majiayu000 Agent Ecosystem Reference
Persistent reference for the epost_agent_kit Claude Code ecosystem. Use this when planning changes, creating new components, or auditing the agent system.
Claude Code Official Components
| Component | Location | Auto-discovered | Purpose |
|---|---|---|---|
| Agents | .claude/agents/ |
Yes | Autonomous subprocesses for complex tasks |
| Skills | .claude/skills/ |
Yes (nested) | Domain knowledge and workflows |
| Commands | .claude/commands/ |
Yes (nested) | User-triggered slash commands (merged into skills) |
| Hooks | .claude/settings.json |
Yes | Event-triggered actions (PreToolUse, PostToolUse, Stop, etc.) |
| Output Styles | .claude/output-styles/ |
Yes | Custom response formatting |
| MCP Servers | .mcp.json (project) / settings.json (user) |
Yes | External tool integrations |
| Plugins | .claude-plugin/plugin.json |
Via --plugin-dir |
Distributable component bundles |
NOT official: .claude/workflows/ — Custom reference documentation only, not auto-discovered by Claude Code.
Commands & Skills Merge
Commands and skills are functionally equivalent and share the same frontmatter fields. Both create /slash-command entries. Skills are the preferred format for new additions. Commands in .claude/commands/ continue to work.
Key difference: Skills use SKILL.md in a directory; commands are standalone .md files in commands/.
Nested Directory Support
Skills support nested directories. Claude Code auto-discovers SKILL.md files at any depth under .claude/skills/. All skills in this ecosystem use flat names (e.g., kit-agent-development/SKILL.md, web-nextjs/SKILL.md).
Agent Frontmatter Reference
| Field | Required | Source | Type | Description |
|---|---|---|---|---|
name |
Yes | Official | string | Identifier (lowercase, hyphens, 3-50 chars) |
description |
Yes | Official | string | Triggering conditions with <example> blocks |
model |
Yes | Official | string | inherit, sonnet, opus, haiku |
color |
Yes | Official | string | blue, cyan, green, yellow, magenta, red |
tools |
No | Official | array | Restrict available tools (default: all) |
skills |
No | Ecosystem | array | Preload skills into agent context at startup |
memory |
No | Ecosystem | string | Persistent learning: user, project, local |
permissionMode |
No | Ecosystem | string | default, acceptEdits, dontAsk, bypassPermissions, plan |
hooks |
No | Ecosystem | object | Per-agent scoped hooks (same format as settings.json hooks) |
disallowedTools |
No | Ecosystem | string | Comma-separated tools to deny (advisory) |
allowedTools |
No | Ecosystem | string | Comma-separated tools to allow |
Source legend: Official = confirmed in upstream anthropics/claude-code plugin-dev docs. Ecosystem = documented in our reference and observed working but not confirmed upstream.
Skill Frontmatter Reference
| Field | Required | Type | Description |
|---|---|---|---|
name |
Yes | string | Skill identifier |
description |
Yes | string | When to use (third-person with trigger phrases) |
context |
No | string | fork — run in isolated subagent |
agent |
No | string | Which subagent type for context: fork |
user-invocable |
No | boolean | false hides from / menu |
disable-model-invocation |
No | boolean | true prevents auto-trigger |
allowed-tools |
No | array | Restrict tools in forked context |
hooks |
No | object | Per-skill hooks |
model |
No | string | Override model for this skill |
argument-hint |
No | string | Placeholder text for skill argument |
epost-kit metadata extensions (nested under metadata:):
| Field | Type | Description |
|---|---|---|
keywords |
array | Search terms for skill discovery |
platforms |
array | Target platforms: ios, android, web, backend, cli, all |
triggers |
array | File patterns that trigger skill (e.g., .kt, .swift) |
agent-affinity |
array | Preferred agents for this skill |
connections.extends |
array | Parent skills this inherits from |
connections.requires |
array | Skills that must be co-loaded |
connections.conflicts |
