dsh-skills-anywhere

Your skills, anywhere. Install an Agent Skill once, use it in every agent: a live skill provider for DeepSeek Harness (dsh) and an MCP server for Claude Code, Cursor, Codex and friends.

English | 中文

Agent Skills are portable by design: a folder with a SKILL.md. Every agent still looks only in its own folder, so a skill you install for Claude Code is invisible to Codex, Cursor and DeepSeek Harness, and the ones you wrote for them are invisible back. dsh-skills-anywhere reads all of those folders where they live and serves them everywhere: as a live skill provider inside dsh, and as an MCP server for Claude Code, Cursor, Codex and any other MCP client.

Inside dsh, it registers one extra provider on the built-in ctx.skills registry, so the model's normal skill tool and /name invocation simply see more skills:

• Every other agent's skill directories. 60+ agents out of the box: Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, Windsurf, Kiro, Goose, OpenCode, Roo, Cline, Qwen Code, Trae and more. Project-level and user-level.
• Claude Code plugin marketplaces. The skills nested inside ~/.claude/plugins/marketplaces/*/plugins/*/skills/*, including the official Anthropic marketplace.
• Any git repository full of skills. Point at anthropics/skills, a sub-directory, a branch, a tag, or a commit. It is shallow-cloned into a local cache, refreshed in the background, and pinned in a lock file.
• Zero copies, zero symlinks. Files are read where they live and re-read on every load. Edit a skill in Cursor and dsh sees the change. Nothing to import, nothing to keep in sync.

Hundreds of skills would bloat every model request, so the provider keeps a catalog budget: at most 50 skills enter the model's session catalog by default, and the rest stay one find_skills call away through two small tools the plugin adds, with /name invocation untouched.

The same pool is available outside dsh too: dsh-skills-anywhere mcp serves it to any MCP client (Claude Code, Cursor, Codex, Windsurf…) as find_skills / open_skill tools and skill:// resources, so one install of a skill reaches every agent you use.

It also deduplicates symlinked and byte-identical installs (the skills CLI links one canonical copy into several agents), repairs common frontmatter drift instead of silently dropping a skill, and renames colliding names (discord/configure vs telegram/configure) so every skill stays reachable. A small CLI shows you exactly what dsh will see and why.

Quick start

# The package is not on npm yet; install the release tarball (no build step, no git needed).
PKG=https://github.com/noteflowai/dsh-skills-anywhere/releases/download/v0.3.1/dsh-skills-anywhere-0.3.1.tgz

# 1. Install into the dsh profile you use (web is the default UI profile)
dsh plugin --profile web add $PKG

# 2. See what the model will get, without booting dsh
npx -y $PKG list

# 3. Add a whole repository of skills
npx -y $PKG add anthropics/skills

Once the npm release is out, dsh-skills-anywhere replaces $PKG everywhere below, which is how the rest of this README writes the commands.

Start dsh as usual. The skill catalog now includes everything above; load a skill with the skill tool or /skill-name exactly as before.

On a machine with only Claude Code installed, list already finds the 31 skills inside the official plugin marketplace, none of which dsh sees on its own:

$ npx dsh-skills-anywhere list
NAME                 FROM                                                    PATH
discord-access       claude plugin discord @ claude-plugins-official         ~/.claude/plugins/marketplaces/.../discord/skills/access/SKILL.md
frontend-design      claude plugin frontend-design @ claude-plugins-official ~/.claude/plugins/marketplaces/.../frontend-design/skills/frontend-design/SKILL.md
skill-creator        claude plugin skill-creator @ claude-plugins-official   ~/.claude/plugins/marketplaces/.../skill-creator/skills/skill-creator/SKILL.md
...
31 skills, 6 renamed — run `dsh-skills-anywhere doctor` for details

dsh plugin --profile web add github:noteflowai/dsh-skills-anywhere

# $DSH_HOME/profiles/web/pnpm-workspace.yaml
allowBuilds:
  'dsh-skills-anywhere@https://codeload.github.com/noteflowai/dsh-skills-anywhere/tar.gz/<sha>': true

What gets discovered

Lower rank wins a duplicate name inside the dsh registry. The built-in dsh roots keep their ranks (.dsh/skills 100, .agents/skills 200, ~/.dsh/skills 400, ~/.agents/skills 500), so a skill you wrote for dsh always beats the same name found elsewhere. .agents/skills and .dsh/skills are deliberately not re-scanned here.

