Auto‑Score Flags

Programmatic signals to grade risk and block merges when necessary.

Signals

Lifecycle scripts present (+3)
preinstall/install/postinstall/prepare detected in package.json or script files.
child_process in lifecycle files (+4)
exec, spawn, execSync during install is a major warning sign.
Network calls in lifecycle files (http/https/net) (+4)
Downloading data/binaries or beaconing during install increases risk.
Access to sensitive paths/env (.ssh, .aws, .npmrc, process.env.*TOKEN|KEY) (+5)
Likely credential theft or environment exfiltration.
Obfuscation patterns (_0x[a-f0-9]+, large encoded arrays, Buffer.from(base64)) (+3)
Conceals behavior; often paired with payload delivery.
Remote binary hosts not on trusted GitHub Releases/CDN (+4)
Unknown hosts are common in malicious native add-ons.
New maintainer recently or repo removed (+3)
Ownership changes can precede compromise.
Name looks like a typo/brandjack of top packages (+5)
Typosquats target common imports and CI scripts.
Writes to HOME or user directories during install (+3)
Modifies ~/.npmrc, ~/.ssh, or caches unexpectedly.
Contacts Discord/Telegram/Pastebin during install (+4)
Common exfil endpoints for malware campaigns.
Dynamic eval/Function with obfuscated strings (+3)
Hides runtime behavior; often decodes payloads.
chmod +x and executing downloaded binaries (+5)
Strong indicator of a payload dropper.
Install‑time telemetry of CI metadata (+2)
Grayware behavior that can leak sensitive org information.

Compute a simple risk score; block merges when score exceeds your threshold.
Log which flags fired; link to triage results to guide reviewers.
Whitelist known safe build-time packages with native compiles from trusted sources.
Require checksum/signature verification when remote binaries are allowed.
Alert when maintainer changes or repository becomes inactive.