Company questions

Company Questions Skill

Reusable operating procedure for building and extending the company-wise LeetCode-style question bank at C:\Users\saikiran\Desktop\DSA\company questions\. This is a separate, sibling pipeline to /dsa-master (which produces the pseudocode + Mermaid topics list/ bank) — this one produces real, runnable Python code with at least two distinct solution methods per question, LeetCode-style Input/Output/Example sections, framed per-company.

If $ARGUMENTS names one or more companies, scope this run to those; otherwise ask which company(s) to generate for.

Current State

Fixed 12-company list (matches topics list/ and prompt/prompt.md): Meta, Google, Amazon, Apple, NVIDIA, Microsoft, Oracle, Anthropic, OpenAI, JPMorgan, Goldman Sachs, Jane Street.

Before starting new work, list C:\Users\saikiran\Desktop\DSA\company questions\ to confirm which companies already have a file, so you extend rather than duplicate.

When to use this skill

• User asks for company-specific interview questions with actual runnable code (not pseudocode).
• User asks to add/extend a company's file, or wants multiple solution approaches per problem.
• User asks something like "give me Google-style LeetCode questions with code."

Procedure

1. Scope the work

• Target company name(s) — file per company: C:\Users\saikiran\Desktop\DSA\company questions\<Company>.md.
• Pick a question count per file, 10-15, Easy → Medium → Hard, with an exact ordered title list of real/well-known problems genuinely associated with (or plausible for) that company's interview loop — decide this list yourself before dispatching agents; do not let a write agent invent its own list.
• Language is Python 3 for all code, across every company/file, for consistency.

2. Write pass — Sonnet 5 (one agent per file, company-question-writer subagent, run in parallel, foreground)

Dispatch one company-question-writer agent per company via the Agent tool, each with:

• The exact absolute output path.
• The company name.
• The exact ordered title list and count.

That agent's own instructions (see .claude/agents/company-question-writer.md) cover the full template and hard rules — required sections: Difficulty, Problem, Input, Output, Example, Method 1 (brute force) + Explanation + Complexity, Method 2 (algorithmically distinct, optimized) + Explanation + Complexity, optional Method 3, Company Angle. Every code block must be real, runnable, self-contained Python with its own imports and its own if __name__ == "__main__": demo whose printed output matches the stated Example output.

3. Verify pass — Haiku (one agent per file, company-question-verifier subagent, run in parallel, foreground)

Dispatch one company-question-verifier agent per file. Unlike /dsa-master's verifier, this one actually executes every Python code block (via Bash) and checks the printed output against each question's stated Example — not just structural presence. Reports PASS, QUESTION_COUNT, CODE_EXEC, ISSUES.

4. Fix pass — Sonnet 5 (only for files that failed verification)

One company-question-writer agent per failing file: pass it the exact reported issues (including which method/question failed execution or output mismatch) plus the same title list/template, have it read the current file and rewrite/fix the broken question(s) in place with the Write tool.

5. Model/process constraints (do not deviate)

• Reading/execution/verification tasks → Haiku (company-question-verifier). Writing/generation tasks → Sonnet 5 (company-question-writer).
• This is a fixed, static process — a known list of companies/files dispatched once each, not a dynamic/open-ended loop.
• Prefer the Workflow tool (static pipeline: write → verify → fix) if available; otherwise fall back to direct parallel Agent tool calls per pass.

6. Update memory after every batch

• C:\Users\saikiran\Desktop\DSA\memory\progress.md — append a row/section per new company file (name, question count, verification status incl. code-exec results).
• This skill's "Current State" section — keep the list of companies already covered up to date.