---

name: code-janitor description: Scan source code for dead code, duplicate logic, zombie functions, and semantic clones. AI-powered code cleanup recommendations. trigger: /janitor

Code Janitor

Detects dead code, duplicate logic, and zombie patterns in JavaScript/TypeScript projects.

Commands

/janitor scan [path]           -- Full scan of project (default: current dir)
/janitor scan --layer=dead     -- Dead code detection only (wraps Knip)
/janitor scan --layer=clones   -- Clone detection only (wraps jscpd)
/janitor scan --layer=zombie   -- Zombie logic patterns only
/janitor scan --layer=semantic -- Semantic clone detection (LLM-assisted)
/janitor report [path]         -- Generate markdown report from last scan

Instructions

When the user invokes /janitor scan, follow this flow:

1. Detect Project Type

• Check for package.json, tsconfig.json in the target path
• Identify JS/TS project structure

2. Layer 1: Dead Code (--layer=dead)

• Run: npx knip --reporter json 2>/dev/null
• If knip not installed, run npx knip (npx auto-installs)
• Parse JSON output for unused exports, files, dependencies
• If knip fails or is unavailable, skip this layer with a note

3. Layer 2: Clone Detection (--layer=clones)

• Run: npx jscpd --reporters json --output /tmp/janitor-clones/ [path] 2>/dev/null
• Parse JSON output for Type 1-3 code clones
• If jscpd fails or is unavailable, skip this layer with a note

4. Layer 3: Zombie Logic (--layer=zombie)

• Run: node /Users/rafiimanggalajapamel/.claude/skills/code-janitor/zombie-scanner.mjs [path]
• Detects: empty catch blocks, console.log debris, no-op functions, commented-out code, unreachable code after return, TODO/FIXME markers
• Always available (zero dependencies)

5. Layer 4: Semantic Clones (--layer=semantic)

• Run: node /Users/rafiimanggalajapamel/.claude/skills/code-janitor/function-extractor.mjs [path]
• Pipe output to: node /Users/rafiimanggalajapamel/.claude/skills/code-janitor/similarity.mjs
• Review top candidate pairs yourself (Claude) and judge if they are semantically equivalent
• This layer is optional and LLM-assisted

6. Aggregate Results

• Pipe all layer outputs as JSON to: node /Users/rafiimanggalajapamel/.claude/skills/code-janitor/reporter.mjs
• Input format: {"project":"name","scannedAt":"ISO","dead":{...},"clones":{...},"zombie":{...},"semantic":{...}}
• Output: formatted markdown report

7. Present Report

• Show the markdown report to the user
• Highlight high-severity findings first
• Offer to help fix specific findings

Default Scan

When no --layer is specified, run all layers sequentially (1-4) and aggregate.

Path Handling

• Default path: current working directory
• Skip: node_modules, .git, dist, build, coverage, .next, .nuxt

Notes

• All scanner scripts are zero-dependency (Node.js stdlib only)
• Zombie scanner and function extractor work via regex (no tree-sitter needed)
• Layer 4 (semantic) requires Claude judgment for final determination