Stop supply-chain
attacks before they run

Command Line Options

Commands

Deep Scan

The --scan flag activates all six intelligence modules for a comprehensive security audit of your dependency tree.

# During install:
safenpm install --scan

# Standalone (no install, analysis only):
safenpm scan

# JSON output for CI:
safenpm scan --json

What gets checked

Typosquat detection checks every package name against 80+ popular npm packages using Levenshtein distance and 5 attack pattern detectors including char-swap, substitution (0→o, rn→m), scope confusion, and edit distance analysis.

Behavioral diffing compares install scripts against a cached baseline from your last install. If a package upgrade introduces new curl calls, DNS lookups, or other suspicious patterns, you'll see the exact new warnings and risk delta.

Maintainer change detection queries the npm registry for the publisher of each version. If the account publishing the current version differs from who published the previous one, that's flagged as a potential account takeover.

Lockfile integrity audit parses package-lock.json and checks for git dependencies, custom registries, missing integrity hashes, weak hash algorithms, and stale lockfiles.

Reputation scoring evaluates every package based on maintainer count, license, repository presence, dependency weight, install scripts, and version maturity, producing a 0-100 score and a project-level health grade.

Community threat intel runs automatically on every install — not just with --scan. Each time you run safenpm install, packages are checked against the live decentralized intelligence network, surfacing flagged packages with report counts and attack categories before any scripts execute.

Allowlisting

Some packages (like bcrypt, sharp, or node-gyp native modules) legitimately need install scripts. Allowlist them to skip sandboxing.

CLI flag

safenpm i --allow bcrypt,sharp,@img/sharp-darwin-arm64

Project config (.safenpmrc)

# .safenpmrc in project root
bcrypt
sharp
@img/*
@mycompany/*

Global config

# ~/.safenpmrc
bcrypt
node-gyp

Allowlisted packages still run, but outside the sandbox. The allowlist is loaded from three sources (CLI flags, project .safenpmrc, home ~/.safenpmrc) and merged. Scope wildcards (@scope/*) match all packages in that scope.

CI/CD Integration

GitHub Actions

- name: Install with safenpm
  run: |
    npm install -g @abgunaydin/safenpm
    safenpm install --json --scan --no-report | tee safenpm-report.json

- name: Check for blocks
  run: |
    BLOCKED=$(cat safenpm-report.json | jq '.summary.blocked')
    TYPOS=$(cat safenpm-report.json | jq '.summary.typosquats')
    if [ "$BLOCKED" -gt 0 ] || [ "$TYPOS" -gt 0 ]; then
      echo "::error::safenpm found $BLOCKED blocked scripts and $TYPOS typosquat suspects"
      exit 1
    fi

JSON output structure

{
  "version": "1.0.0",
  "backend": "sandbox-exec (macOS)",
  "packages": [...],
  "typosquats": [...],
  "lockfileAudit": { "score": 85, "issues": [...] },
  "reputationSummary": { "overallScore": 72, ... },
  "summary": {
    "total": 5, "blocked": 1, "clean": 3,
    "allowed": 1, "warnings": 8,
    "typosquats": 0, "maintainerChanges": 0,
    "lockfileIssues": 2, "reputationScore": 72
  }
}

Doctor as CI gate

# Fail the build if project health is D or F:
safenpm doctor --json
# (exits non-zero on score < 60)

Standalone scan (no install)

# Just scan existing node_modules:
safenpm scan --json

Audit log

Every run is logged to ~/.safenpm/audit.log in JSONL format with auto-rotation at 5MB. Query it with safenpm audit or safenpm audit --json.

Static Analysis Rules

safenpm's analyzer runs 19 rules across 3 severity tiers before scripts are sandboxed. Each rule has a weight; the total is capped at 100.

High severity (weight: 30)

curl/wget downloads, netcat, require('https'), fetch(), require('dns'), process.env access, ~/.ssh access, sensitive dotfile reads (.aws, .npmrc, .docker, .kube), eval(), base64 encoding/decoding, pipe-to-shell execution.

Medium severity (weight: 15)

/etc/passwd reads, $HOME access, child_process spawning, hex/unicode obfuscation, net/dgram socket creation.

Low severity (weight: 5)

node-gyp native builds, prebuild-install, node-pre-gyp. These are common in legitimate packages.

Risk levels

Scores map to levels: critical (60+), suspicious (30-59), low (1-29), clean (0).

Doctor — Project Health Report

safenpm doctor runs every analysis module and produces a letter grade from A+ to F, with weighted scores across 6 categories.

# Run a health check:
safenpm doctor

# JSON output for CI gates:
safenpm doctor --json

What gets graded

Lockfile (20% weight) — checks for missing lockfile, git/file dependencies, missing integrity hashes, weak algorithms, and staleness.

Install Scripts (25% weight) — static analysis of all postinstall hooks. Critical-risk scripts heavily penalize the score.

Typosquats (20% weight) — scans every package name. A single high-confidence typosquat drops this section to 10/100.

Reputation (15% weight) — aggregate reputation score from all packages. Low-reputation packages (missing license, no repo, single maintainer) drag this down.

Behavior (10% weight) — compares install scripts against cached baselines. Significant risk increases are flagged.

Project Hygiene (10% weight) — checks for .safenpmrc, engines field, wildcard version ranges, and package.json health.

Grade scale

A+ (97+), A (93+), A- (90+), B+ (87+), B (83+), B- (80+), C+ (77+), C (73+), C- (70+), D (60+), F (below 60). Doctor exits non-zero on D or F for CI gating.

Actionable fixes

Every finding includes a fix: suggestion — specific commands to run, config changes to make, or packages to replace. If auto-fixable issues are found, doctor prompts you to run safenpm fix.

Auto-Fix

safenpm fix scans for fixable issues and applies corrections automatically.

# Preview what would be fixed:
safenpm fix --dry-run

# Apply all fixes:
safenpm fix

# JSON output:
safenpm fix --json

What gets fixed

Typosquat replacement — high and medium confidence typosquats are automatically removed and replaced with the legitimate package. For example, axois gets uninstalled and axios gets installed in its place.

Malicious package removal — packages that were blocked during safenpm install (network or filesystem violations) can be removed from node_modules and package.json in one step.

How it works

Fix uses npm uninstall to cleanly remove packages, then npm install --ignore-scripts for replacements. If npm uninstall fails (corrupted state), it falls back to manual cleanup: deleting the directory from node_modules and stripping the entry from all dependency fields in package.json.

Package Diff Viewer

safenpm diff shows exactly what changed in your dependencies since the last time you took a snapshot — like git diff for your node_modules.

# Save current state as baseline:
safenpm diff --snapshot

# After updating packages, see what changed:
npm update
safenpm diff

# JSON output:
safenpm diff --json

What gets compared

Install script changes — line-level diffs of postinstall/preinstall hooks with red/green coloring. Shows when scripts are added, removed, or modified between versions.

Dependency changes — new dependencies added, old ones removed, and version range changes within each package.

File changes — new files added or removed from the package directory. Catches when a package ships unexpected new binaries or scripts.

Workflow

Run safenpm diff --snapshot after a clean install to establish your baseline. Then after any npm update or version bump, run safenpm diff to review changes before deploying. Integrate into CI by checking the JSON output for unexpected script modifications.