into the receipt. What was verified is what you build with.
Measured cost
Token counts and wall time in the receipt are per-run means from real runs. The token figure on a catalog card is an estimate and is labeled as one.
Tamper disclosure
The receipt lists every scaffold file the building agent modified or deleted. An empty list means built pristine. Modifications are disclosed, not hidden.
Pending means pending
A pack without a completed verification run says "Verification pending" and publishes no numbers. We never invent them.