Preface

In the plugin ecosystem of DeepSeek Harness (DSH), everything is a plugin. When developers introduce third-party plugins, they need to verify their security, such as whether they leak Tokens, establish malicious network connections, or contain obfuscated code. The dsh-security-guard plugin provides a static scanning solution to help detect the security status of installed plugins in the DSH environment.

Plugin Overview

This is a DSH plugin maintained by weisofns. It provides a 28-rule static scanner, risk scoring, whitelist/blacklist policies, a local Web dashboard, and risk pop-up functionality inside DSH.

Core Features

  • Static Scanner: Built-in 28 rules covering scenarios such as Token theft, data exfiltration, dynamic execution, obfuscation, network behavior, supply chain risks, persistence, and privilege escalation.
  • Startup Self-Check: Automatically scans installed DSH plugin packages under the profile after installation.
  • DSH Tool Registration: Registers the security_scan_plugins, security_scan_plugin, and security_guard_report tools for the model.
  • Alerting Mechanism: Supports console output, event bus, optional Webhook, level filtering, and rate limiting.
  • Local Web Dashboard: Provides a visual panel and supports a JSON API.
  • Remote Rule Updates: Supports remoteSources remote rule JSON, with optional scheduled automatic updates.

Detection Rules

The plugin includes 28 built-in detection rules, mainly categorized as follows:

  • Token Related (TOK): Hardcoded Tokens, Token exfiltration, reading DSH credentials, and environment variable credentials.
  • Data Exfiltration (EXF): Suspicious outbound domains, DNS-ICMP tunneling, and clipboard read-and-exfiltrate behavior.
  • Dynamic Execution (EXE): Dynamic code execution, subprocess execution, and remote code loading.
  • Code Obfuscation (OBF): High-entropy strings, execution after decoding, and minimal JSFuck obfuscation.
  • Network Monitoring (NET): Unauthorized listening, WebSocket exfiltration, and proxy configuration tampering.
  • File System (FS): Out-of-scope file writes, deletion of system files, and full-disk file searches.
  • Process Operations (PRO): Process injection and privilege escalation.
  • Supply Chain Security (SUP): Suspicious install scripts, dependency confusion, and Git repository poisoning.
  • Persistence (PER): Startup item implantation and Shell configuration tampering.
  • Sensitive Services (PRIV): Requests to sensitive DSH services and interception of core services.

Installation

Install using the official plugin loading method. Please completely quit DSH Host first, then run:

dsh plugin --profile web add ./dsh-security-guard-2.1.0.tgz

After restarting DSH, the plugin starts working automatically.

Configuration

The plugin is configured through cordis.patch.yml. Key configuration items include:

  • enabled: Whether to enable the plugin.
  • scanOnStart: Whether to scan on startup.
  • autoUpdateRules: Whether to automatically update rules.
  • enableWebPanel / webPanelPath: Whether to enable the Web dashboard and its path.
  • remoteSources: A list of remote rule JSON URLs.
  • whitelist / blacklist: Whitelist and blacklist policies.
  • minLevel: Minimum alert level.
  • webhookUrl: Webhook URL.

Usage

  1. Interactive Scanning: In the DSH conversation, enter “Scan my plugin security” or “Check which plugins have risks”.
  2. Model Invocation: The model automatically calls the registered tools to return a risk report.
  3. Web Dashboard: Access http://127.0.0.1:<DSH port>/security-guard in a browser to open the visual panel.

Compatibility

  • Node.js version >= 20.
  • Zero third-party runtime dependencies.
  • Depends on the official DeepSeek Harness plugin loading method (dsh.bundle.patch + cordis.patch.yml).
  • @deepseek-ai/dsh-tools is an optional dependency.

Notes

The plugin runs with the current DSH process permissions. It is recommended to review the source code and license before installation.