agentmd-creator

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

AgentMD Creator

AgentMD 生成工具

Create AI agent configuration files through guided briefing process for general-purpose and business-domain agents.
通过引导式简报流程,为通用型及业务领域AI Agent创建配置文件。

Workflow

工作流程

6-step process: Brief → Load References → Generate → Customize → Apply Best Practices → Present
6步流程:简报 → 加载参考资料 → 生成配置 → 自定义调整 → 应用最佳实践 → 交付与迭代

Step 1: Briefing Questions

步骤1:简报问题

Ask user these questions ONE AT A TIME (not all at once). Adapt based on answers.
1. Agent Purpose (REQUIRED)
"Jaki jest główny cel tego agenta? Co ma robić?"
Examples: Research assistant, Sales support, Customer service, Content creator, HR coordinator
If unclear: Provide examples and ask for specific scenarios If too broad: Help narrow down to primary function
2. Domain/Industry (can infer from purpose)
"W jakiej dziedzinie lub branży będzie pracował?"
Examples: General purpose, Sales/CRM, Marketing, Finance, HR, Legal, Healthcare, Real Estate
If general purpose: Can skip domain-specific questions If unclear: Suggest domain based on purpose
3. IDE/Platform (REQUIRED)
"Jakiego narzędzia używasz lub planujesz używać?"
Options:
  • Claude Code → Generate
    CLAUDE.md
  • Cursor → Generate
    .cursor/rules/*.mdc
  • Windsurf → Generate
    .windsurf/rules/*.md
  • Antigravity → Generate
    .antigravity/rules.md
  • JetBrains Junie → Generate
    .junie/guidelines.md
  • GitHub Copilot → Generate
    .github/copilot-instructions.md
  • Multiple/Universal → Generate
    AGENTS.md
If "don't know": Recommend
AGENTS.md
(universal format) If multiple: Generate
AGENTS.md
as base, offer platform-specific additions
4. Specificity Level (default: Moderate if unsure)
"Jak szczegółowy ma być agent?"
Options:
  • General (50-100 lines): Basic role, boundaries, key references
  • Moderate (100-200 lines): Detailed processes, tools, communication style
  • Detailed (200-300 lines): Comprehensive workflows, templates, edge cases
If unsure: Default to Moderate (good balance) Guidance: "Start with Moderate, can always expand later"
5. Special Requirements (OPTIONAL)
"Czy są jakieś specjalne wymagania? (compliance, security, specific tools, workflows)"
If none: Proceed with defaults from domain If many: Prioritize top 2-3 most critical
请逐个向用户提问(不要一次性全部抛出),并根据用户的回答调整后续问题。
1. Agent核心目标(必填)
“这个Agent的主要目标是什么?它要完成哪些工作?”
示例:研究助手、销售支持、客户服务、内容创作、HR协调员
若用户表述模糊: 提供示例并请用户说明具体场景 若用户表述过于宽泛: 协助用户聚焦到核心功能
2. 所属领域/行业(可从目标推断)
“它将在哪个领域或行业中工作?”
示例:通用型、销售/CRM、市场营销、金融、人力资源、法律、医疗健康、房地产
若为通用型: 可跳过领域相关的特定问题 若用户表述模糊: 根据Agent目标推荐对应领域
3. 开发环境/平台(必填)
“你正在使用或计划使用什么工具?”
可选选项:
  • Claude Code → 生成
    CLAUDE.md
  • Cursor → 生成
    .cursor/rules/*.mdc
  • Windsurf → 生成
    .windsurf/rules/*.md
  • Antigravity → 生成
    .antigravity/rules.md
  • JetBrains Junie → 生成
    .junie/guidelines.md
  • GitHub Copilot → 生成
    .github/copilot-instructions.md
  • 多平台/通用 → 生成
    AGENTS.md
若用户表示“不清楚”: 推荐使用
AGENTS.md
(通用格式) 若用户选择多个平台:
AGENTS.md
为基础,同时提供各平台的专属配置补充
4. 配置详细程度(默认:中等,若用户未明确说明)
“你希望配置文件的详细程度如何?”
可选选项:
  • 通用版(50-100行):基础角色定义、边界规则、核心工具参考
  • 中等版(100-200行):详细流程说明、工具清单、沟通风格规范
  • 详细版(200-300行):完整工作流、模板示例、边缘场景处理
若用户未明确说明: 默认使用中等版(平衡实用性与简洁性) 引导建议: “可以先从中等版开始,后续再根据需要补充细节”
5. 特殊需求(可选)
“是否有特殊要求?(如合规性、安全性、特定工具集成、专属工作流等)”
若无特殊需求: 按对应领域的默认规则生成 若需求较多: 优先处理最关键的2-3项需求

