Athang-codes

ShrinkCode — Development skill for Claude Code

Development community

A Claude Skill that compresses bloated codebases into fewer lines without changing behavior: finds duplicate blocks, ranks the worst functions, and proves every change with its own measurement tools.

How to install ShrinkCode

This entry records only its repository, not the path inside it, so there is no exact command to give. Open Athang-codes/ShrinkCode and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What ShrinkCode does

A Claude Skill that compresses bloated codebases into fewer lines without changing behavior: finds duplicate blocks, ranks the worst functions, and proves every change with its own measurement tools.

Alternatives in Development

  • Semble — Fast and Accurate Code Search for Agents 6.1k ★
  • Breach Check — HIBP k-anonymity check on a password wordlist 4.5k ★
  • Code Review Graph — Local knowledge graph for Claude Code 3.7k ★

README

shrinkcode

![shrinkcode in a terminal: it finds the duplicate twins, ranks the worst functions, and proves itself with its own selftest — real output of the bundled tools](assets/demo.svg)

**Install:** **[download `shrinkcode.zip`](https://github.com/Athang-codes/ShrinkCode/releases/latest) → upload it in claude.ai, or unzip it into `~/.claude/skills/` for Claude Code.** One file, no build step, nothing to configure.

A Claude Skill that compresses bloated codebases into fewer lines — **without changing behavior.** Same inputs, same outputs, same edge cases, same error handling, same ordering guarantees. Just leaner code, proven with real tooling instead of vibes.

Turns "take this 10k-line project and make it 5k lines but it has to work exactly the same" from a hope into a repeatable, auditable, git-checkpointed pipeline.

Why this is different from "just ask an LLM to refactor"

  • Tiered risk model. Every change is classified Mechanical / Structural / Semantic before it's made, and the verification bar scales with the tier — a provably-dead import gets a quick test-suite run, an async-ordering change gets an explicit sign-off requirement and a rejection-path test.
  • Real metrics, not impressions. Bundled scripts measure line count (code/blank/comment separated), duplication (exact + near-duplicate detection), and cyclomatic complexity — before and after — so "smaller" and "simpler" are both provable, and cramming logic onto fewer lines can't be disguised as a win.
  • Git-checkpointed batches. Every verified change lands as its own commit with a message recording what changed and how it was verified. Nothing is ever one giant unreviewable diff; any regression is instantly bisectable and any single batch is independently revertible.
  • A written plan before execution, for anything beyond a trivial job — the equivalent of a PR description written before the PR exists, so you approve the approach before the AI starts rewrit