Run npx dsh-skills-anywhere agents for the full agent table and which directories exist on your machine.

Skill format

Any directory with a SKILL.md following the Agent Skills specification, plus dsh's flat <name>.md form. name, description, license, compatibility, allowed-tools, metadata, and dsh's disable-model-invocation / user-invocable are all understood. Unknown frontmatter (Claude Code's argument-hint, context, ...) is preserved under metadata.frontmatter. scripts/, references/ and assets/ are exposed through the skill's resource directory like any dsh skill.

In the default lenient mode a missing name falls back to the directory, an invalid name is normalised to kebab-case, and a missing description is derived from the first paragraph. Each repair is recorded and shown by doctor. Set lenient: false to match the strict behaviour of the built-in provider.

Git sources

npx dsh-skills-anywhere add anthropics/skills                      # default branch
npx dsh-skills-anywhere add anthropics/skills@v1.0.0               # tag or branch
npx dsh-skills-anywhere add vercel-labs/agent-skills/skills        # sub-directory
npx dsh-skills-anywhere add https://github.com/o/r/tree/main/dir   # GitHub tree URL
npx dsh-skills-anywhere add git@gitlab.com:group/skills.git        # any git URL
npx dsh-skills-anywhere add ./local/skills-repo --project          # local repo, project-scoped
npx dsh-skills-anywhere add o/r --ref 3f2a9c1 --rank 300           # pin a commit, set precedence

Sources come from three places, merged in this order: the plugin config.sources, the user file ~/.dsh/skills-anywhere/sources.json, and the project file <project>/.dsh/skills-anywhere.json (commit it to share skills with your team). The CLI edits the last two.

Each repository is shallow-cloned once into ~/.dsh/skills-anywhere/cache/<host>/<owner>/<repo> and refreshed when dsh starts, every syncIntervalMs (6 hours by default), and whenever a sources file changes. The resolved commit of every source is written to ~/.dsh/skills-anywhere/lock.json. Discovery only ever reads the cache, so a failed refresh means yesterday's skills, never an empty catalog. The catalog is invalidated as soon as a refresh brings changes; dsh never waits on the network.

Catalog budget and the find_skills / open_skill tools

dsh publishes every model-invocable skill's name and description into the session, on every request. With marketplaces and a few git sources that is hundreds of lines of context. The provider therefore ranks its skills and marks only the first catalog.limit (default 50) as model-invocable; the remainder is published with model invocation off, which keeps it out of the catalog but still loadable by you with /name.

Two tools, registered by the dsh-skills-anywhere/tools row, make the hidden part reachable for the model:

• find_skills(query, limit?) searches every skill by keyword (name, description, whenToUse, origin), catalog or not, and says which matches are listed.
• open_skill(name) loads any skill by exact name, including ones the budget hid. Skills whose own frontmatter says disable-model-invocation: true are still refused, exactly as the built-in skill tool does.

- id: skills-anywhere
  config:
    catalog:
      limit: 30                       # 0 = unlimited (old behaviour)
      pin: [frontend-design]          # always listed
      hide: [example-skill]           # never listed, still searchable and /name-invocable
- id: skills-anywhere-tools
  config:
    findLimit: 10

Author-disabled skills never count against the budget. Which skills stay listed follows the precedence order below, so project-level skills win over user-level, which win over marketplaces and git sources. The tools row needs the tool runtime (ctx.tools); in a profile without one it stays pending and the provider works alone.

CLI

dsh-skills-anywhere list [--all] [--json]     Skills the provider publishes (--all shows hidden duplicates)
dsh-skills-anywhere agents [--json]           Supported agents and which directories exist here
dsh-skills-anywhere sources [--json]          Configured git sources and their synced commits
dsh-skills-anywhere add <source> [--ref] [--path] [--rank] [--project]
dsh-skills-anywhere remove <source> [--project]
dsh-skills-anywhere sync [--force] [--json]   Clone or refresh every source now
dsh-skills-anywhere doctor [--json]           Repaired, skipped, renamed and duplicate skills, with reasons
dsh-skills-anywhere mcp                       Serve the same skills to any MCP client over stdio

All commands accept --cwd <dir> to pick the project. The CLI uses the same code path as the plugin and never needs dsh running.