Step 2: Load Appropriate References

步骤2:加载对应参考资料

Based on user's answers, read relevant reference files:
Always read:
  • references/best-practices.md
    - Universal best practices
  • references/platforms.md
    - Platform-specific guidance for chosen IDE
Read based on domain:
  • references/use-cases.md
    - Load section matching user's domain/purpose
根据用户的回答,读取相关参考文件:
必须读取:
  • references/best-practices.md
    - 通用最佳实践指南
  • references/platforms.md
    - 所选开发环境的专属配置规范
按领域读取:
  • references/use-cases.md
    - 加载与用户所选领域/目标匹配的章节

Step 3: Generate Configuration File

步骤3:生成配置文件

Create configuration file following this structure:
按照以下结构创建配置文件:

For AGENTS.md / CLAUDE.md (Universal Format)

通用格式(AGENTS.md / CLAUDE.md)

markdown
undefined
markdown
undefined

[Project/Agent Name]

[项目/Agent名称]

[One-line description of agent's purpose]
[Agent核心目标的一句话描述]

Role

角色定义

[Specific role definition with domain expertise]
[包含领域专业能力的具体角色说明]

Tools and Systems

工具与系统

[List of tools, platforms, credentials locations]
[工具清单、平台名称、凭证存放位置]

Knowledge Base

知识库

[Locations of documentation, templates, policies]
[文档、模板、政策文件的存放位置]

Communication Style

沟通风格

[Tone, format, timing preferences]
[语气、格式、响应时机的偏好要求]

Workflows

工作流

[Key processes this agent handles]
[Agent负责的核心业务流程]

Important Notes

重要提示

[Domain-specific gotchas, warnings, critical information]
[领域专属注意事项、警告信息、关键规则]

Boundaries

边界规则

Always Do

必须执行

[Non-negotiable actions and practices]
[不可妥协的操作与规范]

Ask First

需先请示

[Actions requiring human approval or verification]
[需要人工批准或验证的操作]

Never Do

禁止执行

[Forbidden actions, compliance requirements, safety measures]
undefined
[违规操作、合规要求、安全措施]
undefined

For Cursor (.mdc format)

Cursor平台(.mdc格式)

Create modular files in
.cursor/rules/
:
markdown
---
name: "core-agent-role"
description: "Core agent behaviors and boundaries"
alwaysApply: true
---
.cursor/rules/
目录下创建模块化文件:
markdown
---
name: "core-agent-role"
description: "Core agent behaviors and boundaries"
alwaysApply: true
---

[Agent Role]

[Agent角色定义]

[Instructions]

Split into multiple files:
- `001-core-role.mdc` - Role and boundaries
- `100-workflows.mdc` - Process workflows
- `200-domain-knowledge.mdc` - Domain-specific knowledge
[具体指令]

建议拆分为多个文件:
- `001-core-role.mdc` - 角色定义与边界规则
- `100-workflows.mdc` - 业务流程说明
- `200-domain-knowledge.mdc` - 领域专属知识

For Windsurf

Windsurf平台

markdown
undefined
markdown
undefined

[Agent Name] Rules

[Agent名称] 规则

Role

角色定义

[Definition]
[具体说明]

