registry / kbglow / 0.8.0

kbglow@0.8.0

Your Mac's keyboard backlight blinks while your AI agent (e.g. Claude Code) is waiting for approval

AI Security Review

scanned 1d ago · by lpm-firewall-ai

LPM flags this version as an AI-agent control-surface risk. Install-time setup mutates detected foreign AI-agent hook/plugin configurations to run kbglow commands. No exfiltration endpoint was found.

Static reason

High-risk behavior combination matched malicious policy.

Trigger

npm postinstall when supported agent configuration directories exist

Impact

Future agent events execute the package-owned binary through configured shell hooks.

Mechanism

automatic cross-vendor AI-agent hook and plugin injection

Policy narrative

Installing the package runs setup without an explicit setup command. The setup detects several unrelated AI CLI configuration directories and writes or refreshes hooks/plugins that execute the bundled binary during agent events. This is an unconsented postinstall mutation of foreign, cross-vendor AI-agent control surfaces.

Rationale

The observable purpose is a keyboard notification utility, but its npm postinstall automatically injects executable hooks across third-party AI-agent configurations. That meets the firewall block boundary despite no observed network or credential theft. Product guard normalized a concrete AI-agent control hijack publish_block to the blockable dangerous-capability shape.

Evidence

package.jsonscripts/setup.jsbin/kbglow~/.claude/settings.json~/.gemini/settings.json~/.qwen/settings.json~/.config/opencode/plugins/kbglow.js

kbglow@0.8.0

AI Security Review

Decision evidence

Source & flagged code

Source & flagged code

Findings

Source & flagged code