Introduction

The plugin ecosystem of DeepSeek Harness (DSH) supports a high degree of customization. In scenarios that require internet access, existing official search services often require an API Key, and the model-side web_search tool is pre-masked in DSH 0.1.5+ versions, making it impossible to pass depth parameters.

The dsh-web-search-scrape plugin directly scrapes public result pages from search engines and social platforms using Node.js built-in fetch, registers a provider named web-scrape for ctx.web, and registers the web_search_deep tool to bypass the masking mechanism, enabling local retrieval without an API Key.

Core Features

The plugin provides the following capabilities:
* Dual Backend Switching: Supports local (local scraping, default) and official (official API).
* Six-Tier Depth Control: From T1 rapid tier to T6 research tier, with explicit parameter passing or automatic tier selection.
* Dual-Pool Engine: P0 search engine pool (Google, Bing, Baidu, etc.) + P1 social platform pool (Weibo, Bilibili, Reddit, etc.).
* Settings Page Card: Provides a visual configuration card in browser settings, with persistence support.
* Verification Mechanism: T4 and above support subagent triage and webpage content verification.

Backend and Billing

Search Backend

The plugin switches backends via the settings card. The default is local, completely free.

If official search is required (backend: official), allowOfficial must be enabled in settings. By default, allowOfficial is false, and in this case, even a misconfiguration will not generate any network requests or billing.

Billing Control

Billing only occurs in the following two cases, both of which have hard switches:
1. Official Search: After enabling allowOfficial, it is billed based on DeepSeek Messages API tokens.
2. Deep Subagents: T4 and above tiers dispatch LLM subagents for triage or verification, billed based on the current session model.

To avoid silently incurring fees, the plugin defaults to limiting maxAutoTier to 3 (i.e., automatic tier selection is capped at T3, pure scraping only). To restore T4-T6 subagent capabilities, the tier parameter must be passed explicitly.

backend: local          # 使用本地抓取
allowOfficial: false    # 禁止官方搜索，防止误触
maxAutoTier: 3          # 限制自动档位，防止静默产生 LLM 费用

Tiers and Tools

Six-Tier Retrieval Depth

The plugin provides six retrieval tiers from T1 to T6, with a default result cap of 130 items.

Tier Name Duration Engine Pool Social Pool Verification Method
T1 Rapid ~4s 1 0 None
T2 Quick ~10s 2 0 Heuristic Rating
T3 Standard ~20s All 1 Heuristic Rating
T4 Enhanced ~60s All All Subagent Triage
T5 Deep ~150s All All Subagent + ≤6 page verification
T6 Research ~280s All All Subagent + ≤15 page multi-round reading

Tool Masking Handling

In DSH 0.1.5+ versions, the model-side web_search tool has been moved to the agent preset plane, which shadows the plugin’s same-name registration, preventing the tier parameter from being passed.

The plugin solves this issue by registering a tool named web_search_deep. This tool is not subject to preset shadowing, uses the same parameters as web_search, and can be used under any preset.

Installation and Configuration

Prerequisites

  • DSH version must be >= 0.1.5-rc.2.
  • Ensure host compatibility during installation, especially for the ctx.settings.installSection and createSnapshotStore API versions.

Directory Structure

The plugin repository provides ESM source code directly, with no build required:

.
├── package.json
├── cordis.patch.yml
└── lib
    ├── index.js      # 宿主侧：Provider 与工具注册
    └── client.js     # 浏览器侧：设置页卡片

Usage Examples

  1. Select backend as local in the settings page card.
  2. Set allowOfficial: false in the configuration file to avoid billing.
  3. Call web_search_deep in code or prompts, passing the tier parameter (e.g., tier: 5).
  4. Use maxAutoTier: 3 to limit the upper bound of automatic tier selection.

Applicable Scenarios and Notes

  • Scenarios: Developers who need to control result count (default 130) and retrieval depth; long-running sessions that want to avoid API Key management or paid services.
  • Notes: The plugin runs with the permissions of the current DSH process. Before installation, it is recommended to review the source code and license (MIT). T4-T6 tiers involve subagent calls and will incur corresponding model token fees, so configuration should be done carefully.

Summary

dsh-web-search-scrape provides a local scraping solution that requires no API Key and allows control over retrieval depth and result count. It also resolves the tool masking issue in newer DSH versions through a tool renaming mechanism, making it suitable for DSH users who require fine-grained internet access capabilities.