Use as an MCP server

Skills are not a dsh-only idea, and neither is this provider. dsh-skills-anywhere mcp starts a Model Context Protocol server over stdio that exposes the identical pool (agent directories, Claude Code marketplaces, git sources, same dedupe and rename rules) to any MCP client:

Skills are also exposed as skill://<name> resources (with completion), for clients that let you @-mention resources. Skills whose frontmatter sets disable-model-invocation: true are never listed or opened. The server needs no dsh installation at all.

Claude Code (as a plugin; this repo doubles as a plugin marketplace)

claude plugin marketplace add noteflowai/dsh-skills-anywhere
claude plugin install dsh-skills-anywhere@noteflowai

Or register the bare server instead: claude mcp add skills-anywhere -- npx -y dsh-skills-anywhere mcp. Either way, restart Claude Code once so it connects.

Cursor (.cursor/mcp.json or ~/.cursor/mcp.json)

{ "mcpServers": { "skills-anywhere": { "command": "npx", "args": ["-y", "dsh-skills-anywhere", "mcp"] } } }

Codex (~/.codex/config.toml)

[mcp_servers.skills-anywhere]
command = "npx"
args = ["-y", "dsh-skills-anywhere", "mcp"]

Until the npm release lands, replace dsh-skills-anywhere in the manual commands with the release tarball URL (the plugin manifest already does this) (npx -y https://github.com/noteflowai/dsh-skills-anywhere/releases/download/v0.3.1/dsh-skills-anywhere-0.3.1.tgz mcp). Add --cwd <dir> when the client does not start the server inside the project you are working on. Git sources sync in the background on start, exactly as in dsh. Programmatic use: import { createSkillsAnywhereServer } from 'dsh-skills-anywhere/mcp' returns the McpServer and the provider so you can attach your own transport.

Configuration

Override the row in your profile's cordis.patch.yml. A patch replaces the whole config block, so restate every key you care about:

- id: skills-anywhere
  config:
    agents: true
    excludeAgents: [openclaw]
    claudePlugins: true
    sources:
      - anthropics/skills
      - { repo: vercel-labs/agent-skills, path: skills, ref: main, rank: 650 }
    excludeSkills: [example-skill]

The dsh-skills-anywhere/tools row accepts findLimit (default 10), findMaxLimit (50), and find / open booleans to register only one tool.

How precedence and duplicates work

1. Roots are scanned in rank order. Within one rank, the agent table order, then path.
2. Entries pointing at the same file (symlinks) collapse to the first. Entries with the same name and byte-identical body collapse to the first. Both appear in doctor as hidden duplicates.
3. Entries that still share a name but differ are all kept. If one of them is yours (an agent directory) it keeps the bare name and the others are prefixed with their plugin, repository, or agent (telegram-configure). If every member comes from a marketplace or a git source, all of them are prefixed, so you get discord-access and telegram-access rather than a meaningless bare access. doctor lists the renames.
4. The dsh registry then merges this provider's candidates with the built-in ones by rank.

Security notes

• The plugin reads skill files. It never writes to your agent directories.
• Git sources run git on your machine at plugin start and on the refresh interval. Pin a commit for anything you do not fully trust, and review lock.json.
• A skill is instructions the model will follow. Adding a source is a trust decision, exactly like installing a plugin.
• Skills are read with Node's filesystem API, not through dsh's sandboxed ctx.fs; the built-in provider does the same for its bundled root.

Development

pnpm install
pnpm run check        # typecheck + lint + tests + build
pnpm pack             # tarball for `dsh plugin --profile <name> add ./dsh-skills-anywhere-*.tgz`

Tests run against the real @deepseek-ai/dsh-skill registry and real git repositories in temp directories.

Contributing

Issues and pull requests are welcome. Adding an agent is a one-line change in src/agents.ts. See CONTRIBUTING.md.

License

MIT © Note Flow AI