返回 Skill 列表
extension
分类: AI Agent 能力无需 API Key

dialogue-structured-elicit

通过框架感知的问题从用户那里收集结构化信息。当您需要引出流程上下文、需求、约束或决策时使用。触发词包括“收集需求”、“引出上下文”、“结构化问题”、“需要理解”。

person作者: jakexiaohubgithub

Dialogue: Structured Elicit

Gather structured information through framework-aware questioning.

When to Use

Use this skill when you need to:

  • Gather process context before designing capability flows
  • Elicit requirements or constraints from the user
  • Clarify ambiguous information
  • Get decisions on alternatives you've identified

Elicitation Patterns

Process Context Elicitation

When gathering context for process design, elicit information across six areas. Four are mandatory; two are contextual.

Mandatory Questions (must elicit before proceeding)

| Area | Purpose | Example Question | |------|---------|------------------| | Phase | Determine SDLC phase | "Which phase does this process belong to?" | | Purpose | Understand process goal | "What should this process accomplish?" | | Inputs | Identify input documents/data | "What information flows into this process?" | | Outputs | Identify deliverables | "What artifacts does this process produce?" |

Contextual Questions (elicit when relevant)

| Area | Purpose | When to Ask | |------|---------|-------------| | Actors | Understand who's involved | When unclear who participates; multi-team processes | | Constraints | Surface limitations | When regulatory, security, or technical constraints likely |

Minimum Coverage Requirement

Do not proceed to decomposition until all four mandatory areas have answers.

If the user cannot answer a mandatory question:

  1. Note the gap as an observation
  2. Propose a reasonable default with rationale
  3. Get explicit confirmation before proceeding

Example: "I don't have information about inputs yet. Based on similar Phase 3 processes, I'd expect requirements documents and stakeholder notes. Does that seem right?"

Coverage Checklist

Before proceeding from elicitation, verify:

[ ] Phase identified
[ ] Purpose stated
[ ] Inputs defined (or default confirmed)
[ ] Outputs defined (or default confirmed)
[ ] Elicited context logged as observation

Information Composition Elicitation

To estimate formal/tacit/emergent composition:

| Question | Indicates | |----------|-----------| | "Is there documented procedure for this?" | High formal if yes | | "Does this require expertise that's hard to document?" | High tacit if yes | | "Does new understanding emerge during execution?" | High emergent if yes | | "Could someone follow written instructions to do this?" | Low tacit if yes |

Pattern Selection Elicitation

When AI involvement is unclear:

| Question | Helps Determine | |----------|-----------------| | "Should a human always make this decision?" | Human-Only vs other | | "Could AI do this entirely without human review?" | AI-Only suitability | | "Is human judgement essential, or just review?" | Partnership vs AI-Led |

Using AskUserQuestion

Structure questions using the AskUserQuestion tool:

Use AskUserQuestion with:
- question: Clear, specific question
- header: Short label (max 12 chars)
- options: 2-4 distinct choices with descriptions
- multiSelect: true if multiple answers valid

Example: Phase Elicitation

question: "Which SDLC phase does this process belong to?"
header: "Phase"
options:
  - label: "Initiation (Phase 1)"
    description: "Project setup, feasibility, initial stakeholder engagement"
  - label: "Planning (Phase 2)"
    description: "Resource allocation, timeline, risk assessment"
  - label: "Requirements (Phase 3)"
    description: "Gathering and documenting what the system should do"
  - label: "Design (Phase 4)"
    description: "Architectural decisions, component design"

Example: Collaboration Pattern

question: "How should humans and AI collaborate on this step?"
header: "Pattern"
options:
  - label: "Human-Only"
    description: "Human does all work; AI not involved"
  - label: "Human-Led"
    description: "Human drives; AI assists and supports"
  - label: "Partnership"
    description: "Both essential; continuous collaboration"
  - label: "AI-Led"
    description: "AI drives; human reviews and approves"

Recording Elicitation Results

After eliciting information, use the dialogue-log-observation skill to record:

  • Key requirements noted
  • Constraints identified
  • User preferences stated

This preserves the elicited context for future reference.

Escalation

If the user cannot answer or indicates uncertainty:

  • Note the uncertainty as an observation
  • Propose reasonable defaults with rationale
  • Ask if defaults are acceptable

Phase 1 (Initiation) Specific Elicitation

Phase 1 has 75% tacit information composition. Standard elicitation patterns need adaptation.

Key Differences from Later Phases

| Aspect | Later Phases | Phase 1 | |--------|--------------|---------| | Information source | Documents, code, systems | Stakeholder minds | | Capture urgency | Can revisit documents | Must capture now or lose | | AI role | Can lead elicitation | Facilitates human articulation | | Validation | Check against formal specs | Probe for unstated assumptions |

Phase 1 Elicitation Questions

Opportunity/Problem Framing

| Question | Why Important | |----------|---------------| | "What triggered this initiative?" | Captures origin story before it's forgotten | | "What problem are you solving?" | Distinguishes problem from solution | | "Why now? What changed?" | Surfaces temporal context | | "What would happen if we did nothing?" | Tests problem significance | | "What other ways could this problem be framed?" | Explores alternatives |

Stakeholder Mapping

| Question | Why Important | |----------|---------------| | "Who requested this?" | Identifies primary stakeholder | | "Who must approve?" | Identifies decision authority | | "Who will be affected but isn't at the table?" | Surfaces hidden stakeholders | | "Who might resist? Why?" | Surfaces political dynamics | | "Whose expertise is essential?" | Identifies knowledge holders |

Rationale Capture (Critical)

These questions preserve tacit knowledge:

| Question | Captures | |----------|----------| | "Why do you think that?" | Reasoning behind positions | | "What alternatives did you consider?" | Negative knowledge | | "What assumptions are we making?" | Embedded constraints | | "What would change your mind?" | Decision criteria | | "What have we tried before?" | Historical context |

Capture Window Awareness

Phase 1 insights are highly perishable. When a significant insight emerges:

  1. Log it immediately using dialogue-log-observation
  2. Confirm you captured it correctly with the human
  3. Note if there's more context you should probe for

Do not wait until the end of a conversation to capture Phase 1 observations.

Example: Phase 1 AskUserQuestion

question: "What problem is this initiative trying to solve?"
header: "Problem"
options:
  - label: "Clear and documented"
    description: "Problem statement exists; I can point you to it"
  - label: "Clear but undocumented"
    description: "We know the problem but haven't written it down"
  - label: "Still being defined"
    description: "We're not sure yet; exploring options"
  - label: "Multiple competing framings"
    description: "Different stakeholders see it differently"

If "Still being defined" or "Multiple competing framings", switch to Facilitator mode:

  • Help structure the exploration
  • Surface different perspectives
  • Document alternative framings as they emerge
  • Avoid premature convergence

Phase 1 Process Suggestions

When Phase 1 work is substantial, consider offering structured processes:

| User Need | Suggest Process | Default Mode | |-----------|-----------------|--------------| | "I have an idea/opportunity" | PROC-1.1 Opportunity Identification | QUICK | | "Need to align stakeholders" | PROC-1.2 Stakeholder Alignment | QUICK | | "Problem isn't clear" | PROC-1.3 Problem Framing | QUICK | | "Need to justify investment" | PROC-1.4 Business Case | QUICK |

Offer these lightly: "Would you like to work through this systematically? I can guide you through a structured process that typically takes about 15 minutes."

If complexity emerges during elicitation, suggest upgrading to FULL mode.


Part of the Dialogue Framework