surshar / claude-code

Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.

Built by Anthropic · github.com/anthropics

claude-code / plugins / plugin-dev / skills / plugin-structure / SKILL.md

Daisy S. HollmanMerge pull request #11780 from anthropics/daisy/plugin-dev/public-marketplace c508e59 · on Nov 18, 2025 History

476 lines · 13.5 KBPreview CodeRaw

1	---
2	name: Plugin Structure
3	description: This skill should be used when the user asks to "create a plugin", "scaffold a plugin", "understand plugin structure", "organize plugin components", "set up plugin.json", "use ${CLAUDE_PLUGIN_ROOT}", "add commands/agents/skills/hooks", "configure auto-discovery", or needs guidance on plugin directory layout, manifest configuration, component organization, file naming conventions, or Claude Code plugin architecture best practices.
4	version: 0.1.0
5	---
6
7	# Plugin Structure for Claude Code
8
9	## Overview
10
11	Claude Code plugins follow a standardized directory structure with automatic component discovery. Understanding this structure enables creating well-organized, maintainable plugins that integrate seamlessly with Claude Code.
12
13	Key concepts:
14	- Conventional directory layout for automatic discovery
15	- Manifest-driven configuration in `.claude-plugin/plugin.json`
16	- Component-based organization (commands, agents, skills, hooks)
17	- Portable path references using `${CLAUDE_PLUGIN_ROOT}`
18	- Explicit vs. auto-discovered component loading
19
20	## Directory Structure
21
22	Every Claude Code plugin follows this organizational pattern:
23
24	```
25	plugin-name/
26	├── .claude-plugin/
27	│ └── plugin.json # Required: Plugin manifest
28	├── commands/ # Slash commands (.md files)
29	├── agents/ # Subagent definitions (.md files)
30	├── skills/ # Agent skills (subdirectories)
31	│ └── skill-name/
32	│ └── SKILL.md # Required for each skill
33	├── hooks/
34	│ └── hooks.json # Event handler configuration
35	├── .mcp.json # MCP server definitions
36	└── scripts/ # Helper scripts and utilities
37	```
38
39	Critical rules:
40
41	1. Manifest location: The `plugin.json` manifest MUST be in `.claude-plugin/` directory
42	2. Component locations: All component directories (commands, agents, skills, hooks) MUST be at plugin root level, NOT nested inside `.claude-plugin/`
43	3. Optional components: Only create directories for components the plugin actually uses
44	4. Naming convention: Use kebab-case for all directory and file names
45
46	## Plugin Manifest (plugin.json)
47
48	The manifest defines plugin metadata and configuration. Located at `.claude-plugin/plugin.json`:
49
50	### Required Fields
51
52	```json
53	{
54	"name": "plugin-name"
55	}
56	```
57
58	Name requirements:
59	- Use kebab-case format (lowercase with hyphens)
60	- Must be unique across installed plugins
61	- No spaces or special characters
62	- Example: `code-review-assistant`, `test-runner`, `api-docs`
63
64	### Recommended Metadata
65
66	```json
67	{
68	"name": "plugin-name",
69	"version": "1.0.0",
70	"description": "Brief explanation of plugin purpose",
71	"author": {
72	"name": "Author Name",
73	"email": "author@example.com",
74	"url": "https://example.com"
75	},
76	"homepage": "https://docs.example.com",
77	"repository": "https://github.com/user/plugin-name",
78	"license": "MIT",
79	"keywords": ["testing", "automation", "ci-cd"]
80	}
81	```
82
83	Version format: Follow semantic versioning (MAJOR.MINOR.PATCH)
84	Keywords: Use for plugin discovery and categorization
85
86	### Component Path Configuration
87
88	Specify custom paths for components (supplements default directories):
89
90	```json
91	{
92	"name": "plugin-name",
93	"commands": "./custom-commands",
94	"agents": ["./agents", "./specialized-agents"],
95	"hooks": "./config/hooks.json",
96	"mcpServers": "./.mcp.json"
97	}
98	```
99
100	Important: Custom paths supplement defaults—they don't replace them. Components in both default directories and custom paths will load.
