nishant-Tiwari24/cracked

What it is

cracked is a pack of 19 agent skills for tech placement and coding-interview preparation. Installed into Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI or other agent runners, it lets the assistant act as a timed DSA interviewer, an online-assessment grader, a resume reviewer, a system-design and HR interviewer, and a preparation planner. Skills are invoked explicitly, e.g. /mock-interview or /roast-my-resume, or the model picks them up when you describe what you want.

The target audience is final-year and pre-final-year CS students preparing for campus placements, new graduates applying off-campus, and engineers switching jobs. Company-specific behaviour covers 470+ companies: reported rounds, OA formats, frequently reported LeetCode questions and values are cached in a brief built by company-intel company-intel skill.

Install and first run

Pick one install path per agent; installing both plugin and manual routes duplicates every skill.

For Claude Code:

claude plugin marketplace add nishant-Tiwari24/cracked
claude plugin install cracked@cracked

Plugin installs update themselves. For Codex use codex plugin marketplace add ... and codex plugin add ...; for other agents, copy editable files with npx skills@latest add nishant-Tiwari24/cracked -a <agent> and include setup-cracked and cracked-profile when prompted install instructions. Editable files are updated by hand with npx skills@latest update.

Then run /setup-cracked once. It asks up to six short questions (status, target roles, target companies, language, DSA level, deadline), saves your profile, builds briefs for target companies, checks for Python, SQLite, gh and text-to-speech, and suggests first sessions setup-cracked. After that you can type prompts such as:

> roast my resume ~/Downloads/resume.pdf
> mock interview for Google, medium, voice mode
> /oa-sim Amazon

What each part does

• Interviews: mock-interview runs a timed DSA round with withheld constraints, a hint ladder, real code execution and a hire verdict; oa-sim creates assessment files in your editor and grades AC/WA/TLE against hidden tests from an executed reference solution; system-design-mock, hr-round, core-viva (OS, DBMS, CN, OOP, SQL executed in SQLite) and pattern-coach cover the other round types mock-interview, oa-sim.
• Profile: roast-my-resume gives an ATS score and rewrites that leave missing numbers as [N] instead of inventing them; jd-match, resume-blueprint, defend-my-project, make-it-hireable, github-glowup and referral-dm cover job-description matching, project defence and outreach roast-my-resume.
• Strategy: prep-plan, rejection-review and offer-decode build week-by-week plans, post-mortems and CTC breakdowns; all rely on the shared cracked-profile memory layer cracked-profile.

Sessions write scorecards to a local journal under ~/.cracked/, so weak spots persist across runs and plans can be rebuilt from them.

Constraints

• Some skills depend on local tools: Python and SQLite for executed tests and SQL, gh for GitHub-related work, and text-to-speech for voice mode. /setup-cracked reports what is missing.
• Skills hand off to each other; in agents without a native Skill tool, install the whole pack, not single skills, and always include cracked-profile.
• Data stays in plain Markdown in ~/.cracked/ (override with CRACKED_HOME); the README states there is no account, server or tracking, but verification is left to the user.
• Company question data comes from the community dataset leetcode-company-wise-problems and is labelled as frequently reported, not leaked.
• Quality claims (12/12 eval passes, +0.38 over the same model without skills) are the author's, run through Claude Code's plugin eval harness evals.
• Licensed MIT LICENSE.

HotRepo's take

On 2026-11-10 the repository entered HotRepo's new-project board with 15 stars gained and 4 forks over the roughly 90-minute window between 10:45 and 12:15 UTC, reported at 122 stars and 10 forks. The sample is short and single-snapshot, so it supports only that-day activity rather than a trend. What is notable from the technical side is the design of the interview skills: they execute candidate code and grade against hidden tests verified by an executed reference solution, rather than asking the model to guess correctness. That makes the eval harness claim more checkable than a prompt-only skill pack would be.