Allowlists

Allowlists

vexes ships with built-in allowlists for packages that legitimately trigger certain signals. Allowlisted packages are downweighted, not suppressed -- a compromised version of an allowlisted package still triggers if new dangerous patterns appear.

How allowlisting works

When a signal fires for an allowlisted package, it gets a knownGood: true evidence flag. The composite scoring engine applies a 0.2x weight multiplier to known-good signals, reducing their contribution to the risk score by 80%.

This means:

• esbuild's legitimate postinstall is flagged at LOW severity instead of HIGH
• If esbuild's postinstall suddenly starts accessing process.env.AWS_SECRET_ACCESS_KEY, that new signal is NOT downweighted and fires at full severity

Known postinstall packages

These packages have install scripts for legitimate reasons (downloading platform-specific binaries, installing git hooks, etc.):

Build tools

esbuild, @esbuild/*, swc, @swc/core, lightningcss, @parcel/watcher, turbo, @vercel/turbo, vite, node-sass, sass

Native modules

sharp, @img/sharp-*, @img/sharp-libvips-*, canvas, better-sqlite3, sqlite3, bcrypt, argon2, fsevents, keytar

Build systems

node-gyp, node-pre-gyp, @mapbox/node-pre-gyp, prebuild-install, grpc, @grpc/grpc-js, protobufjs, protobuf

Browsers / testing

puppeteer, playwright, electron, electron-builder, cypress

ORM / database

prisma, @prisma/client, @prisma/engines

Dev tools

lefthook, husky, simple-git-hooks, patch-package, core-js

Popular package databases

Used for typosquat detection. A package name within Levenshtein distance 1-2 of a popular package is flagged.

npm (~150 popular packages)

lodash, chalk, react, axios, express, debug, tslib, commander, moment, uuid, webpack, typescript, eslint, prettier, jest, next, vue, tailwindcss, prisma, zod, pino, winston, and many more.

PyPI (~100 popular packages)

requests, numpy, pandas, flask, django, scipy, matplotlib, pillow, pyyaml, cryptography, pydantic, fastapi, sqlalchemy, boto3, tensorflow, torch, pytest, black, ruff, openai, anthropic, and many more.

Extending allowlists

Currently, allowlists are built into the source code (src/core/allowlists.js). Future versions will support extending them via .vexesrc.json:

{
  "allowlists": {
    "postinstall": ["my-internal-build-tool"]
  }
}

To modify the built-in allowlists, edit src/core/allowlists.js directly.