101
102	Path rules:
103	- Must be relative to plugin root
104	- Must start with `./`
105	- Cannot use absolute paths
106	- Support arrays for multiple locations
107
108	## Component Organization
109
110	### Commands
111
112	Location: `commands/` directory
113	Format: Markdown files with YAML frontmatter
114	Auto-discovery: All `.md` files in `commands/` load automatically
115
116	Example structure:
117	```
118	commands/
119	├── review.md # /review command
120	├── test.md # /test command
121	└── deploy.md # /deploy command
122	```
123
124	File format:
125	```markdown
126	---
127	name: command-name
128	description: Command description
129	---
130
131	Command implementation instructions...
132	```
133
134	Usage: Commands integrate as native slash commands in Claude Code
135
136	### Agents
137
138	Location: `agents/` directory
139	Format: Markdown files with YAML frontmatter
140	Auto-discovery: All `.md` files in `agents/` load automatically
141
142	Example structure:
143	```
144	agents/
145	├── code-reviewer.md
146	├── test-generator.md
147	└── refactorer.md
148	```
149
150	File format:
151	```markdown
152	---
153	description: Agent role and expertise
154	capabilities:
155	- Specific task 1
156	- Specific task 2
157	---
158
159	Detailed agent instructions and knowledge...
160	```
161
162	Usage: Users can invoke agents manually, or Claude Code selects them automatically based on task context
163
164	### Skills
165
166	Location: `skills/` directory with subdirectories per skill
167	Format: Each skill in its own directory with `SKILL.md` file
168	Auto-discovery: All `SKILL.md` files in skill subdirectories load automatically
169
170	Example structure:
171	```
172	skills/
173	├── api-testing/
174	│ ├── SKILL.md
175	│ ├── scripts/
176	│ │ └── test-runner.py
177	│ └── references/
178	│ └── api-spec.md
179	└── database-migrations/
180	├── SKILL.md
181	└── examples/
182	└── migration-template.sql
183	```
184
185	SKILL.md format:
186	```markdown
187	---
188	name: Skill Name
189	description: When to use this skill
190	version: 1.0.0
191	---
192
193	Skill instructions and guidance...
194	```
195
196	Supporting files: Skills can include scripts, references, examples, or assets in subdirectories
197
198	Usage: Claude Code autonomously activates skills based on task context matching the description
199
200	### Hooks
201
202	Location: `hooks/hooks.json` or inline in `plugin.json`
203	Format: JSON configuration defining event handlers
204	Registration: Hooks register automatically when plugin enables
205
206	Example structure:
207	```
208	hooks/
209	├── hooks.json # Hook configuration
210	└── scripts/
211	├── validate.sh # Hook script
212	└── check-style.sh # Hook script
213	```
214
215	Configuration format:
216	```json
217	{
218	"PreToolUse": [{
219	"matcher": "Write\|Edit",
220	"hooks": [{
221	"type": "command",
222	"command": "bash ${CLAUDE_PLUGIN_ROOT}/hooks/scripts/validate.sh",
223	"timeout": 30
224	}]
225	}]
226	}
227	```
228
229	Available events: PreToolUse, PostToolUse, Stop, SubagentStop, SessionStart, SessionEnd, UserPromptSubmit, PreCompact, Notification
230
231	Usage: Hooks execute automatically in response to Claude Code events
232
233	### MCP Servers
234
235	Location: `.mcp.json` at plugin root or inline in `plugin.json`
236	Format: JSON configuration for MCP server definitions
237	Auto-start: Servers start automatically when plugin enables
238
239	Example format:
240	```json
241	{
242	"mcpServers": {
243	"server-name": {
244	"command": "node",
245	"args": ["${CLAUDE_PLUGIN_ROOT}/servers/server.js"],
246	"env": {
