Guided Conversation Skill
Purpose
Minimize user friction by asking up to 4 clarifying questions before executing actions. Ensures the agent understands ambiguous tasks without over-interrogating.
When to Use
- - User request is ambiguous or missing critical details
- Action is high-impact (file delete, payment, external API call)
- User has not provided enough context for safe execution
Constraints
- - Max 4 questions per task
- Stop asking once sufficient clarity is reached
- Never ask for information already available in context/memory
- If still ambiguous after 4 questions, make best-effort interpretation and proceed (with user notification)
Output
After clarification:
- 1. Summarize understood task in 1 sentence
- List assumptions made
- Proceed to execution (or ask for confirmation if required by risk profile)
Integration
- - Hook: pre-action phase (before tool calls)
- Risk-based: high-risk actions always require confirmation regardless of clarity
- Low-risk: can proceed automatically after clarification
Examples
User: "Zrób backup"
Agent: "Which folder? Where to store? Full or incremental?" (3 questions)
User: "Wyślij maila do klienta"
Agent: "Who is the client? What content? Any attachments?" (3 questions)
Configuration
- -
maxQuestions: 4 (default) - INLINECODE1 : true (default)
- INLINECODE2 : true (default)
Related Skills
- -
interaction-pipeline (status updates) - INLINECODE4 (risk tolerance)
引导式对话技能
目的
在执行操作前，通过最多提出4个澄清性问题来最小化用户摩擦。确保智能体理解模糊任务，同时避免过度追问。
使用场景
- - 用户请求模糊或缺少关键细节
- 操作影响重大（文件删除、支付、外部API调用）
- 用户未提供足够上下文以确保安全执行
约束条件
- - 每个任务最多4个问题
- 一旦获得足够清晰度即停止提问
- 绝不询问上下文中已有的信息
- 若4个问题后仍模糊不清，则尽力解释并继续执行（需通知用户）
输出
澄清后：
- 1. 用一句话总结理解的任务
- 列出所做的假设
- 继续执行（或根据风险等级要求确认）
集成方式
- - 钩子：操作前阶段（工具调用之前）
- 基于风险：高风险操作无论清晰度如何始终需要确认
- 低风险：澄清后可自动继续执行
示例
用户：Zrób backup
智能体：哪个文件夹？存储位置？完整备份还是增量备份？（3个问题）
用户：Wyślij maila do klienta
智能体：客户是谁？什么内容？是否有附件？（3个问题）
配置
- - maxQuestions：4（默认）
- autoProceedAfterClarity：true（默认）
- requireConfirmationHighRisk：true（默认）
相关技能
- - interaction-pipeline（状态更新）
- adaptive-execution-profile（风险容忍度）
