joshrotenberg/skillet

ملف joshrotenberg

آخر نشاط مصدر مسجل: ٢٧ مارس ٢٠٢٦ في ٢٣:٢٢

آخر تحديث لفهرس SkillsMP: ١٧ أكتوبر ٢٠٢٦ في ٠٠:٠٩

skills مجمعة: ٦

نجوم GitHub: ٨
تفرعات GitHub: ٠

ملف README للمستودع

مصدر ملف README للمستودع: README.md

قراءة الملف الأصلي

عرض ملف README كاملًا

<p align="center"> <img src="assets/logo.png" alt="Skillet" width="360"> </p> <p align="center"> <a href="https://github.com/joshrotenberg/skillet/actions/workflows/ci.yml"><img src="https://github.com/joshrotenberg/skillet/actions/workflows/ci.yml/badge.svg" alt="CI"></a> <a href="https://crates.io/crates/skillet-mcp"><img src="https://img.shields.io/crates/v/skillet-mcp.svg" alt="crates.io"></a> <a href="https://github.com/joshrotenberg/skillet/blob/main/Cargo.toml"><img src="https://img.shields.io/badge/license-MIT%2FApache--2.0-blue" alt="License"></a> </p> # skillet MCP-native skill discovery for AI agents. ## What is skillet Skillet serves agent skills as MCP prompts. Your agent connects, searches for skills relevant to the current task, and uses them immediately via `prompts/get` -- no installation, no restart. It also works as a CLI for browsing skills and managing repos. Skills are just SKILL.md files in git repos. Anyone can create a skill repo and share it. Skillet indexes them, runs full-text search, and delivers them as MCP prompts at runtime. Think of it like git: skillet is the tool, repos are distributed. Skills follow the [Agent Skills specification](https://docs.anthropic.com/en/docs/claude-code/skills) and work with Claude Code, Cursor, Copilot, Windsurf, Gemini CLI, and any compatible agent. ## Install ### Homebrew (macOS / Linux) ```bash brew install joshrotenberg/brew/skillet-mcp ``` ### Shell script ```bash curl --proto '=https' --tlsv1.2 -LsSf https://github.com/joshrotenberg/skillet/releases/latest/download/skillet-mcp-installer.sh | sh ``` ### Cargo ```bash cargo install skillet-mcp ``` ### Docker ```bash docker pull ghcr.io/joshrotenberg/skillet:latest docker run --rm -i ghcr.io/joshrotenberg/skillet:latest ``` Pre-built binaries for macOS, Linux, and Windows are also available on the [releases page](https://github.com/joshrotenberg/skillet/releases). ## Quick start ### Give your agent access to skills (fastest) Add skillet to your agent's MCP config and you're done -- your agent can search for skills and use them as prompts at runtime: ```json { "mcpServers": { "skillet": { "command": "skillet" } } } ``` Or with Docker (zero install): ```json { "mcpServers": { "skillet": { "command": "docker", "args": ["run", "-i", "--rm", "ghcr.io/joshrotenberg/skillet"] } } } ``` That's it. Skillet auto-discovers skills from the [official repo](https://github.com/joshrotenberg/skillet/tree/main/skills) plus 14 community repos (Anthropic, Vercel, Firebase, Supabase, Google Gemini, Microsoft, and more) via the [suggest graph](#decentralized-discovery-suggest-graph). Your agent now has access to `search_skills` and the full [MCP interface](#mcp-interface). For custom or team repos, add `--remote` or `--repo` args: ```json { "mcpServers": { "skillet": { "command": "skillet", "args": [ "--repo", "/path/to/local-skills", "--remote", "https://github.com/acme/team-skills.git" ] } } } ``` ### Use the CLI ```bash # Browse everything skillet search '*' # Search by keyword, category, tag, or owner skillet search rust skillet search '*' --category development skillet search '*' --owner joshrotenberg # See what categories exist skillet categories # Show details about a skill skillet info joshrotenberg/rust-dev ``` ### Embed skills in your project Add skills directly to any repo -- no separate repo required: ```bash # Generate a skillet.toml with an inline skill skillet init --skill # Write your skill prompt $EDITOR SKILL.md # That's it. Run skillet from this directory and the skill is served. ``` ### Manage repos ```bash # Add a remote skill repo skillet repo add https://github.com/acme/team-skills.git # Add a local skill directory skillet repo add /path/to/local-skills # See what's configured skillet repo list # Remove a repo skillet repo remove https://github.com/acme/team-skills.git ``` ### What it looks like **Your agent, with skillet as an MCP server:** ``` User: Set up a new Rust project with CI Agent: Let me search for relevant skills. [calls search_skills(query: "rust", category: "development")] Found joshrotenberg/rust-dev. Let me use its instructions. [calls prompts/get(name: "joshrotenberg_rust-dev")] I'll follow the rust-dev skill conventions for project setup... ``` **CLI:** ``` $ skillet search rust joshrotenberg/rust-dev Rust development standards and conventions joshrotenberg/rust-ci Rust CI/CD pipeline setup Found 2 skills matching "rust" ``` ## Features ### Skills as MCP prompts Every skill in the index is registered as an