Changelog

The interview editor now runs a real language server for the room's language, the same kind your IDE uses. Candidates get completions, hover docs and signature help while they type, so nobody has to remember whether it's push or append under pressure.

• Works in TypeScript, JavaScript, Python, Rust, Go, C, C++, Java, PHP, Ruby and Bash.
• Snippets and auto-imports apply properly, so accepting a completion fills in arguments instead of pasting placeholder text.
• Switching the room's language swaps the language server right away.
• Heavier servers like Rust and Java take a few seconds to warm up. The editor shows a small "Loading autocomplete" pill until they're ready.

Removing a question from an assessment, changing its function signature or deleting a test case could wipe out results for candidates who had already taken it. That's fixed.

• Removing a question now archives it. Past submissions still show it with a "Removed" badge.
• Each graded result keeps a copy of the test it ran against, so later edits can't change a past score.
• The question editor tells you when a question is shared across assessments or already has candidate answers, and asks before a change would clear its tests.
• Library questions that are in use can't be deleted.

When a candidate ran out of time, their attempt was marked expired without being scored. Anyone who had answered questions ended up with a blank result.

• Expired attempts are now scored on whatever the candidate submitted.
• The editor saves the candidate's last changes a few seconds before the deadline, so nothing typed at the end gets lost.
• Submitting right as time runs out moves the candidate to the completed screen instead of leaving the dialog stuck open.
• The per-question button is now "Submit answer", so it's clear it doesn't end the whole assessment.
• The invite dialog and API now tell you if the invitation email failed to send.

A few customers told us CoderScreen felt slow when building tests. We went after the causes.

• Database queries now go through a connection pool close to our servers, which cuts time off nearly every request.
• Added missing indexes and removed repeated queries behind the assessment, submission and invite pages.
• Adding, editing or reordering test cases updates the page in place instead of refetching the whole assessment.

CoderScreen now has a public REST API. Create a key from the new API page in settings and you can list interviews, create assessments, invite candidates and pull results from your own scripts or ATS.

We also host an MCP server on top of the same API, so assistants like Claude or Cursor can do those things for you. The settings page gives you a config you can paste straight into your MCP client, with your new key already filled in.

• Keys belong to your organization and you can revoke them at any time.
• Responses use a stable public format that won't change when we change things internally.

• The AI chat in interview rooms now offers Claude Haiku 4.5, Claude Sonnet 4.5, GPT-5 Mini and GPT-5, replacing older models that were being retired. GPT-5 Mini is the new default.
• If the shell in a room's terminal exits (say someone types exit), the terminal starts a fresh session instead of getting stuck reconnecting.

Three bugs from customer feedback, all fixed:

• The editor highlighted every language as TypeScript. Python, Go, Rust, Java, C, C++, PHP and the rest now get their own highlighting and indentation.
• Bash and Java files didn't run at all. They do now.
• If a room's code runner was restarted mid-interview, Run could fail until you started a new interview. The runner now restores your files before each run.

Live interviews are great once you've got a shortlist. Assessments help you get there. Build a coding test, send it to candidates by email, and they take it on their own time while CoderScreen grades their code.

• Build assessments from a reusable question library, and give each question its own weight.
• Write test cases once and they're graded in 11 languages. Mark some as hidden so candidates can't code to the answer.
• Set an overall time limit and per-question limits.
• Candidates get a dedicated test-taking flow and a confirmation email when they finish.
• Review weighted scores for every submission, then archive the ones you're done with.

Read more about Assessments.