build-cli-plugin
정보
이 스킬은 추상 베이스 클래스 패턴을 사용하여 CLI 플러그인이나 어댑터를 구축하기 위한 템플릿을 제공합니다. 플러그인 계약 정의, 설치 전략(심볼릭 링크/복사/추가) 구현, 그리고 탐지, 등록, 멱등성 작업 처리 방법을 다룹니다. CLI 도구를 새로운 프레임워크 지원으로 확장하거나 다중 대상 애플리케이션을 위한 플러그인 시스템을 생성할 때 사용하세요.
빠른 설치
Claude Code
추천npx skills add pjt222/agent-almanac -a claude-code/plugin add https://github.com/pjt222/agent-almanacgit clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/build-cli-pluginClaude Code에서 이 명령을 복사하여 붙여넣어 스킬을 설치하세요
문서
Build a CLI Plugin
Add a new plugin or adapter to a CLI tool's pluggable architecture using the abstract base class pattern.
When to Use
- Adding support for a new target framework to a CLI installer
- Building a plugin system for a multi-target command-line tool
- Extending an existing adapter architecture with a new strategy variant
- Porting content delivery to a framework that uses a different file layout
Inputs
- Required: Framework or target the plugin supports (name, config paths, conventions)
- Required: Path to the base class or plugin contract
- Required: Installation strategy:
symlink,copy,file-per-item, orappend-to-file - Optional: Content types the plugin handles (e.g., skills only, skills + agents, full support)
- Optional: Scope support (project-level, global, both)
Procedure
Step 1: Define the Contract
The base class establishes the interface all plugins must implement:
export class FrameworkAdapter {
static id = 'base'; // Unique identifier
static displayName = 'Base'; // Human-readable name
static strategy = 'symlink'; // Installation strategy
static contentTypes = ['skill']; // What this adapter handles
async detect(projectDir) { return false; }
getTargetPath(projectDir, scope) { throw new Error('Not implemented'); }
async install(item, projectDir, scope, options) { throw new Error('Not implemented'); }
async uninstall(item, projectDir, scope, options) { throw new Error('Not implemented'); }
async listInstalled(projectDir, scope) { return []; }
async audit(projectDir, scope) { return { framework: this.constructor.displayName, ok: [], warnings: [], errors: [] }; }
supports(contentType) { return this.constructor.contentTypes.includes(contentType); }
}
Static fields define the plugin's identity and capabilities:
id: Used in--framework <id>option and result reportingdisplayName: Shown in human-readable outputstrategy: Determines how content reaches the targetcontentTypes: Filters which items this adapter receives
If the base class does not exist yet, create it first. The pattern scales to any number of plugins.
Got: A base class with static identity fields and abstract methods.
If fail: If the base class has methods that don't apply to all plugins (e.g., not all frameworks support audit), provide default implementations that return sensible no-ops.
Step 2: Choose the Installation Strategy
| Strategy | When to use | Example |
|---|---|---|
| symlink | Target reads source files directly. Cheapest, stays in sync. | Claude Code reads .claude/skills/<name>/ symlinks |
| copy | Target needs files in its own directory. Modifications don't propagate. | Some IDEs index only their own dirs |
| file-per-item | Target expects one file per item with specific format. | Cursor .mdc rules files |
| append-to-file | Target reads a single instructions file. | Aider CONVENTIONS.md, Codex AGENTS.md |
Strategy determines the implementation shape:
- Symlink:
symlinkSync(source, target)— handle relative vs. absolute paths - Copy:
cpSync(source, target, { recursive: true })— handle overwrites - File-per-item:
writeFileSync(target, transform(content))— may need format conversion - Append-to-file: Wrap content in markers for idempotent insert/replace/remove
Got: Strategy selected with clear rationale based on how the target framework discovers content.
If fail: If unsure, check the framework's documentation for how it discovers configuration or instruction files. Default to symlink if the framework reads arbitrary directories.