array | Skills that cannot coexist |
connections.enhances |
array | Optional complementary skills |
Invalid fields: version (not recognized by Claude Code)
Directory Structure
.claude/
├── agents/ # 17 specialized agents
│ ├── epost-project-manager.md # Progress tracking & roadmaps
│ ├── epost-planner.md # Planning & research coordination
│ ├── epost-fullstack-developer.md # Fullstack development
│ ├── epost-code-reviewer.md # Quality assurance & security
│ ├── epost-debugger.md # Root cause debugging
│ ├── epost-tester.md # Test orchestration
│ ├── epost-researcher.md # Technology research
│ ├── epost-docs-manager.md # Auto-updating documentation
│ ├── epost-git-manager.md # Git workflow automation
│ ├── epost-brainstormer.md # Creative ideation
│ ├── epost-kit-designer.md # Kit authoring specialist
│ ├── epost-a11y-specialist.md # Multi-platform accessibility (WCAG 2.1 AA)
│ ├── epost-muji.md # MUJI UI library
│ ├── epost-code-simplifier.md # Code simplification
│ ├── epost-journal-writer.md # Development journals
│ ├── epost-mcp-manager.md # MCP integration
│ └── epost-ui-ux-designer.md # UI/UX design
├── skills/ # Domain knowledge (flat layout)
│ ├── kit-agents/ # Ecosystem hub (this file)
│ ├── kit-agent-development/ # Agent creation guide
│ ├── kit-skill-development/ # Skill creation guide
│ ├── kit-hooks/ # Hook authoring
│ ├── kit-cli/ # CLI tool patterns
│ ├── core/ # Operational rules
│ ├── data-store/ # Agent persistent data store convention
│ ├── cook/ # Build/implement features
│ ├── fix/ # Fix issues and errors
│ ├── plan/ # Planning workflow
│ ├── test/ # Test orchestration
│ ├── debug/ # Debug workflow
│ ├── scout/ # Codebase exploration
│ ├── review-code/ # Code review
│ ├── git-commit/ # Git commit workflow
│ ├── git-push/ # Git push workflow
│ ├── git-pr/ # Git PR creation
│ ├── web-nextjs/ # Next.js patterns
│ ├── web-frontend/ # React/frontend patterns
│ ├── figma/ # Figma MCP + extraction (all platforms)
│ ├── design-tokens/ # Vien 2.0 token architecture (all platforms)
│ ├── web-ui-lib/ # UI component library (web, in platform-web)
│ ├── ui-lib-dev/ # UI library pipeline + integration guidance (all platforms)
│ ├── web-a11y/ # Web accessibility (WCAG 2.1 AA)
│ ├── ios-development/ # Swift/SwiftUI patterns
│ ├── ios-a11y/ # iOS accessibility
│ ├── simulator/ # iOS simulator management
│ ├── android-development/ # Kotlin/Compose patterns
│ ├── android-a11y/ # Android accessibility
│ ├── backend-javaee/ # Jakarta EE patterns
│ ├── a11y/ # Cross-platform a11y foundation
│ ├── knowledge-retrieval/ # Knowledge lookup + base management
│ ├── knowledge-capture/ # Post-task knowledge capture
│ └── ... # (other cross-cutting skills)
├── workflows/ # Reference docs (NOT auto-discovered)
│ ├── feature-development.md
│ ├── bug-fixing.md
│ └── project-init.md
├── output-styles/ # Response formatting (5 styles)
└── settings.json # Hooks, permissions, env vars
Naming Conventions
| Component | Pattern | Example |
|---|---|---|
| Agent files | epost-<role>.md |
epost-planner.md |
| Commands | <category>/<action>.md |
web/cook.md |
| Skills | <name>/SKILL.md |
web-nextjs/SKILL.md |
| Kit skills | kit-<topic>/SKILL.md |
kit-agent-development/SKILL.md |
Description Patterns
| Component | Pattern |
|---|---|
| Agents | [Role] agent that [does X]. Use for [contexts]. |
| Skills | [Domain] [description]. Use when [contexts]. |
| Commands | [Imperative verb] [object/scope] |
Plugin System
Distributable via .claude-plugin/plugin.json:
{
"name": "plugin-name",
"description": "Plugin description",
"version": "1.0.0",
"commands": ["./.claude/commands/"],
"agents": ["./.claude/agents/"],
"skills": ["./.claude/skills/"]
}
Load with: claude --plugin-dir /path/to/plugin
Sub-Skills
kit-agent-development— Agent creation, frontmatter, system prompts, triggeringkit-skill-development— Skill creation, progressive disclosure, validation
Related Documents
CLAUDE.md— Project overview and guidelines.claude/skills/core/SKILL.md— Operational rules.claude/agents/— Agent definitions