Claude 市场兼容层
Claude Code compatibility layer for DeepSeek Harness: load Claude Code skills, commands and agents — from a project's.claude directory or from Claude Code marketplaces — as native dsh skills.
이것은 DeepSeek Harness(DSH) 플러그인입니다. 이 사이트는 GitHub README, 설치 정보, 유지보수 상태, 공개 보안 시그널을 모아 보여줍니다.
업스트림에서 중국어 README를 제공하지 않아 저장소 원본 내용을 표시합니다.
dsh-claude-marketplace
Claude Code compatibility layer for DeepSeek Harness: load Claude Code skills, commands and agents — from a project's .claude directory or from Claude Code marketplaces — as native dsh skills.
This is the ecosystem bridge: install any Claude Code plugin once and its components become callable dsh skills (prefixed cc-), reusing the huge Claude Code plugin/skills ecosystem inside DeepSeek Harness.
Install
dsh plugin --profile <name> add dsh-claude-marketplace
Usage
Ask the model to use the cc_* tools, or drive them directly:
| Tool | Purpose |
|---|---|
cc_scan | Scan <project>/.claude + installed plugins; register every component as a dsh skill |
cc_marketplace_add | Register a marketplace (local path, owner/repo[#ref], or git URL) |
cc_marketplace_list | List registered marketplaces and their plugins |
cc_plugin_install | Install a plugin and register its skills/commands/agents |
cc_plugin_list | List installed plugins |
Example flow:
cc_marketplace_add("anthropics/claude-code")
cc_marketplace_list()
cc_plugin_install(plugin="commit-commands", marketplace="claude-code-plugins")
cc_scan()
Imported components are named cc-<plugin>-<component> (kebab-case, matching dsh's skill name contract), so they can never collide with native dsh skills. Original names are preserved in the label field (my-plugin:review).
What is supported (v0.1)
- Skills —
skills/<name>/SKILL.md, plugin-rootSKILL.md - Commands —
commands/*.md(flat, loaded as skills) - Agents —
agents/**/*.md(loaded as skills with a delegation header) - Marketplaces —
.claude-plugin/marketplace.jsonwith sources: relative path,github,url,git-subdir,npm,archive - Local
.claude— project-level skills/commands/agents (project trust)
Not yet (v0.2+)
- Hooks — parsed but not executed (security: hooks are arbitrary code; an opt-in policy is planned)
- MCP / LSP servers — use dsh's native MCP client instead
- Themes / monitors / workflows / output styles
allowed-tools,model,effortfrontmatter semantics (documented, ignored for now)- Auto-refresh of the skill catalog on file changes (run
cc_scanmanually)
Configuration
| Field | Default | Meaning |
|---|---|---|
cacheDir | ~/.dsh/claude-plugins | Cache root for marketplaces and installed plugins |
skillPrefix | cc | Prefix for imported dsh skill names |
Security
Installed plugins are copied into the cache (never referenced in place). Hooks are not executed in v0.1. Path traversal in marketplace relative sources is rejected. Review plugin sources before installing, exactly as you would for Claude Code itself.
License
MIT
보안 및 설치 증거
이 점수는 공개 저장소 메타데이터와 이 사이트에 등록된 설치 증거에만 기반하며, 코드 보안 감사와 다릅니다.
공개 플러그인 카탈로그에서 왔으며, 공개 GitHub 저장소로 연결됩니다.
저장소가 MIT 라이선스를 선언했습니다.
최근 180일 내 코드 업데이트가 있습니다.
재현 가능한 정확한 설치 메타데이터가 아직 등록되지 않았습니다. 저장소 설명에 따라 직접 확인하세요.
검사한 패키지 메타데이터에 설치 라이프사이클 스크립트가 선언되지 않았습니다.