Kalokoby

All guides › By use case

By use case

Review mode and sign-off: decide only what changed

Review a run one step at a time with the keyboard, keep approvals of unchanged steps, compare screens with swipe or onion skin and hand over a signed record of who accepted what.

The agent shared its fourth run of the week. Forty steps, and thirty-five of them look exactly like the run you accepted on Monday. Review mode puts the five that changed in front of you and leaves the rest approved, with your name and the date you approved them.

The problem

Reviewing every step of every run is how review stops happening. People skim, approve in bulk, and the one screen that broke on a phone in German goes through with the rest. When the client later asks who accepted it, the answer is a chat message.

What changes with Kaloko

A step whose screenshots, criteria and automatic checks are the same as in the last accepted run keeps the approval a person gave it there. Everything else waits in review mode: failed and returned steps, criteria the AI was unsure about, steps that changed, steps with open comments. Once you accept the run, Kaloko can issue a signed record of the acceptance.

How it works

  1. Open the review link

    kaloko share prints a line like 12 steps unchanged, approvals carried over; 5 need review with a link ending in #review. On the canvas the Review button or Q opens the same queue.

  2. Decide with the keyboard

    J and K move between steps, A approves, R returns the step with a note, C comments. The accepted run sits beside each step, so you see what moved.

  3. Look closer where it matters

    in the lightbox compare side by side, as a pixel diff, with *Swipe* (drag a divider across both screenshots) or with *Onion skin* (this run faded over the other). G shows the step in every language and viewport at once.

  4. Point at the problem

    a comment can carry a box, an arrow or a line drawn on the screenshot, and an image you paste. A developer or tester can mark a region that changes on every run (a clock, a carousel) so pixel diffs skip it; the agent pulls those regions into the scenario with kaloko stability --pull.

  5. Accept the run

    when every step is decided, accept or return the whole run. Required approvers (Team and up) make a run count as accepted only once each named person has approved it.

  6. Hand over the record

    on Business and Enterprise an accepted run has a signed acceptance record: who approved what and when, the criteria results and the hashes of the screenshots, signed by Kaloko. Download it as PDF or JSON from the run, or with kaloko signoff <run-id>.

Skills and prompts

The agent prepares the review but never approves for a person:

Share the run of TASK-123 and tell me what still waits for me in review mode, with the link.

Expected outcome: the agent runs kaloko share, then kaloko review, and answers with the counts and the #review link.

Download the signed acceptance record of the accepted run of TASK-123 for the client and check that it verifies.

Expected outcome: kaloko signoff <run-id> --out handover/ writes the JSON and the PDF and checks the signature; the agent refuses if the run is not accepted.

What you get

FAQ

Can a carried approval accept a run on its own?

No. It only spares you the unchanged steps. Accepting the run, and every approval a required approver gives, stays a person's decision. Anyone can revoke a carried approval and the step waits again; Business organizations can turn carrying off in Settings.

What does the signed record prove?

That this run, with these screenshots and these results, was accepted by these people at that time. If anyone changes a byte of the JSON or the PDF, verification fails.

Do reviewers need a paid seat?

No. Viewers read, comment, approve and return for free in every plan.

What it looks like

A real run of Kaloko on its own public pages, refreshed daily. This is the canvas your team gets.

10 of 11 steps passed · 2026-10-02Open the run in Kaloko ↗

More guides

Accepting a task with an AI agent, in the pull requestRegression before a release: compare the run with the accepted baselineSEO and landing-page checks on production, read-only, every dayE-mail flows: capture the message next to the screen that sent it

Install once, then work through your agent

Kaloko runs where your code and your agent are. The service stores and versions the results, shows the canvas and collects approvals.

  1. Add the CLI to the project
    npm install --save-dev kaloko

    Needs Node 20 or newer. Update later with npm update kaloko.

  2. Create the config and install the skill
    npx kaloko init --agent claude --org <your-org>

    The skill is copied to .claude/skills/kaloko. npx kaloko doctor checks Chrome, the config and the token.

  3. Create your organization and a token

    Create an organization; you become its admin. The start page offers a tester token in one click, later under Settings → API tokens. Put it into the project .env:

    KALOKO_TOKEN=qwk_…
    TYPESAFE_API_KEY=…   # optional: semantic evaluator

Then just ask your agent

The skill teaches your agent the whole loop: it writes the acceptance plan and the scenario from the task, walks the screens, evaluates, shares the canvas, reads what reviewers said and fixes it. You don’t type the commands; you look at the canvas.

What the agent runs (or run it yourself, e.g. in CI)

The same loop by hand:

npx kaloko start --scenario docs/tasks/TASK-123/qa/scenario.yml --env local
npx kaloko walk        # playwright steps; agent/manual steps: kaloko capture
npx kaloko evaluate
npx kaloko share --pr
npx kaloko feedback    # what reviewers said, with ids to answer