Bob Rowsse Walakira

← Work

Solo · Language server + core library · npm

DryGuard

Flags duplicate functions before they land — fingerprinting new code against the workspace for agents and humans alike.

Editor (LSP)AI agents (MCP)CI / pre-commit@dryguard/corefingerprintBlock dupe

Problem

Agentic coding tools often regenerate logic that already exists in a codebase instead of finding and reusing it. Humans do the same under deadline pressure. By the time a conventional lint pass catches the duplication, the copy is already in the tree — and agents that never check for duplicates never see the problem at all.

What I did

I built DryGuard around one similarity engine reused across every surface, so adding a new IDE stays a packaging problem rather than a rewrite.

  • @dryguard/core fingerprints functions — TypeScript and JavaScript via ts-morph, Python and Go via tree-sitter — and scores structural similarity, with an optional semantic tier for look-alikes that share intent but not shape.
  • @dryguard/lsp-server exposes diagnostics and code actions to any LSP-capable editor; a thin VS Code / Cursor client launches that server and adds an AI-assisted “merge this duplicate” quick fix.
  • @dryguard/mcp-server lets agents call check_similarity before they write a line, with an auto-refreshing workspace index.
  • @dryguard/cli runs the same checks headlessly for CI and pre-commit, including baselines so an existing codebase can adopt DryGuard without drowning in pre-existing duplicates.

Result

55 npm downloads in the last month across @dryguard/core and @dryguard/lsp-server.

The packages are on npm under MIT. The same engine powers the editor, agent, and CI paths described above.

What I'd do next

First-class JetBrains and Neovim clients on top of the existing LSP server, and tighter MCP tools so agents check similarity before they write.

Let's talk.

About agentic systems, full-stack product work or developer tooling. Email is the fastest way to reach me.