Dsa master

DSA Master Skill

Reusable operating procedure for building and extending the DSA interview question bank at C:\Users\saikiran\Desktop\DSA\topics list\. Full topic/question specification lives in C:\Users\saikiran\Desktop\DSA\prompt\prompt.md — read that first for the exact per-file question lists and the required per-question template.

If $ARGUMENTS names one or more topics, scope this run to those; otherwise ask what to generate/extend, or (if the user asked for a gap analysis) propose topics yourself before proceeding.

Current State (keep this section updated as the bank grows)

21 files, 320 questions, grouped as:

• 10 core DSA files (136 Q) — Arrays, LinkedLists, StacksAndQueues, TreesAndGraphs, DynamicProgramming, GreedyAlgorithms, SearchingAndSorting, Hashing, RecursionAndBacktracking, AdvancedTopics
• 6 senior/quant gap-fill files (96 Q) — SystemDesign, ConcurrencyAndMultithreading, LowLevelDesignOOP, BitManipulation, MathProbabilityAndRandomizedAlgorithms, AdvancedGraphsStringsAndPatterns
• 5 pointers/fundamentals files (88 Q) — Pointers, TwoPointers, ProgrammingBasicsPart1/2/3

Before starting new work, list C:\Users\saikiran\Desktop\DSA\topics list\ to confirm this inventory is still accurate (files may have been added since this skill was last updated).

When to use this skill

• User asks to add a new DSA/CS topic to the bank, or more questions to an existing one.
• User asks to regenerate or fix a file in the bank.
• User asks a gap-analysis question ("what's missing for role X?") — after answering, offer to generate the missing files using this same procedure.

Procedure

1. Scope the work

• Define the new/target file name(s): <Topic>.md, saved into C:\Users\saikiran\Desktop\DSA\topics list\.
• Pick a question count per file in the 12–20 range (never fewer than 10), with an exact ordered title list — write these titles out explicitly before dispatching agents; do not let an agent invent its own list, or counts/consistency drift across files.
• Order titles beginner → intermediate → advanced.
• For companies to reference throughout (Frequently Tested Patterns, Company Angle lines), always use: Meta, Google, Amazon, Apple, NVIDIA, Microsoft, Oracle, Anthropic, OpenAI, JPMorgan, Goldman Sachs, Jane Street.

2. Write pass — Sonnet 5 (one agent per file, run in parallel, foreground)

Each write-agent prompt must specify:

• The exact absolute output path (C:\Users\saikiran\Desktop\DSA\topics list\<Topic>.md).
• The exact numbered title list, in order — non-negotiable, verbatim.
• The required file structure: # <Topic> — DSA Interview Questions, then ## Frequently Tested Patterns Across Companies (5–8 bullets), then one ## Question N: [Title] block per question with: Difficulty, Problem, Pseudocode (fenced ```pseudo), Explanation (bullets + Complexity), Visual Flow (fenced ```mermaid flowchart, question-specific, 6–14 nodes, never generic/reused), Company Angle.
• Rules: pseudocode is never real runnable code; explanations are bullets, not prose, and never restate the problem; every diagram must be distinct and algorithm-specific; difficulty must progress across the file.
• For system-design / low-level-design topics, adapt "Pseudocode" to the core algorithm/data-structure of the design and "Complexity" to Big-O of that core piece plus a scalability/extensibility note.
• Tell the agent to reply with only a one-line confirmation (WROTE <filename> with <N> questions) — never the full file content — to keep it out of the main conversation's context.

3. Verify pass — Haiku (one agent per file, run in parallel, foreground)

Each verify-agent prompt must specify:

• The exact absolute path to read.
• The exact expected question count and topic title.
• Checks: file non-empty; correct top-level heading and Frequently Tested Patterns section present; exactly N ## Question headings numbered sequentially; every question has pseudocode + explanation + complexity + mermaid + company angle; no empty/placeholder diagrams; no truncation at end of file.
• Reply format: PASS: true|false, QUESTION_COUNT: <n>, ISSUES: <comma-separated or "none">.

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

One agent per failing file: pass it the exact reported issues and the same title list/template, have it read the current file and rewrite what's broken in place with the Write tool.

5. Model/process constraints (do not deviate)

• Reading/verification tasks → Haiku. Writing/thinking/generation tasks → Sonnet 5.
• This is a fixed, static process — a known list of files dispatched once each, not a dynamic/open-ended loop (no loop-until-budget, no unbounded agent counts).
• Prefer the Workflow tool (static pipeline: write → verify → fix) if its safety classifier is available; if the Workflow tool errors as unavailable, fall back to direct parallel Agent tool calls running the same three passes manually — this has worked reliably before.

6. Update memory after every batch

• C:\Users\saikiran\Desktop\DSA\memory\progress.md — append a table row/section per new file (name, count, verification status) and update the running grand total.
• The Claude Code memory system (project memory in the session's memory directory) — update the file inventory, question-count totals, and a short why/how-to-extend note so a future session can resume without re-deriving structure.
• Update this skill's "Current State" section and prompt.md's topic tables so the spec never drifts from what's actually on disk.