In AI-assisted security, simply providing knowledge is often insufficient to complete complex tasks. Models may possess extensive knowledge, but lack workflow guidance and control, which can lead to low efficiency or deviation from the goal. DeepSeek Harness (DSH) aims to address this problem. dsh-redteam-model is a plugin implemented based on DSH Web, providing security researchers with a security research workflow covering penetration testing, red team assessments, code auditing, and other domains.

Core Features

This plugin includes ten security mode presets and seventeen runtime plugins:

  1. Ten security mode presets:

    • redteam: General security researcher
    • pentest: Penetration testing
    • code-audit: Code auditing
    • binary-analysis: Binary analysis
    • attack-defense: Attack-defense assessment
    • av-evasion: AV evasion
    • incident-response: Incident response and forensics
    • cloud-security: Cloud security attack and defense
    • ctf-solver: CTF solving
    • asset-mapping: Asset mapping

  2. Seventeen runtime plugins:

    • dsh-stage-gate: Gate structure validation
    • dsh-route-boost: Per-round governance envelope
    • dsh-sec-enforce: Deterministic tool interception
    • dsh-refusal-guard: Refusal prevention
    • dsh-trace-vault: Process trace vault
    • dsh-auto-advance: Auto-advance mechanism
    • dsh-product-subagents: Subagents
    • dsh-mcp-studio: MCP tool console
    • dsh-redteam-results: Results showcase
    • dsh-hunter: Asset hunting
    • dsh-campaign-memory: Campaign memory
    • dsh-mode-group: Mode selector
    • dsh-session-pulse: Session status panel
    • dsh-attack-atlas: Attack surface atlas
    • dsh-scanner-tools: Scanner tool wrapper
    • dsh-semgrep-audit: Static analysis
    • dsh-webshell-mgr: Webshell management

  3. Other features:

    • Workflow-based design, addressing the lack of guidance in AI.
    • Supports offline deployment.
    • Integrates scanning tools (nuclei, httpx, ffuf, nmap, etc.).
    • Supports the MCP tool console.

Installation and Activation

Prerequisites

  1. DeepSeek Harness (DSH) must be installed first.
  2. Node.js >= 22 is required.

Installation Methods

Run the following command in a terminal:

dsh plugin --profile web add github:SeaOf0/dsh-redteam-model

Option 2: Install via the Web Interface

  1. Open the dsh web settings page.
  2. Find Redteam Manager.
  3. On that page, deploy the ten security modes or seventeen runtime plugins with one click.

Typical Usage

  1. View the gate schema: In any session, call gates_list to view the gate structure definitions for specialized modes.
  2. Use scanning tools: In modes such as pentest, attack-defense, cloud-security, ctf-solver, and asset-mapping, you can directly use scanning tools such as nuclei_scan.
  3. View results: In the redteam-results plugin, view the task ledger and results page.

Applicable Scenarios and Notes

  • Applicable scenarios: Only for security testing with written authorization, CTF competitions, bug bounty programs, and security research scenarios.
  • Note: It must not be used for illegal activities. Workflow design is the core feature, aimed at addressing the lack of workflow guidance in AI.

Summary

dsh-redteam-model provides DeepSeek Harness with structured security research capabilities through preset workflows and runtime plugins. It covers the entire process from asset mapping to incident forensics and is suitable for users who need to conduct systematic security research under an authorized environment.

Project address: GitHub