247	"API_KEY": "${API_KEY}"
248	}
249	}
250	}
251	}
252	```
253
254	Usage: MCP servers integrate seamlessly with Claude Code's tool system
255
256	## Portable Path References
257
258	### ${CLAUDE_PLUGIN_ROOT}
259
260	Use `${CLAUDE_PLUGIN_ROOT}` environment variable for all intra-plugin path references:
261
262	```json
263	{
264	"command": "bash ${CLAUDE_PLUGIN_ROOT}/scripts/run.sh"
265	}
266	```
267
268	Why it matters: Plugins install in different locations depending on:
269	- User installation method (marketplace, local, npm)
270	- Operating system conventions
271	- User preferences
272
273	Where to use it:
274	- Hook command paths
275	- MCP server command arguments
276	- Script execution references
277	- Resource file paths
278
279	Never use:
280	- Hardcoded absolute paths (`/Users/name/plugins/...`)
281	- Relative paths from working directory (`./scripts/...` in commands)
282	- Home directory shortcuts (`~/plugins/...`)
283
284	### Path Resolution Rules
285
286	In manifest JSON fields (hooks, MCP servers):
287	```json
288	"command": "${CLAUDE_PLUGIN_ROOT}/scripts/tool.sh"
289	```
290
291	In component files (commands, agents, skills):
292	```markdown
293	Reference scripts at: ${CLAUDE_PLUGIN_ROOT}/scripts/helper.py
294	```
295
296	In executed scripts:
297	```bash
298	#!/bin/bash
299	# ${CLAUDE_PLUGIN_ROOT} available as environment variable
300	source "${CLAUDE_PLUGIN_ROOT}/lib/common.sh"
301	```
302
303	## File Naming Conventions
304
305	### Component Files
306
307	Commands: Use kebab-case `.md` files
308	- `code-review.md` → `/code-review`
309	- `run-tests.md` → `/run-tests`
310	- `api-docs.md` → `/api-docs`
311
312	Agents: Use kebab-case `.md` files describing role
313	- `test-generator.md`
314	- `code-reviewer.md`
315	- `performance-analyzer.md`
316
317	Skills: Use kebab-case directory names
318	- `api-testing/`
319	- `database-migrations/`
320	- `error-handling/`
321
322	### Supporting Files
323
324	Scripts: Use descriptive kebab-case names with appropriate extensions
325	- `validate-input.sh`
326	- `generate-report.py`
327	- `process-data.js`
328
329	Documentation: Use kebab-case markdown files
330	- `api-reference.md`
331	- `migration-guide.md`
332	- `best-practices.md`
333
334	Configuration: Use standard names
335	- `hooks.json`
336	- `.mcp.json`
337	- `plugin.json`
338
339	## Auto-Discovery Mechanism
340
341	Claude Code automatically discovers and loads components:
342
343	1. Plugin manifest: Reads `.claude-plugin/plugin.json` when plugin enables
344	2. Commands: Scans `commands/` directory for `.md` files
345	3. Agents: Scans `agents/` directory for `.md` files
346	4. Skills: Scans `skills/` for subdirectories containing `SKILL.md`
347	5. Hooks: Loads configuration from `hooks/hooks.json` or manifest
348	6. MCP servers: Loads configuration from `.mcp.json` or manifest
349
350	Discovery timing:
351	- Plugin installation: Components register with Claude Code
352	- Plugin enable: Components become available for use
353	- No restart required: Changes take effect on next Claude Code session
354
355	Override behavior: Custom paths in `plugin.json` supplement (not replace) default directories
356
357	## Best Practices
358
359	### Organization
360
361	1. Logical grouping: Group related components together
362	- Put test-related commands, agents, and skills together
363	- Create subdirectories in `scripts/` for different purposes
364
365	2. Minimal manifest: Keep `plugin.json` lean
366	- Only specify custom paths when necessary
367	- Rely on auto-discovery for standard layouts
368	- Use inline configuration only for simple cases
