Introduction

In the DeepSeek Harness (DSH) development workflow, Agent skills are usually stored as SKILL.md files scattered across system, project, and skill pools. Developers often struggle with skills that are hard to remember, difficult to locate quickly, or that are installed but never used (“zombie skills”). dsh-skill-cockpit is designed to solve these issues by providing a centralized panel for viewing, managing, and testing all Agent skills.

Plugin Positioning

This is a Web client plugin for DeepSeek Harness (DSH). It centralizes scattered skill management in the Settings → Skill Cockpit panel.
* Maintainer: Itailang2333
* Core value: Provides Chinese annotations, usage statistics, and a local test shot feature to help developers organize their skill library.

Core Features

  • Full view: Displays skills from system-level (~/.dsh/skills), project-level (<project>/.dsh/skills, etc.), and the skill pool (~/.dsh/skill-pool).
  • Search and annotations: Supports searching by name, description, or annotation. Each skill can have a one-line Chinese annotation added, helping solve the problem of “installed but forgot what it does.”
  • Management actions: Supports enabling/disabling, deleting, and importing skills. Import supports batch folder processing and conflict preview (dry-run).
  • Usage statistics: Records the number of skill calls and the last used time to help identify unused “zombie skills.”
  • Test shot: Provides a test panel where, after entering a sentence, local lexical scoring returns the Top-5 skills most likely to be triggered (not a real model run).

Installation and Enablement

Install it using the official plugin command, then refresh the browser to load the new panel.

dsh plugin --profile web add <repo path or git+https URL>

After installation, restart DSH, perform a hard refresh in the browser (Ctrl+Shift+R), and then go to Settings → Skill Cockpit.

Typical Usage

  • Remember skills: Add a one-line Chinese annotation to each skill in the list, for example, “Translation skill: handles Chinese-English translation.”
  • Test triggering: Enter a sentence in the test panel, review the Top-5 results returned by the local scoring, and confirm whether the skill triggering logic matches expectations.
  • Organize the skill pool: When importing skills from the skill pool into the enabled list, use the conflict preview to check whether existing configurations would be overwritten.

Applicable Scenarios and Notes

  • Applicable scenarios: Developers who need to centrally manage a large number of Agent skills, want to quickly locate a skill’s purpose, clean up unused skills, or verify skill triggering logic.
  • Data storage: All data files are stored locally (~/.dsh/skill-cockpit.*) and do not involve network requests.
  • Permission limitations: Deletion operations are limited to user-level directories and include path escaping protection.
  • Scoring mechanism: The test shot scoring is based on local vocabulary weights (an approximation) and is not a real model inference result. It is only used to assist in checking trigger-word overlap.

Conclusion

dsh-skill-cockpit provides DSH users with an efficient tool for managing Agent skills through a visualized panel and a localized scoring mechanism. Full documentation or source code can be found on GitHub.