registry / devez-vibe / 1.6.49

devez-vibe@1.6.49

Stable terminal UI for Codex and Claude Agent SDK

AI Security Review

scanned 4h ago · by lpm-firewall-ai

LPM flags this version as an AI-agent control-surface risk. npm installation writes a bundled skill into Codex and Claude user configuration directories. It also prunes files in those installed skill directories.

Static reason

High-risk behavior combination matched malicious policy.

Trigger

npm postinstall runs automatically when the package is installed.

Impact

The package changes broad user-level AI agent control surfaces without an explicit setup command.

Mechanism

Install-time copying and deletion in external AI agent skill directories.

Policy narrative

Installing the package automatically invokes install-skills.mjs. That script copies the bundled luna-loop skill into both Codex and Claude user directories and deletes destination entries that are no longer present in its source. This is an unconsented postinstall mutation of external, user-level AI agent control surfaces.

Rationale

The confirmed postinstall hook modifies both Codex and Claude configuration trees without a user-invoked setup action. This meets the policy threshold for malicious install-hook abuse even though no network exfiltration was found.

Evidence

package.jsoninstall-skills.mjsskills/luna-loop/SKILL.mdskills/luna-loop/**~/.codex/skills/luna-loop~/.claude/skills/luna-loop

Decision evidence

public snapshot

devez-vibe@1.6.49

AI Security Review

Decision evidence

Source & flagged code

Source & flagged code

Findings

Source & flagged code