September Week 4 · Goodfit

Questions grouped by skill with a score for each, a different mix of questions for every candidate in both assessments and AI interviews, Bash added to coding assessments, and an automation builder that asks when your sentence leaves something open

Assessments

Group questions by skill, and ask every candidate a different mix

One total score doesn't tell you whether someone is strong at Excel and weak at customer communication. And when hundreds of people sit the same test, the questions get shared.

In skill assessments

Put questions into skills. In the question list, open the menu and pick Add skill head. Name it (say, Excel) and add questions inside it, straight from Add question or the Question bank.
Ask a few from each skill. Click Asks all 5 · change at the bottom of a skill and set how many to ask. Every candidate gets a different mix, so answers are harder to pass around.
The report scores each skill separately. A Skills section at the top of the report shows points and a percentage per skill, then the questions under each one.

A question list with two standalone questions, then a skill called Excel holding five questions with 'Asks 2 of 5 · change' at the bottom, and a closed skill called Customer communication with 4 questions — Two questions everyone answers, then two of five from Excel.

The same list with both skills closed. Excel shows a highlighted 'asks 2 of 5' label, Customer communication shows '4 questions' — Close the skills to skim a long assessment at a glance.

In AI interviews

Set it per competency. In the interview builder, click Asks all 5 · change under a competency and pick how many to ask.

Six views of a competency card: asking 1 of 3, asking 2 of 5 open and closed, a single question with no option to change, and a red warning reading 'Ask count is higher than the number of questions in this competency.' — One line under each competency. Ignore it and nothing changes.

Assessments

Coding assessments now support Bash

Hiring someone for a Systems Engineer role? Test them on the kind of work they actually do: reading logs, pulling out columns, counting and sorting text.

New Bash questions in the bank. 50 of them, ready to drop into an assessment.

A Bash question's diagram: a table of log lines with dates, codes and messages, where the two codes starting with E are highlighted — Questions look like the logs and files the job actually involves.

Automations

The automation builder asks instead of guessing

Describe a rule in a sentence and the AI fills in the trigger, the conditions and the actions. When your sentence leaves something open, it now asks you instead of picking for you.

One question at a time, with the answers ready to click. Say "reject anyone whose AI interview score is too low" and it asks where the line is, offering 3, 4 and 5. Pick one and the rule updates in place.
Each question points at the card it will change. Trigger, Conditions or Actions lights up while you answer, so you can see where your answer lands.
It fills its own gaps too. If the rule needs a stage, an email template or a number of days and your sentence didn't say, it offers the real ones from your pipeline instead of leaving you an empty dropdown. Other takes any number you want.
Skip and Skip all are always there. The questions are a shortcut, never a gate, and everything stays editable by hand below.

A question card headed One quick question, labelled Conditions, Last one. It asks reject candidates scoring below what, notes that the threshold for a low score was not specified, and offers 3, 4 and 5, an Other box with Use this, and a Skip link — It says what your sentence left out, and the answers are one click.

Rules that reject a candidate ask you first

Anything that rejects, or marks someone rejected, shows Confirm before publishing and says plainly: This rule rejects candidates without a human looking first.
Two safer answers are one click away. Flag my team instead swaps the rejection for an alert to the recruiters on that job. Mark as poor fit instead sorts the candidate on the board without closing them out. Keeping the rejection is something you choose, not something you drift into.

A red-bordered card headed Confirm before publishing, labelled Actions, 1 more after this. It reads This rule rejects candidates without a human looking first, then offers Yes reject automatically in red, Flag my team instead, and Mark as poor fit instead, above a line reading Answer this one to clear the check — The only question you cannot skip.

It tells you when a rule cannot do something

Ask for something automations don't do yet, such as deleting or archiving records, and you get That isn't something a rule can do yet with the reason. Nothing is filled in below, so you can describe a different rule or build one by hand.