skills 完整教程：shanraisshan/claude-code-best-practice: claude-ski，适用于工程实践、代码质量与开发流程优化。
skills 完整教程：shanraisshan/claude-code-best-practice: claude-ski，适用于工程实践、代码质量与开发流程优化。
> 来源: shanraisshan/claude-code-best-practice (64k stars) — Claude Code 完整实践指南
> 原文件: best-practice/claude-skills.md
> 模型推荐: 看 skill 类型挑
shanraisshan 在 64k stars 的仓库里整理的"Claude Code 完整实践指南" — 涵盖启动参数、commands、MCP、memory、settings、skills、subagents 等所有方面。
michael 强调"skill 要有相应的指导功能,指导用户使用",所以加了下面两节让 Agent 和用户对接。
---
1. 用户问"Claude Code 怎么用 X"时,触发对应 skill
2. skill 严格按原文指南跑
3. 涉及配置改动前 ask user 确认(避免破坏现有配置)
4. 完工后让 Agent 跑自检
1. 问 Agent "Claude Code 怎么 X"(X = 子主题)
2. 如果 Agent 改配置,确认后再继续
3. 全程 Agent 自动化,你只需回答授权类问题
---
 <br>
[](../implementation/claude-skills-implementation.md)
Claude Code skills — frontmatter fields and official bundled skills.
<table width="100%">
<tr>
<td><a href="../">← Back to Claude Code Best Practice</a></td>
<td align="right"><img src="../!/claude-jumping.svg" alt="Claude" width="60" /></td>
</tr>
</table>
---
| Field | Type | Required | Description |
|-------|------|----------|-------------|
| `name` | string | No | Display name and `/slash-command` identifier. Defaults to the directory name if omitted |
| `description` | string | Recommended | What the skill does. Shown in autocomplete and used by Claude for auto-discovery |
| `when_to_use` | string | No | Additional context for when Claude should invoke the skill — trigger phrases and example requests. Appended to `description` in the skill listing, counts toward the 1,536-character cap |
| `argument-hint` | string | No | Hint shown during autocomplete (e.g., `[issue-number]`, `[filename]`) |
| `arguments` | string/list | No | Named positional arguments for `$name` substitution in the skill content. Accepts a space-separated string or a YAML list — names map to argument positions in order |
| `disable-model-invocation` | boolean | No | Set `true` to prevent Claude from automatically invoking this skill |
| `user-invocable` | boolean | No | Set `false` to hide from the `/` menu — skill becomes background knowledge only, intended for agent preloading |
| `allowed-tools` | string | No | Tools allowed without permission prompts when this skill is active |
| `disallowed-tools` | string/list | No | Tools removed from Claude's available pool while the skill is active (e.g. block `AskUserQuestion` for a background loop). Accepts a space/comma-separated string or YAML list — the restriction clears on the next message |
| `model` | string | No | Model to use when this skill runs (e.g., `haiku`, `sonnet`, `opus`) |
| `effort` | string | No | Override the model effort level when invoked (`low`, `medium`, `high`, `xhigh`, `max`) |
| `context` | string | No | Set to `fork` to run the skill in an isolated subagent context |
| `agent` | string | No | Subagent type when `context: fork` is set (default: `general-purpose`) |
| `background` | boolean | No | Only applies with `context: fork`. Set to `false` to wait for the forked subagent's result in the invoking turn instead of running in the background (default: `true`). Requires v2.1.218+ |
| `hooks` | object | No | Lifecycle hooks scoped to this skill |
| `paths` | string/list | No | Glob patterns that limit when the skill auto-activates. Accepts a comma-separated string or YAML list — Claude loads the skill only when working with matching files |
| `shell` | string | No | Shell for `` !`command` `` blocks — `bash` (default) or `powershell`. Requires `CLAUDE_CODE_USE_POWERSHELL_TOOL=1` |
| `metadata` | YAML map | No | Free-form YAML map for your own key-value data (e.g., entitlement or catalog fields) read by your own tooling from `SKILL.md`. Claude Code does not act on its contents; drops values that aren't a map. Do not reuse frontmatter field names (e.g., `paths`) as keys |
