@clawos-dev/clawd@0.2.579
Standalone clawd daemon — Claude Code (and future Codex) session server over WebSocket
AI Security Review
scanned 25d ago · by lpm-firewall-aiThe package has a concrete first-party agent-extension setup path that automatically equips its own launched sessions with a privileged local MCP server. Static inspection found no npm lifecycle attack or evidence of credential export to an unrelated endpoint.
Decision evidence
public snapshot- The command-line entry point is the bundled CLI, with no install lifecycle hook declared.
- Daemon startup creates a dispatch MCP configuration containing a daemon bearer token and records it for use.
- The generated configuration runs the package's dispatch server as a local stdio MCP server.
- Claude launches receive every registered MCP configuration through the mcp-config flag.
- Channel extension installation verifies a SHA-256 digest and rejects unsafe archive paths before extracting.
- No automatic npm preinstall, install, or postinstall behavior was found.
- The observed MCP configuration is package-owned and injected into sessions launched by this daemon, not written to a global agent configuration.
Source & flagged code
18 flagged · loading sourcePackage source references child process execution.
dist/app-builder-plugin/mcp-server.cjsView on unpkg · L21117Source file is highly similar to a previously finalized malicious package; route for source-aware review.
dist/app-builder-plugin/mcp-server.cjsView on unpkgPackage contains a possible secret pattern.
dist/app-builder-plugin/mcp-server.cjsView on unpkg · L24884Findings
Affected versions and remediation
This report applies to @clawos-dev/clawd@0.2.579.
See all versions of @clawos-dev/clawd for other recorded verdicts.
- Review the evidence and your use of @clawos-dev/clawd@0.2.579 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.