component（component 搜索）

Component: Claude Code Configuration Generator

Generate integrated Claude Code configuration with syntactically valid YAML frontmatter and semantically coherent cross-references.

Component Architecture

┌─────────────────────────────────────────────────────────────────────────────┐
│                          CLAUDE CODE CONFIGURATION                          │
├─────────────────────────────────────────────────────────────────────────────┤
│                                                                             │
│   COMMANDS ─────────invoke────────► SUBAGENTS                               │
│   (.claude/commands/*.md)           (.claude/agents/*.md)                   │
│   • Entry points                    • Specialized personas                  │
│   • Tool orchestration              • Persistent identity                   │
│   • User-facing /slash              │                                       │
│                                     ▼ load                                  │
│                                   SKILLS                                    │
│                                   (.claude/skills/*/SKILL.md)               │
│                                   • Procedural knowledge                    │
│                                   • Domain expertise                        │
│                                                                             │
│   STYLES ─────────modulate────────► SESSION BEHAVIOR                        │
│   (.claude/styles/*.md)             • Autonomy level                        │
│   • Interaction patterns            • Output format                         │
│                                     • Checkpoint frequency                  │
└─────────────────────────────────────────────────────────────────────────────┘

Relationships:

• Commands invoke Subagents (via Agent tool)
• Subagents load Skills (via skills: array)
• Styles modulate session behavior (parallel to commands, not hierarchical)

Parameter Matrix

Parameter Dependencies

allowed-tools + Agent ──────► enables subagent spawning
skills ──────────────────────► requires skill directories exist
permissionMode: allow ───────► full autonomy (no prompts)
permissionMode: ask ─────────► checkpoint prompts (default)
permissionMode: deny ────────► read-only operations

Component Specifications

Commands (.claude/commands/<name>.md)

Filename becomes /slash-command (no name field).

---
description: Execute [action] on [target]     # REQUIRED
allowed-tools: [Read, Write, Agent]           # Optional: tool access
model: sonnet                                 # Optional: sonnet|opus|haiku
argument-hint: <target>                       # Optional: CLI placeholder
---

## Implementation

[Procedural instructions for command execution]
[Can spawn subagents via Agent tool]

Example (/review):

---
description: Review code changes for quality and correctness
allowed-tools: [Read, Grep, Glob, Agent]
model: opus
argument-hint: <path>
---

## Execution

1. Analyze target path for code changes
2. Spawn reviewer agent with file context
3. Synthesize findings into actionable feedback

Subagents (.claude/agents/<name>.md)

Specialized personas with tool access and skill mastery.

---
description: [Role] specializing in [domain]. [Key traits].  # REQUIRED
allowed-tools: [Read, Write]                                  # Permitted tools
disallowed-tools: [Bash]                                      # Explicit denials
model: sonnet                                                 # Reasoning model
permissionMode: ask                                           # ask|allow|deny
skills: [skill-a, skill-b]                                    # Loaded procedures
---

## Persona

[Identity, decision heuristics, communication style]

Example (security-analyst.md):

---
description: Security analyst for vulnerability detection and threat modeling. Methodical, conservative.
allowed-tools: [Read, Grep, Glob]
disallowed-tools: [Write, Bash]
model: opus
permissionMode: ask
skills: [security-analysis, threat-modeling]
---

## Persona

Identify vulnerabilities with CVSS scoring. Prioritize by exploitability.
Flag sensitive data exposure. Never modify production code directly.

Skills (.claude/skills/<name>/SKILL.md)

Procedural knowledge packages loaded by subagents.

---
name: skill-name                              # REQUIRED: kebab-case, ≤64 chars
description: Procedures for [domain]. Trigger when [conditions].  # REQUIRED: ≤1024 chars
allowed-tools: [Read]                         # Optional: tool restrictions when active
---

## Procedures

[Step-by-step workflows, decision trees, domain knowledge]

Example (security-analysis/SKILL.md):

---
name: security-analysis
description: Static security analysis procedures. Trigger when reviewing code for vulnerabilities, secrets, or injection risks.
---

## Detection Patterns

### Secret Detection
- API keys: `[A-Za-z0-9]{32,}`
- AWS keys: `AKIA[A-Z0-9]{16}`
- Private keys: `-----BEGIN.*PRIVATE KEY-----`

### Injection Risks
- SQL: Unparameterized queries with string concatenation
- XSS: Unescaped user input in HTML templates
- Command: Shell execution with user-controlled arguments

Styles (.claude/styles/<name>.md)

Interaction mode modulators for session behavior.

