my-plugin

---

name: my-plugin description: This skill should be used when the user asks to query, inspect, create, update, or delete My Plugin resources, or mentions My Plugin by name. argument-hint: [resource-id] [action] [--json] disable-model-invocation: false user-invocable: true allowed-tools: Bash(curl:*), mcp__my-plugin-mcp__my_plugin model: sonnet effort: high

context: fork

agent: my-plugin-analyzer

hooks: {}

paths:

- "my_service_mcp/**"

- "tests/**"

shell: bash

---

My Plugin Skill

Target path relative to plugin root: skills/my-plugin/SKILL.md

This template intentionally shows the full skill shape we expect plugin authors to fill in.

Frontmatter Reference

Plugin-safe fields shown in the live example:

• name
• description
• argument-hint
• disable-model-invocation
• user-invocable
• allowed-tools
• model
• effort
• context
• agent
• hooks
• paths
• shell

Broader notes:

• description is the most important field for model-driven invocation.
• context: fork and agent are paired fields when you want isolated execution.
• Keep optional fields commented out unless the plugin actually needs them.
• No additional Claude skill frontmatter exclusions are currently called out for plugin scope in the bundled docs used here.

Mode Detection

MCP mode: Use when the plugin tool is installed and available.

HTTP fallback mode: Use only when MCP tooling is unavailable.

Credential source:

• MCP URL from ${user_config.my_service_mcp_url}
• HTTP URL from $CLAUDE_PLUGIN_OPTION_MY_SERVICE_URL
• HTTP auth from $CLAUDE_PLUGIN_OPTION_MY_SERVICE_API_KEY

MCP Mode

Single tool:

• mcp__my-plugin-mcp__my_plugin

Available fields for the tool call

Common fields used by the action/subaction pattern:

• action
• subaction
• id
• name
• query
• page
• page_size
• sort
• order
• filters
• confirm

Full action reference

• list
• get
• create
• update
• delete

Example invocations

mcp__my-plugin-mcp__my_plugin
  action: "list"
  query: "active"
  page: 1
  page_size: 25
  sort: "created_at"
  order: "desc"

mcp__my-plugin-mcp__my_plugin
  action: "get"
  id: "resource-123"

mcp__my-plugin-mcp__my_plugin
  action: "create"
  name: "example"
  subaction: "validate"

mcp__my-plugin-mcp__my_plugin
  action: "update"
  id: "resource-123"
  subaction: "enable"

mcp__my-plugin-mcp__my_plugin
  action: "delete"
  id: "resource-123"
  confirm: true

HTTP Fallback

Use curl with MY_SERVICE_URL and MY_SERVICE_API_KEY.

# List resources
curl -s "$CLAUDE_PLUGIN_OPTION_MY_SERVICE_URL/api/v1/resources" \
  -H "X-Api-Key: $CLAUDE_PLUGIN_OPTION_MY_SERVICE_API_KEY"

# Get resource
curl -s "$CLAUDE_PLUGIN_OPTION_MY_SERVICE_URL/api/v1/resources/$RESOURCE_ID" \
  -H "X-Api-Key: $CLAUDE_PLUGIN_OPTION_MY_SERVICE_API_KEY"

# Create resource
curl -s -X POST "$CLAUDE_PLUGIN_OPTION_MY_SERVICE_URL/api/v1/resources" \
  -H "X-Api-Key: $CLAUDE_PLUGIN_OPTION_MY_SERVICE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"name":"example"}'

Response Shape

Preferred response contract:

• success
• action
• subaction
• data
• message
• errors
• timestamp

Destructive Operations

Always confirm before delete or irreversible update operations.