S.C.R.U.B. FAQs

Question 1

What is S.C.R.U.B. and how does it reduce cloud LLM costs?

Accepted Answer

S.C.R.U.B. (Source Code Review, Uplift, and Baselining) is a tool that automates code quality tasks. It drastically cuts cloud LLM token usage by using 16 specialized deterministic tools and local LLMs for high-volume tasks like linting, typing, and docstrings, engaging expensive cloud models only for high-level planning and review.

Question 2

Can S.C.R.U.B. be integrated into existing development workflows?

Accepted Answer

Yes, S.C.R.U.B. functions as an MCP (Model Context Protocol) server. This allows seamless integration with cloud LLMs (like Claude), VS Code, and any other MCP client, enabling agentic workflows where cloud models orchestrate S.C.R.U.B. tools rather than generating boilerplate.

Question 3

How does S.C.R.U.B. ensure local LLMs provide high-quality results?

Accepted Answer

It uses a 'teacher-student optimization' approach. One-time cloud model input (the 'teacher') compiles highly effective prompts for the local 'student' LLM. These optimized prompts are then used at runtime, ensuring consistent and high-quality output without recurring cloud costs.

Question 4

What types of code quality tasks does S.C.R.U.B. automate?

Accepted Answer

S.C.R.U.B. automates a wide range of tasks including linting (Ruff), generating docstrings and type annotations, adding comments, analyzing code complexity, suggesting refactoring, generating tests, security scanning (Bandit), and creating SBOMs. It has 16 specialized tools for comprehensive code hygiene and security.

Question 5

How does S.C.R.U.B. leverage local LLMs for performance?

Accepted Answer

It employs a deterministic-first approach with a three-tier pre-filtering system, ensuring local LLMs are only engaged when necessary. It also utilizes batched DSPy calls to a local LLM, optimizing performance and reducing round trips for tasks like code generation.

S.C.R.U.B.

S.C.R.U.B.

主な機能

ユースケース

主な機能

ユースケース