diet-code (npm)

Put your AI coding agent on a diet.

Deterministic dead-code analyzer, cleanup tool, and agent benchmark for TypeScript/JavaScript. This npm package ships a self-contained Rust binary — no Rust toolchain, no native build step, no runtime dependencies.

Install

npm install -g @mohammadpalla/diet-code

or run without installing:

npx @mohammadpalla/diet-code analyze .

The postinstall script downloads the prebuilt binary for your platform (linux/macOS/Windows × x64/arm64) from GitHub Releases and verifies its SHA-256 checksum. No binary leaves your machine; no telemetry.

Usage

diet-code analyze .              # dead-code evidence + summary
diet-code analyze . --json       # stable JSON report
diet-code explain <finding>      # evidence for one finding
diet-code clean --dry-run        # preview the deterministic patch plan
diet-code clean                  # apply it on a diet-code/<timestamp> branch
diet-code benchmark --tasks tasks.json --reps 3

Full docs, methodology, and benchmark results: https://github.com/Mohammad-Palla/diet-code#readme

Prefer building from source? Clone the repo and run cargo install --path crates/diet-code-cli. Not published to crates.io.

Supported platforms

| OS | arch | artifact | |---|---|---| | Linux | x64, arm64 | static musl binary | | macOS | x64 (Intel), arm64 (Apple Silicon) | | | Windows | x64 | .exe |

Anything else prints a clear error with the from-source fallback. You can also bypass the download (DIET_CODE_SKIP_DOWNLOAD=1), point at a mirror (DIET_CODE_DOWNLOAD_BASE_URL=…), or override the binary entirely (DIET_CODE_BINARY_PATH=/path/to/diet-code).

License

MIT — see LICENSE.