cracked: agent skills for company-wise mock interviews, OAs and resume roasts
cracked is a pack of 19 agent skills that turn Claude Code, Codex, Cursor, GitHub Copilot or Gemini CLI into an interviewer, coach and career team for tech placements.
What is cracked?
cracked is a pack of 19 agent skills that turns an AI coding agent in the terminal into an interviewer, coach and career team for tech placements, and the repository is nishant-Tiwari24/cracked.
Naming a company changes how the skills behave: mock interviews, online assessments, HR rounds and resume work adapt to that company, using a cached brief built from reported rounds, OA format, published values and the resume bar, with frequently reported LeetCode questions for 470+ companies. Without a company name, the skills still work in a generic product-company style.
The pack covers practice rounds, profile work and planning in one place. It includes timed mock interviews and online assessments graded against hidden tests, system design and HR rounds, a resume roast and job-description matching, project defence, referral messages, a prep plan and offer decoding, with every session ending in a scorecard logged to a local journal.
cracked had about 330 GitHub stars as of November 18, 2026.
Who is cracked for?
cracked is for final-year and pre-final-year CS students preparing for campus placements, new grads applying off-campus, and engineers switching jobs.
It is built for readers who already work in a terminal with Claude Code, Codex, Cursor, GitHub Copilot or Gemini CLI, and it also works with other agents through editable skill files, including OpenCode, Windsurf and Amp. The skills work with any model, run code instead of guessing about it, and never invent a fact about the reader.
Setup checks for tools some skills need: Python, SQLite, gh and text-to-speech. The setup also asks about status, target roles, target companies, language, DSA level and deadline.
Is cracked free and open source?
cracked is free of charge under the MIT License.
- License: MIT, as stated in the LICENSE file and the README.
- Cost: permission is granted free of charge under the MIT License text in the repository.
- Latest release: v1.3.0, published on November 6, 2026, as the newest entry on the Releases page as of November 18, 2026.
- Skills: 19 skills listed in the plugin manifest, split into core, interview, profile and strategy groups.
- Companies: 470+ companies covered by the company-wise question data.
- Agents: Claude Code, Codex, Cursor, GitHub Copilot and Gemini CLI, plus other agents through editable files.
- Data: session scorecards and profile data live in ~/.cracked/ on the reader's own machine, in plain Markdown, with no account, no server and no tracking stated in the README.
- Form: every skill is a Markdown file, and the validator is intentionally zero-dependency.
What stands out
- Company mode: /company-intel builds a cached brief with reported rounds and OA format, the company's most frequently reported LeetCode questions across 30-day, 3-month, 6-month and all-time windows, published values and the resume bar, each with its source.
- Mock interviews that run code: /mock-interview runs a timed DSA round that withholds constraints, gives a 4-rung hint ladder instead of answers, runs code against hidden edge cases, and, in the newer CHANGELOG entries, grades with random differential testing against a brute force on 2,000 to 20,000 inputs, time-capped.
- Online assessments as real files: /oa-sim presents a full online assessment as real files in the reader's own editor, graded AC / WA / TLE against hidden tests whose expected outputs come from an executed reference solution, with the hidden tests kept outside the candidate folder in the newer CHANGELOG entries.
- A calibrated resume roast: /roast-my-resume gives a brutal roast calibrated to the reader's level and market, with a heuristic ATS-readiness score out of 100 and its breakdown in the newer CHANGELOG wording, and rewrites that never invent a metric, using [N] for missing numbers.
- A journal that remembers: every session ends with a scorecard logged to a local journal in ~/.cracked/, which /prep-plan and /rejection-review use to build plans and drills from real weak spots, while /offer-decode turns CTC into in-hand pay with step-by-step tax math in a shell, red flags called out, and a negotiation draft.
How do I install cracked?
cracked installs as a plugin for Claude Code or Codex, or as editable skill files for any other agent, and the reader then runs /setup-cracked once.
From the project's README, pick one method per agent, because installing both gives every skill twice, and a plugin updates itself while the skills.sh route copies editable files updated by hand.
claude plugin marketplace add nishant-Tiwari24/cracked
claude plugin install cracked@crackedFor Codex, the README gives these commands:
codex plugin marketplace add nishant-Tiwari24/cracked
codex plugin add cracked@crackedFor any other agent, or for editable files, the README gives this command, with manual updates:
npx skills@latest add nishant-Tiwari24/cracked -a <agent> # cursor, copilot, gemini, opencode, windsurf, amp…; omit -a to chooseWhen the installer asks which skills to take, the README says to include setup-cracked and cracked-profile, and to update with the following command:
npx skills@latest updateRunning /setup-cracked once asks up to six short questions, saves the profile, builds a prep brief for each target company, checks the machine for Python, SQLite, gh and text-to-speech, and suggests the first three sessions.
