@cotal-ai/connector-jcode@0.135.12
Cotal connector for Jcode: a host-mode mesh peer driven through Jcode's Harness API bridge.
AI Security Review
scanned 2d ago · by lpm-firewall-aiThe credential mirroring appears to support the Jcode integration, and I found no evidence that the credentials are exfiltrated to an unrelated service. However, it automatically exposes credentials from several other AI clients to the Jcode agent without a per-credential opt-in, which is a meaningful capability risk.
Decision evidence
public snapshot- The package describes itself as a Cotal host-mode mesh peer driven through Jcode’s Harness API bridge.
- The credential inventory includes local Claude, Codex, Gemini, Cursor, and GitHub Copilot credential stores.
- When the host starts, it calls mirrorJcodeCredentials for the per-agent Jcode home.
- That function copies or links the discovered credential files into the Jcode home.
- The Jcode instance is launched with inheritLogins:false, but only after the connector has mirrored credentials.
- The manifest’s scripts list contains typecheck, test, build, and bundle; it declares no install lifecycle hook.
- The credential handling is part of the package’s Jcode host integration, and the files are prepared for a local Jcode instance.
- The inspected code does not establish that these credentials are sent to Cotal or another unrelated host.
- The Jcode launcher’s separate credential-inheritance option is disabled.
Source & flagged code
7 flaggedA single source file combines environment access, network access, and code or shell execution; review context before blocking.
dist/host.jsView on unpkg · L111Findings
Affected versions and remediation
This report applies to @cotal-ai/connector-jcode@0.135.12.
See all versions of @cotal-ai/connector-jcode for other recorded verdicts.
- Review the evidence and your use of @cotal-ai/connector-jcode@0.135.12 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.