Introduction

In the development practice of DeepSeek Harness (DSH), agents often need to invoke a specific skill or expert but do not know what is available locally or in the marketplace. Traditional approaches require manually browsing catalogs or writing complex query logic. dsh-skill-matcher is a Cordis plugin that addresses this pain point. It scans local environments and online open-source catalogs to provide DSH with a unified skill and expert matching layer.

Plugin Introduction

This plugin is maintained by axel286137079-dot and follows the MIT license. It is essentially a Skill & Expert Matcher with functionality consistent with the SKILL.md version, but packaged as a DSH plugin. Its core purpose is to provide automated recommendation and matching services when users ask, “which skill/expert should I use?”

Core Features

The plugin provides three main tool functions and an automatic injection mechanism:

  1. skill_matcher_scan
    Scans local skills and experts (supports multiple environments) while also fetching open-source catalog data online. It merges these data to build an index and caches it locally.

  2. skill_matcher_match(query, topN?, includeExperts?)
    Matches based on the user requirement (query) and returns Top-N recommendations. The results include match scores, hit keywords, and specific installation methods.

  3. skill_matcher_list(scope?)
    Used to browse lists of skills and experts from local installations, the marketplace, or open-source catalogs.

Automatic Injection and Engine

The plugin injects a guiding text into the system prompt. When the user’s query involves skill selection, the system preferentially calls skill_matcher_match.

The core engine is located at lib/engine.js, which is a JS port of the Python script bin/sync_index.py. Its processing flow includes:
* Environment detection
* Parsing frontmatter from SKILL.md or plugin.json
* Fetching remote open-source data
* Priority-based merging
* Keyword recall

It includes 13 built-in open-source seed records and can be used in offline environments.

Installation and Activation

The one-click installer in the project root is recommended:

bash ../install.sh --dsh-only

Or add manually (an absolute path is required):

dsh plugin --profile web add file:<本项目绝对路径>/plugin

After installation, restart the DSH Web service to take effect:

launchctl kickstart -k gui/$(id -u)/com.deepseek.harness.web

Typical Usage

  • Browse lists: Call skill_matcher_list to browse local or marketplace resources.
  • Match skills: Call skill_matcher_match with specific requirements to obtain recommendations and installation commands.
  • Cache mechanism: Index data is written to ~/.dsh/dsh-skill-matcher/cache.json, with a cache lifetime of 24 hours.

Notes

  • Security and compliance: The plugin only performs local scans and fetches catalog data in JSON format; it does not execute any remote code.
  • Offline availability: Thanks to the 13 built-in open-source seed records, the plugin can run in disconnected environments.
  • Neutral ranking: Recommendation order is determined solely by requirement match scores, without other factors.
  • Installation troubleshooting: If dsh plugin add fails due to a pnpm supply-chain policy in the profile (such as an expired minimumReleaseAge), use the install.sh script in the root directory instead. It writes directly to the profile’s package.json and creates an automatic backup.

Conclusion

dsh-skill-matcher provides DSH with standardized skill discovery and matching capabilities. Its purely local nature and lack of remote code execution make it suitable for environments that prioritize security and offline operation. For more details, see: GitHub.