@dryguard/core

Structural duplicate-code detection: AST fingerprints, structural and semantic similarity, config, baselines, and the incremental cache.

This package is the engine. It does not start a server or print a report. Call it from @dryguard/mcp-server, @dryguard/lsp-server, or @dryguard/cli, or import WorkspaceIndex if you are building another client.

What a match means

A candidate is a function or method with at least three statements, or with a branch or loop. DryGuard scores it against functions already indexed in the workspace.

• Structural score (always on): same shape after names are normalized. A renamed copy of calculateTax still scores high. The default line is 0.85.
• Semantic score (off unless you enable semanticTier): overlapping identifiers and call names, so a for loop and a .reduce() that do the same job can match. It is noisier. Default line 0.7.

Anything marked with the ignore comment (dryguard-ignore by default) is left out of the index. A baseline file lists duplicates the team has already accepted; those pairs are not reported again.

TypeScript and JavaScript are parsed with ts-morph. Python and Go use the tree-sitter grammars shipped in this package's grammars/ directory. Python and Go fingerprints treat identifiers more coarsely than the TypeScript path.

Configuration

Every client reads .dryguardrc.json from the workspace root. npx @dryguard/cli init writes a starter. The fields are threshold, semanticTier, include, exclude, ignoreComment, cachePath, and baselinePath.

Full usage, including what each client shows when a duplicate is found: How to use DryGuard.