collect-agent-requirements

name	collect-agent-requirements
description	Collect, normalize, and validate the minimum actionable requirements needed to create a new agent through dialogue. This skill only gathers and structures requirements for downstream blueprint design, and must not generate repository files, code, or fake capabilities.
triggers	["创建agent","新建agent","生成agent","创建智能体","builder","builder agent"]
required_tools	[]
permissions	[]
input_schema	{"type":"object","required":["user_request"],"properties":{"user_request":{"type":"string","description":"用户当前提出的 agent 创建需求。"},"conversation_history":{"type":"array","description":"当前会话中与 agent 需求相关的历史对话，可用于补全上下文。","items":{"type":"object"}},"existing_requirements":{"type":"object","description":"之前轮次已提取出的需求字段，用于增量补全，而不是重复采集。","properties":{"agent_name":{"type":"string"},"agent_id":{"type":"string"},"role":{"type":"string"},"goal":{"type":"string"},"style_constraints":{"type":"array","items":{"type":"string"}},"required_skills":{"type":"array","items":{"type":"string"}},"required_tools":{"type":"array","items":{"type":"string"}},"user_constraints":{"type":"array","items":{"type":"string"}},"memory_requirements":{"type":"array","items":{"type":"string"}},"workflow_preferences":{"type":"array","items":{"type":"string"}}}}}}
output_schema	{"type":"object","required":["status","extracted_requirements","missing_fields","follow_up_questions"],"properties":{"status":{"type":"string","enum":["need_more_info","ready_for_blueprint","conflict_detected"]},"extracted_requirements":{"type":"object","description":"当前轮次归纳后的结构化需求，供 blueprint 设计阶段直接消费。","properties":{"agent_name":{"type":"string"},"agent_id":{"type":"string"},"role":{"type":"string"},"goal":{"type":"string"},"style_constraints":{"type":"array","items":{"type":"string"}},"required_skills":{"type":"array","items":{"type":"string"}},"required_tools":{"type":"array","items":{"type":"string"}},"user_constraints":{"type":"array","items":{"type":"string"}},"memory_requirements":{"type":"array","items":{"type":"string"}},"workflow_preferences":{"type":"array","items":{"type":"string"}}}},"missing_fields":{"type":"array","description":"为进入 blueprint 设计仍然缺失的关键字段。","items":{"type":"string"}},"follow_up_questions":{"type":"array","description":"当信息不足或冲突时，需要向用户继续追问的简洁问题。","items":{"type":"string"}},"conflicts":{"type":"array","description":"已检测到的冲突信息，例如 role 与 goal 不一致、工具约束互相矛盾。","items":{"type":"string"}},"collection_summary":{"type":"string","description":"面向后续 blueprint skill 的简洁总结。"}}}
decision_logic	[{"if":"无法提取 agent 的核心目标或角色定位","return":"need_more_info"},{"if":"已提取到 role、goal、required_skills、user_constraints 等最小建模信息，且不存在明显冲突","return":"ready_for_blueprint"},{"if":"用户描述中存在冲突信息，或关键约束彼此矛盾","return":"conflict_detected"}]
constraints	["只能提炼、归纳和补全需求，不直接生成 blueprint 文件、代码或仓库目录。","只输出 requirements 对象；pending requirements 由 runtime 负责保存。","【关键】`required_skills` 和 `required_tools` 必须从用户的**功能描述**中主动推断，不要等待用户提供这些技术细节。用户只需要说\"我想让它能生成题目\"，系统就应该自动识别需要 `question-generation` skill。","对于没有预定义匹配的功能，可以声明新的工具名称，让 blueprint 设计阶段来设计实现。","优先复用用户原话中的显式要求，只有在用户未说明时才可标记为缺失字段，不得擅自脑补。","追问必须简洁，优先一次性询问最关键的 1 到 3 个缺失点，避免连续碎片化提问。","输出的 extracted_requirements 必须是后续 skill 可直接消费的结构化对象，而不是松散文本。","agent_id 若已出现，需归一化为 snake_case；若尚未提供，可暂时留空，不得强行生成。","必须输出 user_message 字段：向用户说明当前收集进度、已确认的需求、或需要追问的问题，语言要自然、口语化，不要重复 JSON 结构。"]
failure_modes	[{"case":"user_request_ambiguous","effect":"无法确定 agent 的角色、目标或边界，不能进入 blueprint 设计"},{"case":"conflicting_requirements","effect":"需求存在冲突，继续建模会导致 blueprint 不稳定"},{"case":"insufficient_operational_details","effect":"可描述概念，但无法形成可执行需求对象"}]
fallback_strategy	[{"when":"user_request_ambiguous","action":"输出 missing_fields，并生成最少量 follow_up_questions 追问关键字段"},{"when":"conflicting_requirements","action":"输出 conflicts，并要求用户确认优先级或最终取舍"},{"when":"insufficient_operational_details","action":"引导用户补充技能、工具、约束、记忆或交互方式"}]
tool_policy	{"audit_logging":true,"sandbox_execution":false,"require_approval_for_write":false}
subagent_allowed	false
dependencies	[]
availability_checks	[]
enabled	true
metadata	{"category":"builder","stage":"requirements_collection","produces":"requirements_object","consumes":["user_request","conversation_history","existing_requirements"]}

Use this skill to collect and normalize the minimum requirements for creating a new agent.

Execution notes:

1. First extract whatever the user has already specified: role, goal, style, skills, tools, constraints, memory needs, and workflow preferences.
2. Merge the current turn with existing_requirements instead of rebuilding from scratch.
3. If key fields are missing, ask concise follow-up questions focused on the highest-value gaps only.
4. If information conflicts, do not guess; explicitly return conflict_detected and list the conflicts.
5. Do not generate files, code, tools, or blueprint artifacts in this step.
6. 【智能功能推断】不要等待用户提供 required_skills 和 required_tools，而是从用户的功能描述中主动推断：
   • 用户说"生成题目" → 自动添加 question-generation skill 和 question_generator tool
   • 用户说"评估回答" → 自动添加 response-evaluation skill 和 response_evaluator tool
   • 用户说"读取文件" → 自动添加 file-reading skill 和 read_local_file/list_directory tools
   • 用户说"导出报告" → 自动添加 report-export skill 和 report_exporter tool
   • 用户说"记住信息" → 自动添加 memory-persistence skill 和 persistent_memory_store tool
   • 如果用户描述的功能没有匹配的预定义 skill/tool，可以在 required_tools 中声明新工具，让后续的 blueprint 设计阶段来设计实现。
7. Produce a structured requirements object that the next blueprint-design skill can consume directly.
8. Do not ask the user to confirm creation in this step.
9. 接受用户的自然语言描述，从对话中提取需求信息。用户可能用随意的方式描述需求（如"我想创建一个面试助手"），你需要理解意图并提取结构化字段。
10. 如果 requirements 中包含 [待补充] 占位符，必须从用户对话中推断或追问来填充这些字段。[待补充] 表示系统无法自动提取的信息，需要你来补全。