/setup

Set up Cartwalk on any machine, for any store

Five minutes of setup, then one command. It needs Node, Kane CLI and Chrome. It was built and tested on Windows 11; macOS and Linux should work but I have not run them.

  1. Install what it needs

    Node.js 20 or newer

    Required. Runs Cartwalk itself.

    Kane CLI

    Required. The shopper. You need a free TestMu AI account; log in once.

    npm install -g @testmuai/kane-cli
    kane-cli login

    Google Chrome

    Required by Kane. If it is installed somewhere unusual, point Kane at it with the KANE_CLI_CHROME_PATH environment variable.

    Claude Code CLI

    Only for the fix loop. Install it and log in so that `claude` works in a terminal. Not needed for check, report, verify or tamper.

    Check Kane works: kane-cli --version should print a version. Kane's free tier gave me 10,000 credits; a walk costs about 25 to 50.

  2. Get Cartwalk

    No install (downloads from GitHub the first time you run it):

    npx github:ibnweb3/cartwalk help

    Or clone it, if you want to read or change the code:

    git clone https://github.com/ibnweb3/cartwalk
    cd cartwalk
    npm test

    In the examples below, cartwalk means either npx github:ibnweb3/cartwalk or, inside the clone, node bin/cartwalk.js.

  3. Walk your store

    Point it at any store's address, including one running on your own machine:

    cartwalk check https://your-store.example

    A walk usually takes 3 to 7 minutes. Kane opens Chrome headless, reads the shipping banner, adds the most expensive product to the cart, opens the cart and presses Checkout. Here is real output from a walk of Saleor's public demo:

    round 1: PASS  (9 steps, 250s, 53 credits)
      PASS  reached_checkout: the shopper reached checkout, which asks for contact details before payment; Cartwalk does not fill those in, so payment itself was not probed
      PASS  step_budget: 9 steps (budget 20)
      PASS  price_math: 1 x $300.00 = $300.00
      PASS  shipping_promise: subtotal $300.00 is over $75.00 and shipping is free
  4. Read the result

    PASS / FAIL
    The check was decided by plain code. A FAIL names the numbers or Kane's evidence.
    n/a
    The check was not triggered (for example, a cart under the free-shipping threshold).
    ???
    Unknown: Kane did not report the value, so nothing is claimed.
    exit code
    0 pass · 1 a check failed · 2 inconclusive or could not run. Use it in a script or pipeline.
    cartwalk-log.jsonl
    Written in the folder you ran it from (change it with --log file). It holds the instruction, Kane's summary and values, and the verdicts.
  5. Make a report and check the log

    cartwalk report cartwalk-log.jsonl --out report.html
    cartwalk verify --log cartwalk-log.jsonl
    cartwalk tamper cartwalk-log.jsonl

    report writes one HTML file with a button that re-checks the hash chain in the browser. tamper edits a copy of the log, never the original, and shows verification catching it.

  6. Let the coding agent fix it (optional)

    For a Node store that reads its port from the PORT environment variable. Commit your work first: the agent edits files in --fix-dir and its diff goes in the log.

    cartwalk loop http://localhost:3000 --fix-dir ./my-store --serve ./my-store/server.js

    Cartwalk starts your store, walks it, and on a failure hands the failed checks and Kane's evidence to claude (read and edit tools only, no shell) inside --fix-dir. It restarts the store and walks again, up to 3 rounds (--rounds). For other stacks, use check and apply fixes yourself.

    Try the loop on the bundled shop first. node bin/cartwalk.js loop http://localhost:4180 --fix-dir store --serve store/server.js from the clone reproduces the recorded run in about 10 minutes.
  7. Use it in a pipeline

    The exit codes are the interface: 0, 1 or 2. The runner needs Node, Kane CLI (logged in) and Chrome. I have not run Cartwalk in CI yet, so treat this as untested.

Safety and cost. Cartwalk never pays and types nothing into checkout pages, but it does add one item to a cart on the store you point it at, so use a staging or test store when you can. Each walk spends about 25 to 50 Kane credits. The log stays on your machine and contains Kane's description of your store.
/options

Options

--log <file>
Where to write the log.
--timeout <sec>
How long Kane gets for one walk. Default 420.
--steps <n>
Most steps Kane may take. Default 30.
--budget <n>
Step budget for the step_budget check. Default 20.
--rounds <n>
Most walk and fix rounds in a loop. Default 3.
--fix-dir, --serve
Loop only: the folder the agent may edit, and the Node file that starts your store.
/troubleshooting

When something goes wrong

Result is inconclusive
Kane did not finish. Its service can be slow: a healthy walk took between 2.5 and 4.5 minutes for me, and once it did not finish in 7. Run it again, or raise --timeout. A run that ran out of time is not retried and never counted as a store failure.
Kane CLI was not found
Run npm install -g @testmuai/kane-cli, then kane-cli login.
Chrome not found
Install Google Chrome, or set KANE_CLI_CHROME_PATH to its full path.
"something is already serving"
Another program is on that port. Stop it, or use another port.
FAIL on a store that works
Open the log and read Kane's summary. If it describes a real problem, it is real. Cartwalk does not fill in contact details, so it counts a contact-details checkout as reached, not as a failure.
"not a web address"
Use a full address starting with http:// or https://, with no spaces or shell symbols.
Tested on: Windows 11, Node 24, Kane CLI 0.10.6, Chrome 158. Not tested on macOS or Linux.