The README then shows these example prompts:
> roast my resume ~/Downloads/resume.pdf
> mock interview for Google, medium, voice mode
> /oa-sim AmazonHow it was built with Claude
The repository includes an AGENTS.md file for working on the pack, which asks contributors to read CONTRIBUTING.md before editing a skill, run node scripts/validate.mjs after any change, keep the version identical in package.json and .claude-plugin/plugin.json, add a CHANGELOG.md entry, and add no dependencies.
The eval suite is described as behavioural tests run through Claude Code's plugin eval harness, with deterministic checks plus LLM judges, and the repository ships plugin and marketplace manifests in .claude-plugin/.
Try it with your AI. Paste this into your AI assistant after installing cracked, and fill in the bracketed parts.
Help me start interview preparation with cracked, using only the official commands and
interfaces named in its README. First, confirm which agent I use: [Claude Code, Codex,
Cursor, GitHub Copilot, Gemini CLI, or another agent]. If I use Claude Code, show me
these commands first and wait for me to confirm before I run anything: claude plugin
marketplace add nishant-Tiwari24/cracked and claude plugin install cracked@cracked. Then
help me run /setup-cracked once, using my answers: status [final-year student,
pre-final-year student, new graduate, or engineer switching jobs], target roles [roles],
target companies [companies], language [C++, Java, Python, JavaScript, or Go], DSA level
[beginner, intermediate, or strong], and deadline [date]. After setup, suggest a first
session with /mock-interview for [company] at [difficulty], or /roast-my-resume for
[resume path], and explain what each skill will do before it runs. Show risky steps
first and wait for my confirmation. Do not install anything from unknown sources, do not
touch credential files, and do not delete anything.What are cracked's limits?
cracked is for practice only, and its company data, scores and pay estimates carry the limits the repository states, rather than being official or guaranteed.
- Practice only, never live cheating: if the reader is clearly in a live, proctored assessment or interview, the shared honesty rules say to decline to solve it and offer a debrief afterwards or practice on the same pattern once it is over.
- Company questions are labelled frequently reported by candidates from a community dataset, never a company's official, leaked or guaranteed question bank, and briefs carry a freshness date; the company-intel skill warns that data may be stale if the upstream date is more than about 3 months old, and falls back to an archetype when a company is not in the dataset or the reader is offline.
- Scores are judgements, not official results: the resume score is labelled a heuristic ATS-readiness score in the newer CHANGELOG entries, because no ATS publishes one, and the OA cutoff verdict is labelled a heuristic.
- Offer estimates are for planning only, not tax or legal advice, and the offer-decode skill ends by saying to confirm final numbers with HR; tax rules change every year, and the newer CHANGELOG entries update the India rules to FY 2026-27.
- The reader's data stays local in plain Markdown in ~/.cracked/, or CRACKED_HOME if set, and nothing is uploaded by cracked itself, but the privacy notes say the only network traffic is the agent's normal traffic to its model provider, plus downloads of the public company-question dataset and web lookups if the agent has a web tool.
- Voice mode in Claude Code needs a Claude.ai login and a local microphone, and does not work over SSH or in cloud sessions; if no text-to-speech command is available, the skill stays text-only.
- Integrity signals are not proof: the integrity notes say the skill cannot see the screen, browser or other apps, and must not claim to proctor, and the explain-and-modify probe runs only when two or more signals appear.
- If the reader's language toolchain is not installed, mock interviews and online assessments fall back to hand-tracing, stated as such, and agents without a Skill tool need the whole pack installed, always including cracked-profile.
Related on claudefox
- Product Engineering: 45 free agent skills for shipping a product with Claude Code and Codex
- BeautifulRust: Agent Skills for writing, reviewing, and porting idiomatic Rust
- Dashboard Design Skill: a Claude Code skill for clean, honest dashboards
- How to call Gemini 4 Hydrogen from Claude Code: gemini-api CLI, Docs MCP and skills
Problems and feedback
Report problems on the project's Issues page on GitHub, where the repository provides bug and skill-idea issue templates, and read CONTRIBUTING.md before proposing a change to a skill.
Source: cracked repository, cracked README, cracked Releases, cracked LICENSE, cracked CHANGELOG, cracked AGENTS.md, cracked plugin manifest, cracked evals, cracked-profile skill, company-intel skill, setup-cracked skill, mock-interview integrity notes, mock-interview voice mode, OA runner notes, offer-decode skill, cracked contributing guide, cracked Issues
This story first appeared in the claudefox daily briefing for November 19, 2026.