Introduction¶

The core philosophy of DeepSeek Harness (DSH) is “everything is a plugin.” When building agents, developers often face the challenge of selecting the right plugins from ecosystems such as GitHub or npm. Manually browsing github.com/topics/dsh-plugin, comparing ratings, and writing installation commands is inefficient and error-prone.

The DeepSeek Harness Plugin Advisor solves this problem. It is a workflow plugin maintained by haveanote06, designed to automate requirement analysis, ecosystem search, and plugin distribution.

Plugin Overview¶

haveanote06/dsh-plugin-advisor is a plugin advisor tool. Its core workflow is: automatically analyze requirements → check locally installed plugins → search the community ecosystem → score candidates by multi-signal quality and recommend the best plugin → install with one click after user confirmation.

License: MIT

Core capabilities:
* Automatically analyze requirements
* Check locally installed plugins
* Search the community ecosystem
* Multi-signal quality scoring
* One-click installation

Installation and Activation¶

Install it using the official command. Both GitHub format and npm package names are supported as sources.

# Install from GitHub
dsh plugin --profile web add github:haveanote06/dsh-plugin-advisor

# Install from npm (after publication)
dsh plugin --profile web add dsh-plugin-advisor

After installation is complete, restart dsh web and enter normal conversation mode to start using it.

Usage¶

The plugin is triggered automatically when the Agent detects that a requirement exceeds the capabilities of the current tools. It can also be invoked by asking directly in a conversation.

Automatic trigger flow:
1. plugin_inventory: The Agent first checks whether locally installed plugins can cover the requirement.
2. plugin_search: If local plugins are insufficient, the Agent searches the community ecosystem and returns candidate plugins sorted by quality score.
3. The Agent presents the candidates and risk notes. After user confirmation, it runs plugin_install to install with one click.

Direct inquiry example:

“I want to receive a WeChat notification when a task is completed. Do you recommend any plugins?”

Quality Scoring Mechanism¶

The plugin calculates scores based on objective hard signals, while semantic matching is handled by the Agent. Each candidate plugin includes a score breakdown.

Scoring dimensions and weights:

Data sources: GitHub dsh-plugin topic search, the awesome-dsh-plugin curated list, and npm download volumes.

Configuration¶

The plugin supports several configuration items located in ~/.dsh/profiles/web/cordis.patch.yml. All of the following configuration items are optional:

- id: dsh-plugin-advisor
  config:
    catalogTtlHours: 6          # catalog cache TTL
    githubTokenEnv: GITHUB_TOKEN  # raise GitHub API rate limit (60→5000/h)
    profile: web                # target profile for installation
    weights:                    # scoring weight overrides
      curated: 35
      stars: 20
      downloads: 15
      recency: 15
      structure: 15

Security and Precautions¶

• Never install silently: Before installation, candidates and risks must be displayed, and user confirmation must be obtained.
• Command format restrictions: Installation commands are limited to npm package names or github:owner/repo; all other formats are rejected.
• Data privacy: Catalog data is fetched read-only, and no local information is uploaded.
• Cache policy: The catalog cache defaults to 6 hours.

Ecosystem Context¶

The catalog data used by this plugin comes from the independent site skillhub.cn, which has no official affiliation with DeepSeek or High-Flyer.

Additional Resources¶

• GitHub repository
• Community catalog