Workflows

工作流

[Processes]
[业务流程]

Boundaries

边界规则

[Always/Ask/Never structure]

Keep under 6000 characters per file.
[必须执行/需先请示/禁止执行 三级结构]

单个文件内容不得超过6000字符。

For Platform-Specific

其他平台专属格式

Follow platform guidelines from
references/platforms.md
遵循
references/platforms.md
中的平台规范。

Step 4: Customize Based on Specificity and Domain

步骤4:根据详细程度与领域自定义配置

Adapt generated configuration to match both specificity level and domain requirements.
A. Apply Specificity Level:
General (50-100 lines):
  • Role definition
  • Key boundaries (Always/Never only)
  • Critical tools/systems
  • 2-3 most important workflows
Moderate (100-200 lines):
  • Detailed role with context
  • Full boundaries (Always/Ask/Never)
  • All relevant tools and systems
  • Communication style
  • Main workflows with steps
  • Important notes
Detailed (200-300 lines):
  • Comprehensive role with examples
  • Detailed boundaries with reasoning
  • Complete tool ecosystem
  • Communication style with templates
  • All workflows with decision trees
  • Knowledge base organization
  • Edge cases and exceptions
  • Compliance requirements
B. Add Domain-Specific Elements:
Based on chosen domain, enhance with specific requirements. Common domains:
  • Sales/CRM: Lead qualification, CRM fields, follow-up sequences
  • Customer Support: SLA targets, escalation paths, response templates
  • HR: Confidentiality, compliance standards, approval workflows
  • Finance: Data accuracy, approval thresholds, audit trails
  • Marketing: Brand guidelines, content calendar, campaign tracking
  • Legal: Document confidentiality, compliance checklists, attorney escalation
Reference: See
references/use-cases.md
for detailed patterns and complete guidance for each domain.
根据用户选择的详细程度和所属领域,调整生成的配置文件。
A. 匹配详细程度要求:
通用版(50-100行):
  • 角色定义
  • 核心边界规则(仅“必须执行”和“禁止执行”)
  • 关键工具/系统清单
  • 2-3个核心工作流
中等版(100-200行):
  • 带场景背景的详细角色定义
  • 完整边界规则(三级结构)
  • 所有相关工具与系统
  • 沟通风格规范
  • 带步骤说明的核心工作流
  • 重要提示
详细版(200-300行):
  • 包含示例的全面角色定义
  • 带理由说明的详细边界规则
  • 完整工具生态系统
  • 带模板的沟通风格规范
  • 带决策树的全量工作流
  • 知识库组织方式
  • 边缘场景与异常处理
  • 合规要求
B. 添加领域专属内容:
根据所选领域,补充对应的专属规则。常见领域的配置要点:
  • 销售/CRM: 线索资格审核、CRM字段规范、跟进流程
  • 客户支持: SLA指标、升级路径、响应模板
  • 人力资源: 保密要求、合规标准、审批流程
  • 金融: 数据准确性要求、审批阈值、审计追踪
  • 市场营销: 品牌规范、内容日历、 campaign 追踪
  • 法律: 文档保密规则、合规检查清单、律师升级流程
参考: 详见
references/use-cases.md
中各领域的详细模板与指导。

