Risky-Tool Feed Entry Detected 2026-10-02
Detections feed entry — detected by codelake scanners. This package was flagged by our own detection pipeline. The rules that fired and package identifiers are below. Full analysis, timeline and IOCs live on hand-curated advisories at /advisories; this page is the machine-generated receipt for the live detections feed.
Advisory · CLR-2026-25042
@agentgates/[email protected]
Attack-adjacent by design — legitimate author, but capability can be abused.
Detection details
@agentgates/[email protected] published to npm, flagged by codelake scanners on 2026-10-02 08:38 UTC.
Detector rules that fired on this package:
- Install lifecycle hook (postinstall) executes a bundled script that itself contains malicious-behavior findings: scripts/postinstall.mjs
behavior.scripts.install-hook-runs-flagged-script - Source writes to an AI-agent config/instruction file (CLAUDE.md / .claude / .cursor / MCP config / copilot-instructions / agent rules) — agent-control-surface tampering
behavior.static.agent-config-tamper - Source spawns a child process with a non-literal command
behavior.static.child-spawn-dynamic - Source loads a module from a non-literal name — opaque dependency
behavior.static.dynamic-require - Source contains a long base64 string literal — possible embedded payload
behavior.static.long-base64
This is an auto-generated entry in the codelake detections feed. Hand-curated deep case studies live on /advisories.