@marecgents/dsh-hub@0.0.2-rc.12
DeepSeek Harness desktop hub — Tauri 2.x shell + dsh web plugin layer. This npm package ships the plugin half (client UI + HTTP routes + cordis.patch.yml); the desktop shell ships as the NSIS installer (see build/).
AI Security Review
scanned 2mo ago · by lpm-firewall-aiThis is a risky DSH desktop integration with unprompted global package installation and package-owned profile setup, so it warrants a warning. The source supports no concrete malicious behavior requiring a publish block.
Decision evidence
public snapshot- The postinstall script probes four registries and can globally install dsh and pnpm without a user command.
- The desktop sidecar registers this package in its own DSH web profile and creates a junction to it.
- The runtime provides authenticated local routes that create interactive shell sessions and accept terminal input.
- The profile setup is for this package's named DSH integration, not a broad foreign agent configuration.
- The plugin disables itself unless the desktop shell launch marker is set.
- PTY routes require a per-process token and reject non-loopback hosts and foreign origins.
- No source evidence shows credential harvesting, data exfiltration, remote payload execution, or hidden persistence.
Source & flagged code
8 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 unpkgPackage source references child process execution.
bin/dsh-web-sidecar.mjsView on unpkg · L27Findings
Affected versions and remediation
This report applies to @marecgents/dsh-hub@0.0.2-rc.12.
See all versions of @marecgents/dsh-hub for other recorded verdicts.
- Review the evidence and your use of @marecgents/dsh-hub@0.0.2-rc.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.