---
name: style-name                              # REQUIRED
description: [Interaction pattern] for [use case].  # REQUIRED
keep-coding-instructions: true                # Optional: preserve coding behavior
---

## Behavior Modifications

[Output format, checkpoint frequency, autonomy level]

Supervision Gradient:

Tool Pattern Syntax

allowed-tools:
  # Basic tools
  - Read                      # File reading
  - Write                     # File writing
  - Edit                      # File editing
  - Glob                      # Pattern matching
  - Grep                      # Text search
  - LS                        # Directory listing
  - Agent                     # Subagent spawning
  - WebFetch                  # URL fetching
  
  # Bash patterns
  - Bash(git status)          # Exact command
  - Bash(git:*)               # All git subcommands
  - Bash(npm run:*)           # All npm run scripts
  
  # MCP tools
  - mcp__server__tool         # Specific MCP tool
  - mcp__github__*            # All tools from MCP server

Generation Workflow

┌─────────────────────────────────────────────────────────────────────────────┐
│ 1. ANALYZE                                                                  │
│    └─► Identify domain requirements, capabilities needed                    │
├─────────────────────────────────────────────────────────────────────────────┤
│ 2. DESIGN SKILLS                                                            │
│    └─► Package domain procedures (security-analysis, code-review, etc.)     │
├─────────────────────────────────────────────────────────────────────────────┤
│ 3. CREATE SUBAGENTS                                                         │
│    └─► Define personas that load skills, set tool access                    │
├─────────────────────────────────────────────────────────────────────────────┤
│ 4. BUILD COMMANDS                                                           │
│    └─► Create /slash entry points that invoke agents                        │
├─────────────────────────────────────────────────────────────────────────────┤
│ 5. CONFIGURE STYLES                                                         │
│    └─► Set supervision levels for different workflows                       │
├─────────────────────────────────────────────────────────────────────────────┤
│ 6. VALIDATE                                                                 │
│    └─► Check syntax, cross-references, parameter validity                   │
└─────────────────────────────────────────────────────────────────────────────┘

Integration Patterns

Pattern 1: Skill-Backed Agent

skill: code-analysis ──► agent: analyst ──► command: /analyze
                              │
                         skills: [code-analysis]

Agent loads procedural knowledge via skills, invoked by command.

Pattern 2: Multi-Agent Pipeline

command: /ship
    │
    ├─► Agent[planner] ───► Plan
    ├─► Agent[implementer] ► Code  
    ├─► Agent[reviewer] ──► Review
    └─► Agent[deployer] ──► Deploy

Sequential agent stages for complex workflows.

Pattern 3: Skill Composition

Agent loads multiple skills for comprehensive expertise:

# .claude/agents/fullstack.md
---
description: Full-stack developer proficient across domains
skills:
  - frontend-development
  - backend-development
  - security-guidelines
---

Pattern 4: Skill References Skill

Skills can reference procedures from other skills:

# .claude/skills/advanced-security/SKILL.md
---
name: advanced-security
description: Advanced threat modeling. Extends security-analysis procedures.
---

## Foundation

Load base procedures from `security-analysis` skill, then apply:

### Extended Analysis
- Supply chain attack vectors
- Dependency confusion risks
- CI/CD pipeline vulnerabilities

Validation Checklist

Commands

• [ ] No name field (filename is command name)
• [ ] Has description (imperative verb + target)
• [ ] Tool patterns valid
• [ ] No forbidden params: permissionMode, skills, keep-coding-instructions

Subagents

• [ ] Has description (role + domain + traits)
• [ ] skills: arrays reference existing skill directories
• [ ] permissionMode is ask, allow, or deny
• [ ] model is sonnet, opus, or haiku

Skills

• [ ] Has name (kebab-case, matches directory)
• [ ] Has description (max 1024 chars, includes trigger conditions)
• [ ] SKILL.md under 500 lines
• [ ] Directory contains SKILL.md at minimum

Styles

• [ ] Has name and description
• [ ] keep-coding-instructions is boolean if present
• [ ] No tool or model parameters

Cross-References

• [ ] All skills in skills: arrays exist as directories
• [ ] All agents referenced by commands exist
• [ ] No circular skill dependencies

Troubleshooting

Alternatives Considered

JSON Configuration: Single .claude.json file is simpler but lacks prose instructions and becomes unwieldy at scale. Markdown files support rich documentation alongside configuration.

Flat Structure: All components in one directory sacrifices organization but works for small projects. Current structure scales to dozens of components.

References

Detailed documentation by topic:

• references/syntax.md — Complete YAML frontmatter schemas, all parameters
• references/patterns.md — Integration patterns, multi-agent pipelines
• references/examples.md — Full plugin examples with all component types