devez-vibe@1.10.4
Stable terminal UI for Codex and Claude Agent SDK
AI Security Review
scanned 8d ago · by lpm-firewall-aiSource confirms automatic installation and pruning of package-owned skills in user-level agent directories, which warrants a warning under the agent extension lifecycle policy. The bridge’s separate permission RPC does not establish an automatic settings mutation or a concrete attack.
Decision evidence
public snapshot- The npm postinstall hook runs the skill installer and a Python dependency preparation script.
- The installer targets user-level Codex and Claude skill directories, using their standard home paths unless environment overrides are set.
- On postinstall, the installer copies two bundled skills and prunes destination entries missing from the package’s copies.
- The search skill identifies itself as a bundled port; this supports a package-owned extension setup finding.
- The bridge’s settings helper resolves user, project, or local Claude settings paths; its permission update is dispatched through a separate permissions RPC.
- The reviewed postinstall skill installer does not write Claude settings; the bridge permission update has a separate RPC trigger.
- The installer copies package-bundled skills, and the inspected skill source describes a package-aligned web access capability.
Source & flagged code
7 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@1.10.4.
See all versions of devez-vibe for other recorded verdicts.
- Review the evidence and your use of devez-vibe@1.10.4 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.