YouTip

Skills are essentially operation manuals that teach AI to follow a fixed process. Once written, they can be called repeatedly like functions.\\\\n\\\\nWe can think of Skills as encapsulating "how a certain type of task should be done professionally" into a reusable, automatically triggerable capability module.\\\\n\\\\nSkills exist as Markdown files. They do not execute functions directly; instead, through on-demand, progressive loading, they achieve efficient and reusable knowledge transfer.\\\\n\\\\n\\\\n\\\\nThe biggest difference between Skills and traditional Prompts is: on-demand loading + progressive disclosure (only stuffing the thick SOP into the context when needed, greatly saving tokens).\\\\n\\\\n| Comparison Item | Ordinary Prompt | Skills Mechanism |\\\\n| --- | --- | --- |\\\\n| Needs to be re-described every time | Yes | No (described only once) |\\\\n| Context length usage | Full insertion every time | Progressive loading (full content only read when triggered) |\\\\n| Consistency | Depends on prompt quality each time | High (fixed SOP + templates) |\\\\n| Reusability | Manual copy-paste | Auto-matching / slash commands / project sharing |\\\\n| Maintenance method | Re-send every time a prompt is changed | Modify the SKILL.md file, takes effect globally/for the project |\\\\n\\\\nFor example, when we usually write articles, before having Skills, we had to repeat the following steps every time:\\\\n\\\\nHelp me summarize the article → translate → adapt to WeChat Official Account style → add title → output Markdown\\\\nWith Skills:\\\\n\\\\nYou only need one sentence:\\\\n\\\\nUse the "Tech Article to WeChat Official Account" Skill\\\\nAI will automatically execute according to the steps you set.\\\\n\\\\nImagine AI as a **smart but inexperienced fresh graduate intern**:\\\\n\\\\n* **Ordinary Prompt** = You have to teach them how to do things from scratch every time (teach them once today, and you have to reteach them tomorrow)\\\\n* **Rule / Memory** = You post a "Company Code of Conduct" at their workstation (always effective, but can only govern attitude and format)\\\\n* **MCP / Tools** = You install a bunch of software and APIs on their computer (they can call external tools, but don't know when to use them or how to combine them)\\\\n* **Skills** = You directly give them a complete **"Job Training Package"** (PDF + flowchart + SOP + script templates + common scripts), telling them: "When the boss asks you to do this kind of task, just follow the methods in this folder"\\\\n\\\\nMainstream clients that currently support Skills:\\\\n\\\\n| Rank | Tool Name | Free to use Skills? | Recommended for | Default Skill Storage Path | Notes |\\\\n| --- | --- | --- | --- | --- | --- |\\\\n| 1 | Claude Code | Yes (Official) | Everyone | ~/.claude/skills | Standard setter, most complete ecosystem |\\\\n| 2 | Cursor | Yes | Most commonly used for coding | ~/.cursor/skills | Almost seamlessly compatible with Claude Skills |\\\\n| 3 | Trae / OpenCode | Yes | Pursuing cost-effectiveness | See tool settings | Many domestic users |\\\\n| 4 | VS Code + Plugin | Partial support | Already deep into vscode | Configure in plugin settings | Catching up quickly |\\\\n| 5 | Coze/Other domestic platforms | Partial support | Like web versions | Platform's built-in skill market | Some require membership |\\\\n\\\\n**Difference between Skills and MCP:** Skills are for **knowledge reuse**, MCP is for **capability extension**.\\\\n\\\\n## Skills\\\\n\\\\nKnowledge reuse\\\\n\\\\n* Knowledge sharing: experience, best practices, workflows\\\\n* Based on simple Markdown files, anyone can create them\\\\n* Progressive loading, high Token usage efficiency\\\\n* No server or backend setup required\\\\n* Applicable to Web / Desktop / CLI\\\\n\\\\n## MCP\\\\n\\\\nCapability extension\\\\n\\\\n* Function extension: connect APIs, databases, external tools\\\\n* Requires coding skills and server-side configuration\\\\n* Loads all tool definitions at startup\\\\n* Strong integration capability with external systems\\\\n* Higher Token consumption and complexity\\\\n\\\\n* * *\\\\n\\\\n## Core Structure of a Skill\\\\n\\\\nThe core of Skills is: a folder + a SKILL.md file.\\\\n\\\\n**The SKILL.md file contains:**\\\\n\\\\n* Metadata (at least a name and description)\\\\n* Instructions telling the AI how to complete a specific task\\\\n\\\\n\\\\n\\\\nA Skill is essentially a Markdown file (filename fixed as SKILL.md)\\\\n\\\\nmy-skill/\\\\n└── SKILL.md （Only required)\\\\n\\\\nSKILL.md basic template:\\\\n\\\\n---\\\\nname: pdf-processing\\\\ndescription: From PDF inExtract textand tables, Fill out forms, and merge document\\\\n---\\\\n\\\\n# PDF Handle.