This picture is from one run by SkillLift on an invented brief. It is not from the project.

Open source

Interview Coach

Paste what you said in a practice interview. It scores each answer, says what the question was testing, and shows where the answer is thin.

Interview Coach · Noam Segal · MIT

Install

Clone it

git clone https://github.com/noamseg/interview-coach-skill.git
cd interview-coach-skill

Then switch it on

mv SKILL.md CLAUDE.md

Commands are copied word for word from the project's README. Read them before you run them.

We ran it once

The brief: Invented: a job seeker called Wren Talvik, six years at a made-up parcel company (Brindlemoor Parcel), practising for an Operations Lead job at a made-up freight firm (Kettlewick Freight). Three typed-up practice answers, all vague and none with a number, and a request for an analysis in one go with no questions. Fictional; the label is on the page and on the poster. The page carries the skill's text unchanged and only sets it in headings, lists and tables.

It scored each answer on five points (the three answers averaged 1.7, 2.0, 2.3, 1.7 and 1.0), named the habits behind them (saying "we" all the way through, avoiding conflict, no result stated), and said what each question was really testing. It rewrote the weakest answer as a table of original against stronger version with empty brackets for every real fact, and gave a worksheet of questions to fill in over 72 hours. Nothing in the reply states a number or event the person did not supply. The second run, asked to rewrite the returns answer, also left every figure as a bracket.

What we thought: On the thing we most wanted to know, it behaved: given vague answers with no numbers, it said what was missing and left brackets instead of making facts up, on two models. It is more careful than most interview tools there. But that is one command. Its own examples invent numbers, three of its commands that rewrite your words have no rule against it, and we did not run them, so use the analysis freely and check every rewrite of a resume or pitch line by line. It is long, English only, and starts by pushing you to set up a state file.

Open the full output

One run by an AI agent on 2026-11-02, at commit 634a8dd. No person has judged it, and your result will differ.

Why it was picked

It fills the interview-practice gap. It scores each answer on five points, says what the question was really testing, and when it rewrites an answer it leaves brackets where only you know the facts.

Why it exists

Practice with a friend rarely tells you why an answer fell flat. This names the habit, such as saying "we" all the way through, and what to fix first.

What you get

A written analysis: five scores per answer, the pattern behind them, a rewrite of the weakest answer with brackets for your real facts, and a short practice plan.

How it works

Clone it, switch it on, and delete its settings file.
Open the folder in Claude Code and say kickoff, or paste a transcript and ask for an analysis.
Fill in the brackets with true facts before you use any rewrite.

What it needs

Claude Code, or another agent that can read files. English only.

In short

Pick this when you can describe your answers but not why they did not land.

Worth knowing

English only, and aimed at the US (pay talk, state laws). In our run on a practice transcript with no numbers in it, the analysis and the rewrite invented no figure: unknowns stayed as empty brackets, on two runs by two different models. But its own worked examples add numbers and events the person never gave (Example 4 adds a veto and five figures, with a warning; Example 13 adds several with none), and its resume, pitch, LinkedIn and outreach commands have no rule against that. We did not run those. It names real people and employers from a podcast and newsletter with no links, and quotes pay figures for named companies as fact. It keeps your notes in a coaching_state.md file in its folder and can search the web. The download includes a settings file, .claude/settings.json, that pre-approves reading, editing, writing, web fetch and web search. Delete that file and approve each action yourself. Renaming SKILL.md to CLAUDE.md is the project's own install step.

Used in these workflows

Apply for a job step 5