InterviewMentor Claude repo

A software interview is usually four different tests wearing one company badge: one round asks you to code, one asks you to design a large system, one asks you to debug a failure, and one asks you to explain how you handled real people and messy projects. InterviewMentor packages those tests into reusable interviewer files that can run inside Claude Code instead of a human mock interviewer. (github.com) The repo is public on GitHub under PrepLabsAI, carries an MIT license, and was crawled this week with 53 commits, 3 forks, and 13 stars. Its README describes it as a collection of specialized interview “skills” for Claude Code and other agent-style assistants. (github.com) The basic trick is simple: instead of asking a chatbot for random practice questions, you load a role with rules. The repo says each skill is a markdown instruction set that turns the assistant into a specific interviewer with a domain, difficulty, and style. (github.com) Claude Code already has a plugin marketplace system for installing local or shared extensions, and Anthropic’s docs say plugins can add skills, agents, commands, and other capabilities. InterviewMentor uses that system by telling users to add the local `agents` folder as a marketplace inside Claude Code. (code.claude.com) (github.com) The repo is broad enough to mimic an onsite loop instead of a single LeetCode drill. Its listed tracks include entry-level software engineering, mid-level backend and frontend, data engineering, machine learning, debugging, behavior, and systems design. (github.com) What makes it feel more like an interviewer than a flashcard deck is the pushback. In the included URL shortener example, the agent rejects “fast” as an answer, asks for a read-to-write ratio, forces the candidate to calculate queries per second, and then challenges a weak shortcut based on the MD5 hash function. (github.com) The debugging example uses the same pattern in a different setting. It starts with a payment service slowing from 100 milliseconds to 30 seconds, then pushes the candidate away from “check the logs” and toward triage, thread-pool exhaustion, timeouts, and a circuit breaker. (github.com) Under the hood, the template is built around a timed interview flow instead of one-shot grading. The default structure is 5 to 10 minutes of warm-up, 15 to 20 minutes of core concepts, 20 to 30 minutes of problem solving, and a 5-minute wrap-up with a scorecard, strengths, improvement areas, and study resources. (github.com) That timing matters because real interviews are often lost on pacing, not knowledge. The template says the interviewer should adjust difficulty after the warm-up, stay easy if the candidate struggles early, and add harder constraints if the candidate answers quickly. (github.com) The repo also breaks the “software interview” label into narrower jobs, which is closer to how hiring actually works. One system reliability interviewer focuses on exponential backoff, circuit breakers, monitoring metrics, tracing IDs, and disaster recovery targets, while a deployment rollback interviewer focuses on failed releases, rollback choices, database migration compatibility, and feature flags. (github.com 1) (github.com 2) This is still prompt engineering, not a hiring oracle. But as a practice tool it solves a real problem: most candidates can find 500 interview questions online, and far fewer can rehearse a full 45-minute conversation where someone keeps pressing until the vague answer turns into a number, a tradeoff, or a failure mode. (github.com 1) (github.com 2) The project is also new enough that it is still expanding in public. GitHub shows version 1.0.0 as the initial release, version 1.0.1 updating links and installation instructions, and version 1.0.2 on March 18, 2026 adding more agents for more topics. (github.com)