MCP prompt, namespaced as `owner_skill-name`. Agents discover skills via `prompts/list` and fetch content via `prompts/get`. When the index refreshes, prompts are synced automatically and a `prompts/list_changed` notification is emitted. ### Multi-repo support Aggregate local directories and remote git repos. Skillet clones and periodically refreshes remotes in the background. Use `--refresh-interval` to control how often (default: every 5 minutes). ```bash # CLI: explicit repos skillet search rust \ --repo /path/to/local \ --remote https://github.com/acme/team-skills.git # Or configure defaults in ~/.config/skillet/config.toml ``` ### Decentralized discovery (suggest graph) Repos can suggest other repos via `[[suggest]]` entries in their `skillet.toml`. Skillet follows these on startup to build a discovery graph -- no central authority needed. The official repo suggests 14 community and vendor repos (Anthropic, Vercel, Firebase, Supabase, Google Gemini, Microsoft, Redis, Callstack, Kepano, TerminalSkills, Softaworks, Anthony Fu, Daymade, and more) so you get broad coverage out of the box. ```toml # In any repo's skillet.toml [[suggest]] url = "https://github.com/acme/agent-skills.git" description = "Acme team skills" ``` Skills discovered through the suggest graph are stamped with a hop-based trust tier: | Tier | Description | |---|---| | **Direct** | From a directly configured repo (depth 0) | | **Suggested** | One hop away via a `[[suggest]]` link | | **Transitive** | Two or more hops from a configured repo | The suggest walker has configurable safety limits: max depth, max repos, per-repo fan-out caps, clone timeout, URL allow/block lists, and negative caching for failed URLs. Opt out with `--no-suggest` or set `enabled = false` under `[suggest]` in config. ### Release model resolution Skillet respects release conventions when fetching repos: 1. **Consumer pin** (`[[source]]` in config.toml) -- highest priority. Lock a repo to a specific tag, branch, or commit. 2. **Author preference** (`[source]` in skillet.toml) -- the repo author declares `prefer = "release"`, `"main"`, or `"tag:v*"`. 3. **Auto-detect** -- if neither is set, skillet checks for a latest release tag; if none exists, it stays on the default branch. ```toml # Consumer-side pin in ~/.config/skillet/config.toml [[source]] repo = "github.com/someone/skills" version = "v2.1.0" ``` ```toml # Author-side preference in skillet.toml [source] prefer = "release" ``` ### Project manifest (skillet.toml) Embed skills directly in any repository with a `skillet.toml` at the project root. No separate repo needed -- skillet auto-detects the manifest and serves embedded skills alongside repo skills. ```toml [project] name = "my-tool" description = "A CLI tool with agent skills" [[project.authors]] name = "Alice" github = "alice" # Single inline skill (SKILL.md at project root) [skill] name = "my-tool-usage" description = "How to use my-tool" ``` Two layout modes: - **Single skill** (`[skill]`): one SKILL.md at the project root - **Multi-skill** (`[skills]`): a `.skillet/` directory with multiple skills All sections are optional and can be combined. Scaffold one with: ```bash skillet init --skill # single skill skillet init --multi # multi-skill directory ``` ### Zero-config skill discovery Directories with only a `SKILL.md` are fully discoverable. YAML frontmatter in SKILL.md is the primary metadata source: ```markdown --- name: my-skill description: What this skill does version: 1.0.0 trigger: When to activate this skill categories: - development tags: - rust - testing --- ## My Skill Instructions for the agent... ``` When frontmatter is absent, metadata is inferred automatically from the directory name, git remote, and SKILL.md content. `skill.toml` is supported as a legacy fallback but frontmatter is preferred. ### BM25 full-text search Search indexes skill names, descriptions, categories, tags, and SKILL.md content. Results are ranked by relevance using BM25 scoring with field-weighted boosting. ### Persistent disk cache The skill index is cached to disk and refreshed based on TTL (default: 5 minutes). The suggest graph has its own TTL (default: 1 hour). No rebuilding from scratch every time. Use `--no-cache` to bypass. ### Filesystem watching Use `--watch` with the MCP server to auto-reload when local repo files change. Useful during skill development. ### Configurable server exposure Control which MCP tools are exposed: ```bash # Read-only mode skillet --read-only # Explicit allowlist skillet --tools search,categories,info ``` ## MCP interface When running as an MCP server, agents discover skills via tools and fetch skill content as prompts. ### Tools | Tool | Purpose | |---|---| | `search_skills` | Full-text search with category, tag, and model filters | | `list_categories` | Browse all skill categories with counts | | `list_skills_by_owner` | List all skills by a specific publisher | | `info_skill` | Detailed information about a specific skill | | `annotate_skill` | Attach a persistent note to a skill | ### Prompts Every indexed skill is served as an MCP prompt via `prompts/list` and `prompts/get`. Prompt names are namespaced as `owner_skill-name` (e.g. `joshrotenberg_rust-dev`). The prompt description comes from SKILL.md frontmatter and the prompt content is the full SKILL.md text. When the index refreshes (repo pull, filesystem watch, or cache expiration), prompts are synced automatically: new skills are registered, removed skills are unregistered, and a `prompts/list_changed` notification is emitted to connected clients. ## CLI reference ### Use skills | Command | Description | |---|---| | `skillet search <query>` | Search for skills (`*` for all). Supports `--category`, `--tag`, `--owner` | | `skillet categories` | List all skill categories with counts | | `skillet info <owner/name>` | Show detailed information about a skill | ### Author skills | Command | Description | |---|---| | `skillet init [path]` | Generate a `skillet.toml` project manifest. Supports `--skill`, `--multi` | ### Manage repos | Command | Description | |---|---| | `skillet repo add <url_or_path>` | Add a remote or local repo to config | | `skillet repo remove <url_or_path>` | Remove a repo from config | | `skillet repo list` | List configured repos | | `skillet [serve]` | Run the MCP server (default when stdin is not a terminal) | ### Server options | Flag | Description | |---|---| | `--repo <path>` | Local repo directory (repeatable) | | `--remote <url>` | Git URL to clone and serve from (repeatable) | | `--refresh-interval <duration>` | How often to pull from remotes (default: `5m`, `0` to disable) | | `--cache-dir <path>` | Directory to clone remote repos into | | `--subdir <path>` | Subdirectory within repos containing skills | | `--watch` | Watch local repos for changes and auto-reload | | `--http <addr>` | Serve over HTTP instead of stdio (e.g. `0.0.0.0:8080`) | | `--read-only` | Expose only read-only tools | | `--tools <list>` | Explicit tool allowlist (comma-separated) | | `--no-suggest` | Don't follow `[[suggest]]` entries from repos | **HTTP transport note**: `--http` disables origin validation to allow connections from any origin. It is intended for local development and trusted networks. In production, place behind a reverse proxy with authentication and CORS configuration. ## Configuration Skillet reads `~/.config/skillet/config.toml` for defaults. Create it manually or use `skillet repo add` to populate repos: ```toml [repos] remote = ["https://github.com/joshrotenberg/skillet.git"] local = [] follow_suggestions = true # follow [[suggest]] entries from repos suggest_depth = 1 # max recursion depth for suggestions [cache] enabled = true ttl = "5m" # index cache time-to-live suggest_ttl = "1h" # suggest graph cache TTL [suggest] enabled = true max_depth = 2 # max BFS depth for suggest graph max_per_repo = 5 # max suggestions per repo max_repos = 20 # total max repos via suggestions clone_timeout = "30s" negative_cache_ttl = "1h" allow = [] # URL glob allowlist (empty = allow all) block = [] # URL glob blocklist [server] tools = [] # empty = expose all resources = [] # empty = expose all # Consumer-side version pinning [[source]] repo = "github.com/someone/skills" version = "v2.1.0" ``` ## Skill format A skill is a directory with a `SKILL.md` file containing YAML frontmatter for metadata and markdown content for the prompt: ``` owner/skill-name/ SKILL.md # Prompt with YAML frontmatter (required) scripts/ # Optional executable scripts references/ # Optional reference docs assets/ # Optional templates, configs ``` ```markdown --- name: rust-dev description: Rust development standards and conventions version: 2026.02.24 trigger: Use when writing or reviewing Rust code license: MIT author: Josh Rotenberg categories: - development - rust tags: - rust - cargo - clippy - fmt - testing --- ## Rust Development Standards ### Pre-commit Checklist ... ``` All frontmatter fields are optional -- skillet infers what it can from the directory name, git remote, and content. Fully compatible with the [Agent Skills specification](https://docs.anthropic.com/en/docs/claude-code/skills). ## Status Skills are served as MCP prompts with YAML frontmatter as the primary metadata source. The suggest graph discovers skills across 14 default repos plus any you configure. Release model resolution respects author preferences and consumer pins. The CLI covers search, info, categories, init, and repo management. See [open issues](https://github.com/joshrotenberg/skillet/issues) for what's next. ## License MIT OR Apache-2.0

