create-ai-tool

AI Tool Creator

Analyze the user's requirements to determine Skill or Agent and delegate to the corresponding creation skill.

Judgment Criteria

Default is Skill. Only when the conditions below apply, Agent.

Condition	YES → Agent	NO → Skill
Need isolated execution in an independent context?	Large output, parallel processing	Sufficient in the main conversation
Need to restrict tool access?	Read-only, allow only specific tools	No restriction needed
Perform independent tasks via automatic delegation?	Claude judges and delegates the task	User/Claude calls directly

Decision Flow

[요청 분석]
    │
    ├─ 컨텍스트 격리 또는 도구 제한 필요?
    │       │
    │       ├─ YES → Agent → /create-agent
    │       │
    │       └─ NO
    │           │
    │           └─ Skill → /create-skill
    │
    └─ 규칙/컨벤션/가이드라인?
            │
            └─ Skill (user-invocable: false) → /create-skill

Key Differences

Category	Skill	Agent
Context	Shared main conversation	Separately isolated
Purpose	Knowledge/guidelines/workflow	Task delegation
Invocation	`/skill-name` or Claude automatic	Claude automatic delegation
Tool control	`allowed-tools`	`tools`, `disallowedTools`
Rules/conventions	`user-invocable: false`	Not applicable

Routing by use case

Use Case	Type	Key Settings	Delegation
Coding conventions, style guide	Skill	`user-invocable: false`	`/create-skill`
Step-by-step work instructions	Skill	Default	`/create-skill`
Deployment, DB operations (side effects)	Skill	`disable-model-invocation: true`	`/create-skill`
Code review, debugging (isolation)	Agent	`tools`, `disallowedTools`	`/create-agent`
Parallel research, large-scale analysis	Agent	`model`, `tools`	`/create-agent`

Execution

Once the type is determined, load the contents of the corresponding skill and proceed with creation:

Skill → create-skill Load
Agent → create-agent Load

Detailed Judgment

If complex judgment is needed, decision tree refer to.