Dsa system design optimizer

DSA and System Design Optimizer

Use this skill to find practical engineering optimizations in a real codebase. Optimize for product impact, maintainability, and measured evidence, not interview cleverness.

Default mode is audit first. Do not edit code until the user chooses a candidate or explicitly asks for implementation.

Serve experienced engineers with concise evidence and trade-offs first. When the user seems newer or asks for learning help, add a short plain-English explanation of the relevant data structure, algorithm, or system pattern after the recommendation.

Operating Rules

• Ground every recommendation in product behavior, expected scale, and code evidence.
• Prefer simple, boring improvements before complex algorithms or infrastructure.
• Estimate both current and proposed complexity. If complexity is uncertain, say what evidence is missing.
• Treat memory, readability, rollout risk, and tests as first-class constraints.
• If the host supports subagents, delegate independent read-only audits in parallel. If not, inspect locally.
• For destructive, irreversible, or high-blast-radius changes, produce a design/spec first.
• Do not shame simple code. A clear O(n^2) solution may be correct for bounded cold paths.
• Stay language- and vendor-neutral by default. Name specific languages, frameworks, databases, clouds, or vendors only when discovered in the codebase or requested by the user.

First Pass

1. Read project context: README*, CONTEXT.md, docs/, ADRs, architecture diagrams, package manifests, and deployment files.
2. Identify product shape: user workflows, hot paths, data volume, traffic, read/write ratio, latency targets, uptime needs, and team/ops maturity.
3. If context is missing, ask only for the minimum missing scale facts:
   • Product in one sentence
   • Current and target scale
   • The workflow that must stay fast or reliable

Mode Selection

Use DSA audit when the request mentions Big O, slow functions, memory, loops, sorting, search, graph/tree logic, deduplication, ranking, batching, or hot endpoint internals.

Use system design audit when the request mentions architecture, scaling, databases, caching, queues, reliability, infra, services, RAG, LLM serving, agents, multi-user growth, latency, or availability.

Use both when a slow user workflow likely has architectural and code-level causes.

DSA Audit Workflow

1. Explore read-only:
   • Hot paths: endpoints, jobs, render paths, CLI commands, query handlers.
   • Complexity smells: nested loops, repeated scans, repeated sorts, recursion without memoization, queue operations with shifting arrays, N+1 queries, repeated DB/API calls.
   • Data structure smells: lists used for membership, ad hoc deduplication, rebuilt indexes, missing maps/sets, repeated range sums, brute-force graph traversals.
2. Classify candidates using references/dsa-patterns.md and references/dsa-complexity.md.
3. Present findings using references/audit-templates.md.
4. Before implementation, run the grilling loop in references/grilling.md.
5. Implement only after selection. Add or update tests first when practical, then measure or benchmark with realistic input sizes.

System Design Audit Workflow

1. Explore read-only:
   • Entry points, service/module boundaries, data stores, queues, caches, infra, deployment topology, observability, background jobs, third-party dependencies.
   • Scale smells: sync chains on user requests, database bottlenecks, missing indexes, poor shard keys, unbounded scans, no pooling, no cache invalidation plan, shared mutable state, single points of failure, no rate limits, weak observability.
   • Database internals smells: weak query plans, lock/deadlock pressure, bad isolation assumptions, replication lag, write/read amplification, poor partitioning, unsafe failover, missing restore drills.
   • AI-system smells: RAG without evals, no ACL filtering before retrieval, slow retrieval, missing reranking, unbounded agent loops, non-idempotent tools, weak prompt-injection defenses.
2. Classify candidates using references/system-design-patterns.md.
3. Present roadmap using references/audit-templates.md.
4. Run the grilling loop in references/grilling.md before proposing migrations or infra changes.
5. For code-level improvements, implement and verify. For architecture changes, write a spec/ADR with rollout, rollback, metrics, and trade-offs.

Parallel Audit Prompts

When subagents are available, use short, read-only prompts like these:

• Hot path auditor: "Find likely hot paths and complexity smells. Report file/function, evidence, current complexity guess, and risk. Do not edit."
• Data structure auditor: "Find places where arrays/lists, maps, sets, heaps, queues, indexes, or graph structures are missing or misused. Do not edit."
• Data layer auditor: "Find query/data-access inefficiencies, N+1 patterns, missing indexes, bad query plans, lock pressure, partitioning/sharding risks, replication/consistency risks, connection pooling issues, and cache candidates. Do not edit."
• Reliability auditor: "Find sync dependency chains, retries/timeouts/circuit breaker gaps, rate-limit gaps, single points of failure, and observability gaps. Do not edit."
• AI systems auditor: "If the product uses LLMs/RAG/agents, inspect retrieval, evals, tool safety, memory, budgets, and latency risks. Do not edit."

Merge reports, remove duplicates, and rank by user impact.

Evidence Standard

Every recommendation should include:

• File/component evidence
• Current behavior and estimated cost
• Proposed data structure, algorithm, or system pattern
• Expected complexity or reliability improvement
• Trade-offs and failure modes
• Verification plan: tests, benchmark, load test, metric, or rollout signal

If evidence is weak, label it as a hypothesis and explain how to validate it.

Reference Guide

• DSA pattern selection: references/dsa-patterns.md
• Complexity and hidden costs: references/dsa-complexity.md
• System design patterns: references/system-design-patterns.md
• Grilling questions: references/grilling.md
• Report templates: references/audit-templates.md