Step 5: Apply Best Practices

步骤5:应用最佳实践

Review generated configuration against quality criteria from
references/best-practices.md
.
Quality Review Checklist:
  1. Length Check:
    • General: 50-100 lines? ✓
    • Moderate: 100-200 lines? ✓
    • Detailed: 200-300 lines? ✓
  2. Boundaries Complete:
    • "Always Do" section present? ✓
    • "Ask First" section present? ✓
    • "Never Do" section present? ✓
    • At least 3 items in each? ✓
  3. Tools Specific:
    • Platform names included? (e.g., "Salesforce" not "CRM")
    • Credential locations specified? (e.g., "1Password: Sales CRM")
    • File paths provided? (e.g., /templates/sales-emails/)
  4. Content Quality:
    • No generic advice? ("be professional" → specify what professional means)
    • Examples concrete? (specific scenarios, not abstract descriptions)
    • References used? (links to docs, not copied content)
  5. Domain Appropriateness:
    • Domain-specific elements included?
    • Communication style matches domain?
    • Compliance requirements covered (if applicable)?
If any check fails: Revise configuration before presenting to user.
Reference: See
references/best-practices.md
for detailed criteria and examples.
对照
references/best-practices.md
中的质量标准,审核生成的配置文件。
质量审核清单:
  1. 篇幅检查:
    • 通用版:50-100行? ✓
    • 中等版:100-200行? ✓
    • 详细版:200-300行? ✓
  2. 边界规则完整性:
    • 包含“必须执行”章节? ✓
    • 包含“需先请示”章节? ✓
    • 包含“禁止执行”章节? ✓
    • 每个章节至少包含3条规则? ✓
  3. 工具说明明确性:
    • 是否包含具体平台名称?(例如:写“Salesforce”而非“CRM”)
    • 是否指定凭证存放位置?(例如:“1Password:销售CRM凭证”)
    • 是否提供文件路径?(例如:/templates/sales-emails/)
  4. 内容质量:
    • 无泛泛而谈的建议?(将“保持专业”改为明确的专业行为定义)
    • 示例是否具体?(使用真实场景而非抽象描述)
    • 是否引用外部文档?(提供链接而非复制内容)
  5. 领域适配性:
    • 是否包含领域专属规则?
    • 沟通风格是否匹配领域要求?
    • 是否覆盖合规要求(若适用)?
若任意一项未通过: 在交付给用户前修改配置文件。
参考: 详见
references/best-practices.md
中的详细标准与示例。

Step 6: Present and Iterate

步骤6:交付与迭代

  1. Show generated configuration to user
  2. Explain key sections and their purpose
  3. Ask: "Czy chcesz jakieś zmiany lub doprecyzowania?"
  4. Iterate based on feedback
  5. Save to appropriate file location
  1. 向用户展示生成的配置文件
  2. 解释核心章节的作用
  3. 询问用户:“您想要进行一些修改或细化吗?”
  4. 根据用户反馈调整配置
  5. 将配置文件保存到对应位置

Output Format

输出格式规范

File Location:
  • AGENTS.md → Root of project
  • CLAUDE.md → Root or
    .claude/
  • .cursor/rules/*.mdc →
    .cursor/rules/
    directory
  • .windsurf/rules/*.md →
    .windsurf/rules/
    directory
  • .antigravity/rules.md →
    .antigravity/
    directory
  • .junie/guidelines.md →
    .junie/
    directory
  • .github/copilot-instructions.md →
    .github/
    directory
Content:
  • Well-structured markdown
  • Clear section headings
  • Specific, actionable guidance
  • Domain-appropriate examples
  • Complete boundaries section
文件存放位置:
  • AGENTS.md → 项目根目录
  • CLAUDE.md → 项目根目录或
    .claude/
    目录
  • .cursor/rules/*.mdc →
    .cursor/rules/
    目录
  • .windsurf/rules/*.md →
    .windsurf/rules/
    目录
  • .antigravity/rules.md →
    .antigravity/
    目录
  • .junie/guidelines.md →
    .junie/
    目录
  • .github/copilot-instructions.md →
    .github/
    目录
内容要求:
  • 结构清晰的Markdown格式
  • 明确的章节标题
  • 具体、可执行的指导规则
  • 符合领域需求的示例
  • 完整的边界规则章节

