omp-plugin-duplicate-detector

Real-time duplicate code detection for oh-my-pi (omp), built on jscpd.

The plugin indexes your repository in the background and watches the agent as it works. When newly written or edited code duplicates existing logic anywhere in the workspace, the agent is warned immediately, before the copy-paste hardens into a maintenance problem. Duplicate detection can be toggled on or off per project using /duplicates on|off.

Indexing and mutation checks run entirely in a background worker with a persistent on-disk cache, so sessions start instantly and the agent loop is never blocked, even in large repositories.

Installation

Install directly via the omp CLI:

omp plugin install omp-plugin-duplicate-detector

On startup you'll see a short readiness note in the transcript, e.g. Duplicate detector: Ready (1,420 Git files indexed, cached).

Usage

Real-time warnings

Enabled by default. Whenever the agent writes or edits a source file, the change is checked against the workspace index. If it duplicates existing code, a warning naming both locations is surfaced. How it's delivered is controlled by reminderMode:

• steer (default) — shows a warning card in the terminal and steers the agent to reconsider before its next step.
• in-band — appends a system reminder to the tool result instead.
• none — disables real-time warnings; the index stays warm for on-demand scans.

Previously reported duplicates are remembered per session, so iterative edits to the same file don't repeat the same warning.

/duplicates slash command

Toggle duplicate detection on or off on a per-project basis. The preference is stored persistently in ~/.cache/omp/duplicate-detector/projects.json.

/duplicates on                           # enable duplicate detector for this project
/duplicates off                          # disable duplicate detector for this project
/duplicates status                       # show current status for this project

Configuration

Plugin settings (via the extension's settings in your omp config):

Project configuration

Standard jscpd configuration is discovered automatically — .jscpd.json and its .jscpd.rc* / .config/ variants, or a "jscpd" key in package.json:

{
  "minLines": 6,
  "minTokens": 50,
  "ignore": ["vendor/**", "build/**"],
  "ignoreTests": true,
  "customTestPatterns": ["**/custom_fixtures/**"],
  "excludeTestPatterns": ["**/src/services/test-utils-in-prod.ts"],
  "formatsExts": {
    "yaml": ["yml", "yaml"],
    "markdown": ["md", "mdx"]
  }
}

What is filtered by default

To prevent false alarms, noise from test fixtures, and unnecessary tokenization, the detector filters out non-production code, data files, and build artifacts by default:

• Test files and directories (ignoreTests: true): Test files, test suites, test doubles (mocks, stubs, fakes), fixtures, snapshots, and test runners across common languages and frameworks (e.g. Jest, Vitest, Pytest, Go, JUnit, Cargo, etc.) are skipped automatically. Use customTestPatterns to ignore additional paths, excludeTestPatterns to keep specific paths in production scope, or ignoreTests: false to disable test filtering entirely.
• Non-code and data formats: Documentation (.md, .txt), data and configuration files (.json, .yaml, .toml, .csv, .ini), server configs (nginx, apacheconf), diffs, logs, and standalone markup data (.svg, .xml) are excluded. To scan duplicates within specific data or doc formats, opt them in via formatsExts.
• Generated code and lockfiles: Files containing standard auto-generation header markers (such as @generated or DO NOT EDIT), generated file naming conventions (*.generated.*, *.designer.cs), lockfiles (*.lock, *-lock.json), and minified assets (*.min.js, *.map) are ignored.
• Safety and resource limits: Large files (>100 KiB) and files outside the Git repository are skipped to maintain minimal memory usage and fast background indexing.

Scope and performance

• Only Git-tracked files are indexed (git ls-files); outside a Git repository, baseline scanning is skipped.
• Hard caps on file size, file count, and total indexed bytes prevent runaway memory use in large monorepos.
• Tokenized files are cached on disk per workspace with size-bounded eviction, so repeat sessions skip re-tokenization entirely.

Development

Requires Bun and Git.

bun install        # dependencies
bun test           # test suite
bun run typecheck  # TypeScript
bun run check      # lint (Biome)
bun run build:worker  # rebuild dist/detector-worker.js after worker-side changes

License

MIT — Oh My Pi Contributors