thalamus@1.4.1
Mobile and Web client for Claude Code
AI Security Review
scanned 2mo ago · by lpm-firewall-aiThe package contains a real remote-command control path, but it is activated by the user-facing authenticated client rather than an install hook or concealed persistence. The credential upload is explicit, so the demonstrated behavior is dangerous capability rather than proven malware.
Decision evidence
public snapshot- When the CLI runs, it downloads enabled MCP server definitions from its cloud service, including command, arguments, and environment variables.
- Those downloaded definitions are passed as Claude MCP configuration, allowing the cloud service to select commands that the locally launched Claude process can run.
- The explicit connect command sends Anthropic OAuth tokens to the package service.
- An opt-in environment flag can upload unencrypted runtime log content to a user-selected server URL.
- The install hook only extracts bundled platform tools into the package directory.
- Token registration is reachable through an explicit connect command after authentication.
- The OAuth token read for quota polling is sent to Anthropic's usage endpoint, not shown being sent to the Thalamus service.
- No self-dependency or automatic lifecycle hook modifies Claude configuration files.
Source & flagged code
10 flaggedPackage defines install-time lifecycle scripts.
package.jsonView on unpkgSource file is highly similar to a previously finalized malicious package; route for source-aware review.
dist/lib.cjsView on unpkgPackage source references dynamic require/import behavior.
dist/lib.cjsView on unpkg · L2Findings
Affected versions and remediation
This report applies to thalamus@1.4.1.
See all versions of thalamus for other recorded verdicts.
- Review the evidence and your use of thalamus@1.4.1 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.