agent-creator
Create Specialized AI Agents
Building AI agents with consistent, high-quality performance requires deep domain knowledge and optimized system prompts. This skill provides a structured 4-phase methodology to create production-ready agents with embedded expertise.
スキルZIPをダウンロード
Claudeでアップロード
設定 → 機能 → スキル → スキルをアップロードへ移動
オンにして利用開始
テストする
「agent-creator」を使用しています。 Create a code review agent for JavaScript projects
期待される結果:
- Agent Name: js-code-reviewer
- System Prompt: You are an expert JavaScript code reviewer...
- Constraints: [list]
- Workflow Phases: [description]
「agent-creator」を使用しています。 Build a research agent for medical literature
期待される結果:
- Agent Name: medical-lit-researcher
- System Prompt: You are a medical research specialist...
- Evidence Standards: [requirements]
- Citation Format: [specification]
「agent-creator」を使用しています。 Design an agent for customer onboarding
期待される結果:
- Agent Name: onboarding-guide
- System Prompt: You help new customers get started...
- Step Workflow: [process]
- Escalation Triggers: [conditions]
セキュリティ監査
低リスクStatic analysis flagged 81 potential issues in documentation and diagram files. All findings are false positives: DOT files contain GraphViz workflow diagrams, SKILL.md is documentation with no executable code, and metadata fields triggered false positives. The skill is a prompt/documentation skill for agent creation with no actual security risks.
リスク要因
🌐 ネットワークアクセス (1)
📁 ファイルシステムへのアクセス (1)
⚙️ 外部コマンド (30)
品質スコア
作れるもの
Build Domain-Specific Research Agents
Create AI agents optimized for specific research domains like legal document review, scientific literature analysis, or market research. The agent embeds domain-specific terminology, constraints, and workflow patterns.
Develop Code Review Automation Agents
Generate specialized agents for code review tasks with embedded coding standards, language-specific rules, and security scanning patterns. Agents follow consistent review workflows.
Create Customer Support Workflow Agents
Build agents trained on support scripts, knowledge base content, and escalation procedures. These agents handle common queries with consistent responses and proper handoff logic.
これらのプロンプトを試す
Create a new AI agent named [agent-name] for [task-description]. Use the 4-phase SOP methodology to define its system prompt with embedded domain knowledge about [domain-area].
Design an agent that can [primary-capability]. Include tools for [tool-list]. Define the agent constraints and response formats for [use-case].
Create an agent that follows this workflow: Phase 1 [phase-1-description], Phase 2 [phase-2-description], Phase 3 [phase-3-description]. Embed evidence-based prompting for quality outputs.
Generate a production-ready Claude Code agent with: name=[agent-name], purpose=[purpose], constraints=[constraints], escalation-rules=[rules]. Include error handling and validation prompts.
ベストプラクティス
- Define clear agent boundaries and constraints before generating the system prompt
- Use evidence-based prompting to embed verifiable domain knowledge
- Test agent responses with edge cases before deployment
回避
- Avoid creating agents without defined escalation paths for unknown inputs
- Do not skip the 4-phase methodology for complex agent requirements
- Avoid overloading agents with too many conflicting constraints