جارٍ تحميل ملف README للمستودع…

Skills في هذا المستودع

مطوّرو البرمجيات مديرو الشبكات وأنظمة الحاسوب

٢ فئات مهنية · 100% مصنفة

skill

المهنة

الوصف

آخر تحديث

skill-repos

مطوّرو البرمجيات

Curated directory of external skill repositories. Use when the user needs skills not available locally, or to suggest adding new skill sources.

لغة النص الأصلي: الإنجليزية

2026-03-27

contributor

مطوّرو البرمجيات

Contributing to skillet itself. Covers architecture, development setup, testing, and PR workflow.

لغة النص الأصلي: الإنجليزية

2026-03-27

repo-maintainer

مطوّرو البرمجيات

Creating and maintaining skill repos. Covers setup, directory layout, PR review, multi-repo, and hosting.

لغة النص الأصلي: الإنجليزية

2026-03-27

setup

مطوّرو البرمجياتمديرو الشبكات وأنظمة الحاسوب

Set up and configure Skillet skill discovery. Use when the user wants to set up skillet, configure repos, or customize server behavior.

لغة النص الأصلي: الإنجليزية

2026-03-27

skill-author

مطوّرو البرمجيات

Authoring skills for skillet. Covers the skill format, discovery via suggest, and project manifests.

لغة النص الأصلي: الإنجليزية

2026-03-27

user

مطوّرو البرمجيات

Using skillet as a skill consumer. Covers searching, browsing, and using skills as MCP prompts.

لغة النص الأصلي: الإنجليزية

2026-03-27

عرض ٦ من أصل ٦ skills مجمعة.

التثبيت بمساعدة مساعد ذكي

انسخ هذا الطلب إلى المساعد الذكي الذي تستخدمه.

Follow https://skillsmp.com/skill-install/prompt.md to install Agent Skills from https://github.com/joshrotenberg/skillet.