\\\\n\\\\n## UseScenarios\\\\nUse when operations on PDF files are required, for example:\\\\n- Extract text or table data from PDFs\\\\n- Fill out PDF forms\\\\n- Merge multiple PDF files\\\\n\\\\n## Extract text\\\\n- Use `pdfplumber` Extract content from text-based PDFs \\\\n- Scanned PDFs require OCR tools \\\\n\\\\n## Fill out forms\\\\n- Read PDF form fields \\\\n- Populate and generate new files based on input data \\\\n\\\\nMinimum required example:\\\\n\\\\n---\\\\nname: skill-name\\\\ndescription: Explain the functionality of this Skill and applicable Scenarios\\\\n---\\\\n\\\\nExample with optional fields:\\\\n\\\\n---\\\\nname: pdf-processing\\\\ndescription: From PDF inExtract textand tables, Fill out forms, and merge document\\\\nlicense: Apache-2.0\\\\nmetadata:\\\\n author: example-org\\\\n version: "1.0"\\\\n---\\\\n\\\\n| Field | Required | Description |\\\\n| --- | --- | --- |\\\\n| name | Yes | Skill name, up to 64 characters, can only use lowercase letters, numbers, and `-`, and cannot start or end with `-` |\\\\n| description | Yes | Function and usage scenario description, up to 1024 characters, cannot be empty |\\\\n| license | No | License name or link to the license file included with the Skill |\\\\n| compatibility | No | Environment and dependency description (products, system packages, network permissions, etc.), up to 500 characters |\\\\n| metadata | No | Custom key-value pairs for extending metadata (e.g., author, version number) |\\\\n| allowed-tools | No | List of allowed tools (space-separated, experimental feature) |\\\\n\\\\nIf you need reference materials, examples, or execution scripts, you can use a more complex Skill directory structure:\\\\n\\\\nmy-skill/\\\\n├── SKILL.md # Required: instruction + metadata\\\\n├── scripts/ # Optional: executable code\\\\n├── references/ # Optional：Documentation\\\\n└── assets/ # Optional: templates, resources\\\\n\\\\n\\\\n\\\\n### How Skills Work\\\\n\\\\nSkills use progressive loading to efficiently manage context:\\\\n\\\\n* **Discovery:** At startup, the AI only loads the name and description of each skill, keeping only the most basic identification information.\\\\n* **Activation:** When a task matches a skill's description, the AI reads the complete SKILL.md instructions into the context.\\\\n* **Execution:** The AI follows the instructions, loading reference files or running code as needed.\\\\n\\\\nThis design keeps the AI fast while allowing it to acquire more information on demand.\\\\n\\\\n* * *\\\\n\\\\n## Claude Code Skills\\\\n\\\\nNext, let's use Claude Code as an example to create a simple Skill.\\\\n\\\\nClaude Code finds and loads Skills in the following order (the more specific the location, the higher the priority):\\\\n\\\\n| Level | Path | Scope |\\\\n| --- | --- | --- |\\\\n| Enterprise | Configured via management console (managed settings) | All users in the organization |\\\\n| Personal | `~/.claude/skills//SKILL.md` | All your projects |\\\\n| Project | `.claude/skills//SKILL.md` | Current project only |\\\\n| Plugin | `/skills//SKILL.md` | Environments where the plugin is enabled |\\\\n\\\\nEach Skill is a folder; the folder name is the skill identifier (kebab-case lowercase + hyphens recommended).\\\\n\\\\nMinimal structure:\\\\n\\\\n~/.claude/skills/\\\\n└── code-comment-expert/ # Skill folder name\\\\n └── SKILL.md # Only required file (must be all uppercase + .md lowercase)\\\\n\\\\nSKILL.md complete format:\\\\n\\\\n--- # YAML frontmatter Start (at the beginning of the line)\\\\nname: code-comment-expert # Required: Skill name (also serves as the /slash command name)\\\\ndescription: >- # Required: The most critical line! Claude relies on it to determine whether to AddLoad\\\\n Add professional, clear Chinese-English bilingual Comments to the code.\\\\n Suitable for code lacking documentation, having poor readability, or requiring sharing and review.\\\\n Common trigger Scenarios: AddComments, add comments, add document, explain this, improve readability\\\\ntrigger_keywords: # Highly recommended (significantly boosts auto-trigger rate)\\\\n - AddComments\\\\n - Comments\\\\n - Adddocument\\\\n - explain code\\\\n - document\\\\n - comment this\\\\n - readability\\\\nversion: 1.0 # Optional\\\\nauthor: yourname # Optional\\\\n--- # ← YAML End\\\\n\\\\n# Main content starts here (Markdown) — actual instructions executed by Claude\\\\n\\\\nYou are now a 「