Step 3: Implement Detection
Detection tells the CLI which frameworks are present in a project:
// In detector.js — each rule checks for a filesystem marker
const RULES = [
{
id: 'my-framework',
displayName: 'My Framework',
check: (dir) => existsSync(resolve(dir, '.myframework/')),
marker: '.myframework/',
scope: 'project',
},
];
Detection strategies:
- Directory presence:
.claude/,.cursor/,.gemini/ - Config file:
opencode.json,.aider.conf.yml - Instruction file:
AGENTS.md,CONVENTIONS.md - Global markers:
~/.openclaw/,~/.hermes/
Always return the marker in the detection result so users can understand why a framework was detected.
Got: A detection rule that reliably identifies the framework without false positives.
If fail: If the framework has no unique marker (generic directory name), use a combination of markers or require explicit --framework specification.
Step 4: Implement Install with Idempotency
async install(item, projectDir, scope, options) {
const targetDir = this.getTargetPath(projectDir, scope);
const targetPath = resolve(targetDir, item.id);
// Idempotency: skip if already installed (unless force)
if (existsSync(targetPath) && !options.force) {
return { action: 'skipped', path: targetPath };
}
if (options.dryRun) {
return { action: 'created', path: targetPath, details: 'dry-run' };
}
// Ensure parent directory exists
mkdirSync(targetDir, { recursive: true });
// Strategy-specific installation
if (this.constructor.strategy === 'symlink') {
const relPath = relative(targetDir, item.sourceDir);
symlinkSync(relPath, targetPath);
} else if (this.constructor.strategy === 'copy') {
cpSync(item.sourceDir, targetPath, { recursive: true });
}
return { action: 'created', path: targetPath };
}
Idempotency rules:
- Skip if target exists and
--forceis not set - Overwrite if
--forceis set (remove first, then install) - Dry-run always succeeds with
action: 'created' - Return value must always be
{ action, path, details? }
Got: Install creates content at the target path, skips if already present, respects --force and --dry-run.
If fail: If symlink creation fails on Windows/NTFS, fall back to directory junction or copy. Log the fallback.
Step 5: Implement Uninstall with Cleanup
async uninstall(item, projectDir, scope, options) {
const targetDir = this.getTargetPath(projectDir, scope);
const targetPath = resolve(targetDir, item.id);
if (!existsSync(targetPath)) {
return { action: 'skipped', path: targetPath };
}
if (options.dryRun) {
return { action: 'removed', path: targetPath };
}
// Remove the installed content
rmSync(targetPath, { recursive: true });
return { action: 'removed', path: targetPath };
}
Cleanup considerations:
- Remove only what the plugin installed — never delete user-created files
- For append-to-file: remove the marked section, not the entire file
- Leave parent directories intact (other plugins may use them)
Got: Uninstall removes only the plugin's content and nothing else.
If fail: If removal fails (permissions, locked file), return an error result instead of throwing.
Step 6: Implement Listing and Audit
async listInstalled(projectDir, scope) {
const targetDir = this.getTargetPath(projectDir, scope);
if (!existsSync(targetDir)) return [];
const entries = readdirSync(targetDir);
return entries.map(name => {
const fullPath = resolve(targetDir, name);
const broken = lstatSync(fullPath).isSymbolicLink()
&& !existsSync(fullPath);
return { id: name, type: 'skill', broken };
});
}
async audit(projectDir, scope) {
const items = await this.listInstalled(projectDir, scope);
const ok = items.filter(i => !i.broken);
const broken = items.filter(i => i.broken);
return {
framework: this.constructor.displayName,
ok: [`${ok.length} skills installed`],
warnings: [],
errors: broken.map(i => `Broken: ${i.id}`),
};
}
Got: Listing returns all installed items with broken-link detection. Audit summarizes health.
If fail: If the target directory doesn't exist, return empty results (not an error — the framework has nothing installed).
Step 7: Register the Plugin
// In adapters/index.js
import { MyFrameworkAdapter } from './my-framework.js';
register(MyFrameworkAdapter);
Registration makes the adapter available to:
- Auto-detection (
detectFrameworks()→getAdaptersForDetections()) - Explicit selection (
--framework my-framework) - Listing (
listAdapters())
Got: The adapter appears in tool detect output and can be targeted with --framework.
If fail: If the adapter doesn't appear, verify static id matches the detection rule's id and that register() was called.
