devez-vibe@1.9.47
Stable terminal UI for Codex and Claude Agent SDK
AI Security Review
scanned 13d ago · by lpm-firewall-aiStatic source inspection found automatic installation of bundled skills into user-level agent directories, a guarded first-party agent extension setup risk. The other inspected findings did not establish concrete malicious behavior, so this warrants a warning rather than a block.
Decision evidence
public snapshot- The postinstall hook runs the skill installer automatically during package installation.
- The installer targets the user's Codex and Claude skill directories and copies bundled skills there.
- The installer copies and prunes the two named skill directories, changing agent-discoverable content without an install-time consent check.
- The bridge can update Claude permission settings when its permission-update operation is invoked; the inspected code does not show this operation being called during installation.
- The benchmark helper runs a specified local engine and records its results; it is not invoked by the package lifecycle shown.
- The CLI launcher starts the packaged or pointer-selected executable only when the user invokes the CLI. No concrete malicious payload behavior was established in the inspected source.
Source & flagged code
8 flagged · loading sourcePackage defines install-time lifecycle scripts.
package.jsonView on unpkgInstall-time lifecycle script is not statically allowlisted and needs review.
package.jsonView on unpkgSource creates an unconsented AI-agent control surface through install-time mutation or a default unauthenticated remote skill channel.
Findings
Affected versions and remediation
This report applies to devez-vibe@1.9.47.
See all versions of devez-vibe for other recorded verdicts.
- Review the evidence and your use of devez-vibe@1.9.47 before allowing it. Restrict the permissions described in this report.
- Choose an independently verified alternative or release. This report does not establish that other versions are safe.
Evidence last updated: .
Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.