Know what your product still does.
Write what your app should do as sentences. Your coding agent checks what it can in a real browser and records what it saw. A person approves what becomes spec.
A signed-in customer with a saved card can pay without re-entering it.
A discount code lowers the total before the customer pays.
Staging returned 502; not counted as a failure.
- 1
An agent proposes
It reads your product and writes what it should do as plain sentences: features, and the cases under them.
TerminalRead `nmr help seeding`, then propose a spec for this project.Claude DesktopUse the Nimrod connector. Propose a spec for project <project>. - 2
A person approves
Nothing an agent writes is spec until someone on your team approves it. An agent cannot approve anything.
- 3
An agent verifies
It opens your deployed app in a real browser, checks each case, and records a verdict. A pass must name the text that matched, the URL it ended on, or both, or it is stored as blocked.
TerminalRead `nmr help runner`, then execute run <run id>.Claude DesktopUse the Nimrod connector. Execute run <run id> of project <project> in the built-in browser.
Know what is safe to deploy
Every case shows its last verdict and how long ago it was checked, so you see at a glance what still holds.
A spec people can read
Cases are sentences, not test code, so anyone on the team can read what the product promises, not only engineers.
Holds up when the UI changes
A case says what a person should see, not which selector to click. The agent reads the page fresh on every run, so a redesign does not break cases the way it breaks scripted tests.
Nothing to host
Nimrod needs no runner, test server or CI job of its own. The agent session you already use runs the checks.
Why not a markdown file and your agent?
You could. Teams that try it end up with a file nobody trusts, because the agent edits it, nobody knows when a line was last checked, and a dead staging server looks like a broken product. Nimrod adds the three things that file lacks:
An approval gate
The spec changes only when a person says so. An agent proposes; it cannot approve its own edit.
Freshness
Every case shows its last verdict and when it was checked, so you know which lines still hold.
Blocked is not failed
A run that cannot reach your app says so, instead of turning everything red. A staging outage never marks your product broken.
A discount code lowers the total before the customer pays.
Staging returned 502; not counted as a failure.
Set up in two minutes
- 1
Install nmr
One file, put in ~/.local/bin. The script checks the download before it installs anything.
curl -fsSL https://assets.getnimrod.com/cli/install.sh | sh - 2
Sign in
Opens this site in your browser. Approve it, and the machine gets its own token, listed under Agent tokens.
nmr login - 3
Link your repository to a project
Run it in the folder of the product you want to verify. Every command from that folder then uses this project. With no project named, it lets you pick one.
nmr link - 4
Give your agent a job
Paste one of these into the agent session, in the same folder. New features and cases land in Review for you to approve; a run you prepare under Runs records a verdict for each case.
Read `nmr help seeding`, then propose a spec for this project.Read `nmr help runner`, then execute run <run id>.
Nimrod is open to a few teams at a time. Leave your email and we will let you in when there is room.