| `license` | string | No | License covering the skill. Part of the [Agent Skills](https://agentskills.io) spec. Claude Code accepts the field but does not act on it |
| `compatibility` | string | No | Environment requirements for the skill (max 500 chars), such as intended products or system prerequisites. Part of the [Agent Skills](https://agentskills.io) spec. Claude Code accepts the field but does not act on it |
---
| # | Skill | Description |
|---|-------|-------------|
| 1 | `code-review` | Review the current diff for correctness bugs at a chosen effort level (low/medium: fewer, high-confidence findings; high→max: broader coverage) — `--comment` posts findings as inline PR comments |
| 2 | `batch` | Run commands across multiple files in bulk |
| 3 | `debug` | Debug failing commands or code issues |
| 4 | `loop` | Run a prompt or slash command on a recurring interval (up to 3 days) |
| 5 | `claude-api` | Build apps with the Claude API or Anthropic SDK — triggers on `anthropic` / `@anthropic-ai/sdk` imports |
| 6 | `fewer-permission-prompts` | Scan transcripts for common read-only Bash/MCP calls and add a prioritized allowlist to `.claude/settings.json` to reduce permission prompts |
| 7 | `run` | Launch and drive the project's app to see a change working in the real app (not just tests). Requires v2.1.145 |
| 8 | `verify` | Build and run the app to confirm a code change does what it should, without falling back to tests or type checks. Requires v2.1.145 |
| 9 | `run-skill-generator` | Teaches `/run` and `/verify` how to build and launch the project — records a per-project launch recipe at `.claude/skills/run-<name>/`. Requires v2.1.145 |
| 10 | `simplify` | Review changed code for cleanup opportunities (reuse, simplification, efficiency, abstraction level), four review agents in parallel. From v2.1.154 it does **not** hunt for correctness bugs — use `/code-review` for that |
| 11 | `design-sync` | Convert your repo's React design system and upload it to Claude Design — optionally name the design system (e.g., `/design-sync Acme DS`). First-time sync verifies every component and can take hours on large repos. Available on the Anthropic API only (unavailable on Bedrock, Google Cloud Agent Platform, and Microsoft Foundry) |
| 12 | `dataviz` | Design charts, graphs, and dashboards with a color-palette validator for accessible, consistent visualizations — triggers on requests for any chart, graph, plot, or data visualization in any output medium. Introduced v2.1.198 |
| 13 | `doctor` | Setup/health checkup for your Claude Code configuration. The one bundled skill exempt from `disableBundledSkills` — stays typable even when that setting is on. Reclassified from a built-in command to a bundled skill in v2.1.205 |
| 14 | `review` | Fast single-pass, read-only review of a GitHub pull request. For a multi-agent deep dive, use `/code-review <level> <pr#>` instead. From v2.1.186–v2.1.201 used the `/code-review medium` engine; v2.1.202 reverted to a fast single-pass review. Became Skill-tool-invocable in v2.1.108 |
| 15 | `security-review` | Review the current diff for security vulnerabilities and suggest fixes. Pass `--fix` to apply findings or `--comment` to post them as inline GitHub PR comments. Became Skill-tool-invocable in v2.1.108 |
See also: [Official Skills Repository](https://github.com/anthropics/skills/tree/main/skills) for community-maintained installable skills.
---
本 skill 专注于skills 完整教程，shanraisshan/claude-code-best-practice: claude-skills。它将相关流程标准化，帮助用户更快拿到可靠结果，减少重复手工操作。
当你需要在skills 完整教程相关工作中获得稳定、可复用的产出时最适合——无论是单次任务还是纳入日常工作流，都能直接调用。
需要一个具体的项目或任务上下文，最好带有代码仓库或需求文档。
1. 问 Agent "Claude Code 怎么 X"(X = 子主题)
2. 如果 Agent 改配置,确认后再继续
3. 全程 Agent 自动化,你只需回答授权类问题
---
本 skill 专注于skills 完整教程，shanraisshan/claude-code-best-practice: claude-skills。它将相关流程标准化，帮助用户更快拿到可靠结果，减少重复手工操作。
当你需要在skills 完整教程相关工作中获得稳定、可复用的产出时最适合——无论是单次任务还是纳入日常工作流，都能直接调用。
需要一个具体的项目或任务上下文，最好带有代码仓库或需求文档。