Step 8: Write Tests
describe('adapter: my-framework (dry-run)', () => {
it('targets the correct path', () => {
const out = run('install create-skill --framework my-framework --dry-run');
assert.match(out, /\.myframework/i);
});
});
Test at minimum: dry-run path, detection presence, and content type support.
Got: Adapter-specific tests confirm the installation path and behavior.
If fail: If the framework isn't detected in CI (no marker directory), use --framework explicitly in tests.
Validation
- Plugin extends the base class correctly
- Static fields (
id,displayName,strategy,contentTypes) are set - Detection rule identifies the framework without false positives
-
install()is idempotent (skip if exists, respect--force) -
uninstall()removes only plugin-created content -
listInstalled()detects broken symlinks -
audit()reports health accurately - Plugin is registered and appears in
tool detect - Dry-run tests pass
Pitfalls
- Forgetting relative vs. absolute symlinks: Project-scope symlinks should be relative (portable). Global-scope symlinks should be absolute (not dependent on cwd).
- Not handling missing parent directories: Always
mkdirSync(dir, { recursive: true })before creating content. - Append-to-file without markers: Without idempotent markers (
<!-- start:id -->/<!-- end:id -->), repeated installs duplicate content. Always wrap appended content. - Detection false positives: A generic directory name (e.g.,
.config/) may match multiple frameworks. Use specific file markers inside the directory. - Forgetting
supports()check: The installer callssupports(item.type)before dispatching. IfcontentTypesis wrong, the adapter silently skips items.
Related Skills
scaffold-cli-command— build the CLI commands that use this plugintest-cli-application— testing patterns for CLI tools including adapter testsdesign-cli-output— terminal output for install/uninstall results
GitHub 저장소
연관 스킬
content-collections
메타이 스킬은 콘텐츠 콜렉션(Content Collections)을 위한 프로덕션 검증된 설정을 제공합니다. 콘텐츠 콜렉션은 Markdown/MDX 파일을 Zod 검증이 포함된 타입 안전한 데이터 콜렉션으로 변환해주는 TypeScript 최우선 도구입니다. 블로그, 문서 사이트 또는 콘텐츠 중심의 Vite + React 애플리케이션을 구축할 때 타입 안전성과 자동 콘텐츠 검증을 보장하기 위해 사용하세요. Vite 플러그인 구성과 MDX 컴파일부터 배포 최적화 및 스키마 검증에 이르기까지 모든 것을 다룹니다.
polymarket
메타이 스킬은 개발자들이 Polymarket 예측 시장 플랫폼을 활용한 애플리케이션을 구축할 수 있도록 지원하며, 거래 및 시장 데이터를 위한 API 통합 기능을 포함합니다. 또한 WebSocket을 통한 실시간 데이터 스트리밍을 제공하여 실시간 거래와 시장 활동을 모니터링할 수 있습니다. 이를 통해 거래 전략을 구현하거나 실시간 시장 업데이트를 처리하는 도구를 생성하는 데 활용할 수 있습니다.
creating-opencode-plugins
메타이 스킬은 개발자들이 명령어, 파일, LSP 작업 등 25개 이상의 이벤트 유형에 연결되는 OpenCode 플러그인을 만들 수 있도록 돕습니다. JavaScript/TypeScript 모듈을 위한 플러그인 구조, 이벤트 API 명세, 구현 패턴을 제공합니다. OpenCode AI 어시스턴트의 라이프사이클을 사용자 정의 이벤트 기반 로직으로 가로채거나, 모니터링하거나, 확장해야 할 때 사용하세요.
sglang
메타SGLang은 RadixAttention 프리픽스 캐싱을 활용하여 JSON, 정규식, 에이전트 워크플로우를 위한 고속 구조화 생성에 특화된 고성능 LLM 서빙 프레임워크입니다. 특히 반복되는 프리픽스가 있는 작업에서 상당히 빠른 추론 속도를 제공하여 복잡한 구조화 출력 및 다중 턴 대화에 이상적입니다. 제약 디코딩이 필요하거나 광범위한 프리픽스 공유가 있는 애플리케이션을 구축할 때는 vLLM과 같은 대안보다 SGLang을 선택하십시오.
