/code-clone - Code Clone Detector

What this is for

Code clones - identical or near-identical code blocks - are a maintenance liability: bugs must be fixed in N places, refactoring carries hidden scope, and the codebase grows without corresponding value. This skill systematically finds all 4 clone types, validates each cluster, and produces deduplication proposals.

Detects all 4 clone types in a target directory. Produces a structured report with similarity scores, risk tiers, and LLM-validated deduplication proposals.

Usage

/code-clone              # interactive (prompts for directory)
/code-clone <dir>        # scan directory directly
/code-clone -help        # show usage

PROTECTION RULE - never ~/.claude/

Read-only skill. Guard required if write mode added later.

## What You Must Do When Invoked

Step 1

1. -help / -h -> print usage, exit 0.

Step 2

2. Confirm target directory exists.

Step 3

3. Run scripts/clone-scan.ps1 -ProjectDir <dir>.

Step 4

4. LLM reads the JSON output, validates each clone cluster:

Step 5

• Type 1-3: verify cluster, determine risk tier, propose deduplication strategy.

Step 6

• Type 4 candidates: read both files, determine if semantic clone exists,

Step 7

assign confidence level.

Step 8

5. Filter false positives (framework boilerplate, test doubles, generated files,

Step 9

trivial wrappers).

Step 10

6. Write clone-report.md to the working directory.

Clone Types

Type	Name	Detection	Similarity
1	Exact	Whitespace-normalized content hash	1.0
2	Parameterized	Token-normalized (ids/literals as placeholders)	1.0
3	Near-miss	Token Jaccard similarity within Type 2 groups	>= 0.7
4	Semantic	LLM validation of file-level fingerprint pairs	LLM-determined

Risk Tiers

Tier	Meaning
Critical	Both blocks actively maintained, diverging logic risks bugs
Medium	One block is a copy of the other, likely a quick paste
Low	Generated code, test fixtures, intentional duplication

Output

clone-report.md with:

• Executive summary (total clusters, clone density, distribution by type)
• Critical findings (sorted by risk, with locations and deduplication proposal)
• Medium findings (grouped by clone type)
• Low / informational findings (intentional duplicates with justification)
• False positives (dismissed with reason)
• Open questions (all suspected clusters, Type 4 needing manual review)