369
370	3. Documentation: Include README files
371	- Plugin root: Overall purpose and usage
372	- Component directories: Specific guidance
373	- Script directories: Usage and requirements
374
375	### Naming
376
377	1. Consistency: Use consistent naming across components
378	- If command is `test-runner`, name related agent `test-runner-agent`
379	- Match skill directory names to their purpose
380
381	2. Clarity: Use descriptive names that indicate purpose
382	- Good: `api-integration-testing/`, `code-quality-checker.md`
383	- Avoid: `utils/`, `misc.md`, `temp.sh`
384
385	3. Length: Balance brevity with clarity
386	- Commands: 2-3 words (`review-pr`, `run-ci`)
387	- Agents: Describe role clearly (`code-reviewer`, `test-generator`)
388	- Skills: Topic-focused (`error-handling`, `api-design`)
389
390	### Portability
391
392	1. Always use ${CLAUDE_PLUGIN_ROOT}: Never hardcode paths
393	2. Test on multiple systems: Verify on macOS, Linux, Windows
394	3. Document dependencies: List required tools and versions
395	4. Avoid system-specific features: Use portable bash/Python constructs
396
397	### Maintenance
398
399	1. Version consistently: Update version in plugin.json for releases
400	2. Deprecate gracefully: Mark old components clearly before removal
401	3. Document breaking changes: Note changes affecting existing users
402	4. Test thoroughly: Verify all components work after changes
403
404	## Common Patterns
405
406	### Minimal Plugin
407
408	Single command with no dependencies:
409	```
410	my-plugin/
411	├── .claude-plugin/
412	│ └── plugin.json # Just name field
413	└── commands/
414	└── hello.md # Single command
415	```
416
417	### Full-Featured Plugin
418
419	Complete plugin with all component types:
420	```
421	my-plugin/
422	├── .claude-plugin/
423	│ └── plugin.json
424	├── commands/ # User-facing commands
425	├── agents/ # Specialized subagents
426	├── skills/ # Auto-activating skills
427	├── hooks/ # Event handlers
428	│ ├── hooks.json
429	│ └── scripts/
430	├── .mcp.json # External integrations
431	└── scripts/ # Shared utilities
432	```
433
434	### Skill-Focused Plugin
435
436	Plugin providing only skills:
437	```
438	my-plugin/
439	├── .claude-plugin/
440	│ └── plugin.json
441	└── skills/
442	├── skill-one/
443	│ └── SKILL.md
444	└── skill-two/
445	└── SKILL.md
446	```
447
448	## Troubleshooting
449
450	Component not loading:
451	- Verify file is in correct directory with correct extension
452	- Check YAML frontmatter syntax (commands, agents, skills)
453	- Ensure skill has `SKILL.md` (not `README.md` or other name)
454	- Confirm plugin is enabled in Claude Code settings
455
456	Path resolution errors:
457	- Replace all hardcoded paths with `${CLAUDE_PLUGIN_ROOT}`
458	- Verify paths are relative and start with `./` in manifest
459	- Check that referenced files exist at specified paths
460	- Test with `echo $CLAUDE_PLUGIN_ROOT` in hook scripts
461
462	Auto-discovery not working:
463	- Confirm directories are at plugin root (not in `.claude-plugin/`)
464	- Check file naming follows conventions (kebab-case, correct extensions)
465	- Verify custom paths in manifest are correct
466	- Restart Claude Code to reload plugin configuration
467
468	Conflicts between plugins:
469	- Use unique, descriptive component names
470	- Namespace commands with plugin name if needed
471	- Document potential conflicts in plugin README
472	- Consider command prefixes for related functionality
473
474	---
475
476	For detailed examples and advanced patterns, see files in `references/` and `examples/` directories.