Send your first coding screen

Create a screen

Open Screens in the sidebar and choose New screen. Give it the name candidates will see, such as Rate limiter, in the Name this screen field.

In the settings on the right, pick a Time limit. 90 minutes is the default and suits most backend tasks; you can also choose 30 minutes to 4 hours, or a custom length. Once the draft exists, your changes save as you go.

Add a challenge from your repository

Paste the repository URL into Repository URL and choose Check. Kendor previews the import: the number of files, its size, and whether the repository already has a kendor.yaml.

• Hosts: github.com, gitlab.com and bitbucket.org, plus your own GitHub Enterprise Server, GitLab, Gitea or Forgejo.
• Private repositories: an organization owner adds a read-only token under Integrations → Code hosts first. Tokens are stored encrypted, and Kendor only ever shows their last four characters.
• One folder of a monorepo: paste a link to a branch or folder (for example github.com/acme/platform/tree/main/services/billing) and only that part is imported.

Set the Title, Language, and any Databases the project needs, then choose Add challenge. The import is a copy of that commit: later pushes to the repository don't change a screen that's already out.

No repository? Under Or add:, choose Blank project to start from one of Kendor's starters, or Single file for an algorithm-style question.

Check how it runs

A kendor.yaml at the root of the project tells Kendor how to install, start and test it. When the repository doesn't have one, the editor opens Runtime configuration for you. Fill in the form, or switch to Raw YAML, and save: Kendor writes kendor.yaml and restarts the sandbox with it.

A small project with a PostgreSQL database looks like this:

version: 1
runtime:
  type: python
  version: "3.12"
databases:
  - type: postgres
    version: "18"
    init: db/schema.sql
install:
  commands:
    - pip install --user -r requirements.txt
services:
  app:
    dev:
      command: uvicorn main:app --host $HOST --port $PORT --reload
      port: 8000
    test:
      command: pytest --tap-stream
preview:
  service: app

version: 1
runtime:
  type: node
  version: "22"
databases:
  - type: postgres
    version: "18"
    init: db/schema.sql
install:
  commands:
    - npm ci
services:
  app:
    dev:
      command: node --watch src/server.js
      port: 3000
    test:
      command: node --test
preview:
  service: app

Kendor starts the database inside the sandbox and passes its connection string to your app as DATABASE_URL. Choose Preview as candidate to see the running app exactly as a candidate will.

Make sure the tests run

Candidates run the project's tests themselves, from the Tests tab in the editor or from the terminal, as often as they like. The tests are ordinary files in the project, so candidates can read them too.

The Tests tab runs the test command from kendor.yaml. To get one row per test instead of a single pass or fail, have the command print TAP: pytest --tap-stream (Kendor's Python ships with pytest-tap) or Node's built-in node --test both do. Run the suite once in the preview to check it.

Launch the screen

Back in the screen's settings, add Default reviewers and check the Invite expiry (7 days unless you change it). To let candidates use Claude Code or Codex in the terminal, turn on AI assistants in the terminal; it runs on your organization's own API keys.

The Launch checklist shows what's missing. A screen needs a title, at least one challenge, and a valid runtime configuration. Choose Launch and the screen goes Live.

Invite a candidate

Open Candidates and choose Invite candidate. Enter one or more emails, pick the screen (only live screens are listed), and choose Send invite.

Each candidate gets their own link by email. They don't need an account: they enter their name, and the timer starts when they choose Begin screen. If you'd rather send the link yourself, use Copy invite link in the candidate's row.

Review the submission

When the candidate submits, open them from Candidates and choose Open review. You'll find their code and its diff under Solution, plus the Timeline, Terminal, AI agents and Activity tabs.

Each reviewer records a recommendation, from Strong yes to Strong no, with an optional note. Then someone makes the call: Advance or Reject. Neither emails the candidate, so you decide when and how they hear back. Mark passed and Mark did not pass are different: they email the candidate their result straight away. See Review a submission.