Examples of Good Agent Configurations

优秀Agent配置示例

See
references/examples.md
for complete, detailed examples including:
Available examples:
  • Research Assistant (General, ~80 lines) - Systematic research with source credibility
  • Sales Assistant (Moderate, ~150 lines) - CRM coordination with MEDDIC methodology
  • Customer Support (Moderate, ~130 lines) - Ticket handling with escalation paths
  • HR Coordinator (Detailed, ~200 lines) - Recruiting, onboarding, compliance
Quick preview - Research Assistant structure:
markdown
undefined
详见
references/examples.md
中的完整示例,包括:
可用示例:
  • 研究助手(通用版,约80行):系统化研究与来源可信度验证
  • 销售助手(中等版,约150行):CRM协作与MEDDIC方法论落地
  • 客户支持(中等版,约130行):工单处理与升级路径
  • HR协调员(详细版,约200行):招聘、入职与合规管理
研究助手示例结构预览:
markdown
undefined

Research Assistant

Research Assistant

[Role definition]
[角色定义]

Tools and Systems

Tools and Systems

[Specific tools with locations]
[带具体位置的工具清单]

Boundaries

Boundaries

Always Do / Ask First / Never Do

Always Do / Ask First / Never Do

[3-tier boundary structure]

**Use these examples to:**
- Match specificity level to user's choice
- Adapt structure to chosen domain
- Scale boundaries appropriately
- Format tools section correctly

See full examples in `references/examples.md`
[三级边界规则结构]

**使用示例的建议:**
- 匹配用户选择的详细程度
- 适配用户所选领域调整结构
- 合理规划边界规则的数量
- 规范工具章节的格式

完整示例请查看`references/examples.md`

Tips for Success

成功技巧

  1. Start simple, iterate: Begin with general config, add detail based on real usage
  2. User-specific language: If user speaks Polish, use Polish in generated config
  3. Domain expertise: Leverage user's domain knowledge during briefing
  4. Real examples: Ask for actual scenarios to make config concrete
  5. Test and refine: Encourage user to test agent and report issues
  1. 从简到繁,逐步迭代: 先创建通用配置,再根据实际使用情况补充细节
  2. 适配用户语言: 如果用户使用波兰语,配置文件中也使用波兰语
  3. 借助领域专业知识: 在简报阶段充分利用用户的领域经验
  4. 使用真实场景: 请用户提供实际工作场景,让配置更贴合需求
  5. 测试与优化: 鼓励用户测试Agent并反馈问题,持续优化配置

Common Mistakes to Avoid

常见错误规避

  • Too generic: "Be helpful and professional" → Specify what professional means in this context
  • Too long: 500+ lines → Split into modules or remove redundancy
  • Missing boundaries: No clear Never section → Always include safety boundaries
  • Vague tools: "Use CRM" → "Use Salesforce, credentials in 1Password"
  • No examples: Abstract descriptions → Include concrete examples
  • One-size-fits-all: Same config for all domains → Customize per domain
  • 过于泛化: 避免“保持专业和有帮助”这类表述,需明确说明该场景下“专业”的具体定义
  • 篇幅过长: 避免500行以上的配置,可拆分为模块化文件或删除冗余内容
  • 缺失边界规则: 必须包含明确的“禁止执行”章节,设置安全边界
  • 工具描述模糊: 避免“使用CRM”这类表述,应写“使用Salesforce,凭证存放在1Password”
  • 缺乏示例: 避免抽象描述,应包含具体场景示例
  • 千篇一律: 不为所有领域使用相同配置,需根据领域特性自定义

Validation

最终验证

Before finalizing, verify:
  • Role clearly defined
  • Tools/systems specified with access info
  • Boundaries include all three tiers (Always/Ask/Never)
  • Domain-specific guidance included
  • Appropriate length for specificity level
  • User confirms it matches their needs
在定稿前,请确认以下内容:
  • 角色定义清晰明确
  • 工具/系统已指定具体名称与访问信息
  • 边界规则包含完整的三级结构(必须执行/需先请示/禁止执行)
  • 已添加领域专属指导规则
  • 篇幅符合所选的详细程度要求
  • 用户确认配置符合其需求