@viraatdas/rudder@3.9.2
A TUI for managing coding agents: run Claude Code, Codex and opencode side by side, each in its own isolated jj workspace.
AI Security Review
scanned 23h ago · by lpm-firewall-aiThe package silently installs a persistent cloud-management skill into global agent configuration during normal CLI startup, which leaves a meaningful agent extension risk. The inspected helper acts only when later invoked, and credential-related code did not establish unrelated credential forwarding; a warning is appropriate.
Decision evidence
public snapshot- Normal CLI startup installs or updates the bundled skill in global Claude and Codex skill directories without a confirmation step.
- The bundled Google Cloud helper can enable APIs for a specified project when invoked.
- Credential inspection resolves to identity comparison and quota requests to the corresponding providers; it does not show forwarding to an unrelated recipient.
- The flagged runtime operations handle text extraction, secret redaction, subprocess logging, WebSocket headers, and clipboard or paste data.
- The cloud-changing helper requires a later agent interaction and project/API arguments; the inspected source does not show it changing cloud settings during installation or CLI startup.
- The Jujutsu postinstall hook is separate from the agent skill installation behavior.
Source & flagged code
24 flagged · loading sourcePackage defines install-time lifecycle scripts.
package.jsonView on unpkgInstall-time source downloads a native archive from a fixed external host without transport verification, extracts it, and installs an executable payload.
package.jsonView on unpkg · L1Findings
Affected versions and remediation
This report applies to @viraatdas/rudder@3.9.2.
See all versions of @viraatdas/rudder for other recorded verdicts.
- Review the evidence and your use of @viraatdas/rudder@3.9.2 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.