Elicitation Skill — Intent Compiler for AI Agents

English | 简体中文

Elicitation Skill is an AgentSkill that helps an AI assistant turn vague human intent into actionable definitions.

It is not a meeting-notes bot, a questionnaire, or a generic brainstorming prompt. It teaches an agent to ask a few sharp questions first, uncover the hidden judgment behind a fuzzy idea, and then compile the answer into an Intent Brief that can change a product, workflow, skill, memory rule, or agent behavior.

Why this exists

Most AI agents are good at executing once the task is clear. The hard part is often before execution:

• The human has a strong but blurry intuition.
• The product direction is not yet explicit.
• The team feels tension but cannot name it.
• The agent wants to act too early and produces a generic plan.

Elicitation Skill creates a lightweight protocol for this pre-execution layer.

Ask first. Synthesize after. Route into system change.

What it does

Given a fuzzy prompt such as:

• “Help me clarify this product idea.”
• “Let’s do an elicitation session.”
• “What agent/workflow are we missing?”
• “This direction feels right, but I can’t explain why yet.”
• “Turn this vague thought into something executable.”

The agent will:

1. Identify the elicitation type.
2. Ask 3–5 concise, high-leverage questions.
3. Wait for the human’s answers.
4. Produce an Intent Brief.
5. Suggest 1–3 concrete system feedback actions.

When to use it

Use it before execution when the task is still under-defined but high-stakes:

• Product strategy
• Agent / workflow design
• Creative direction
• Brand or taste alignment
• Weekly planning
• Postmortem synthesis
• Strategic tension clarification

Do not use it for routine Q&A or simple tasks that are already clear.

Repository structure

.
├── README.md
├── README.zh-CN.md
├── skill/
│   └── elicitation-agent/
│       ├── SKILL.md
│       └── references/
│           └── question-patterns.md
├── templates/
│   └── intent-brief.md
├── examples/
│   └── website-product-intent-brief.md
├── docs/
│   └── product-note.md
└── AGENTS.md

Installation

Copy the skill folder into your OpenClaw / Codex / agent skills directory:

cp -R skill/elicitation-agent ~/.openclaw/workspace/skills/

Then restart or reload your agent environment if required.

Quick start

Ask your agent:

Let’s do an elicitation session for this vague product idea:
I think our website should feel less like a corporate landing page and more like a world people can enter.

The agent should not immediately write a sitemap. It should ask a few clarifying questions first.

Core output: Intent Brief

The final artifact should distinguish:

• Explicit goal
• Hidden tension
• Agent judgment
• Executable translation
• Feedback target
• Verification method
• What not to do

See templates/intent-brief.md.

Design principles

• Fewer, sharper questions beat long forms.
• Do not summarize too early.
• Do not turn the human into the project manager.
• Do not attribute the human’s insights to the agent.
• Every session must route into at least one concrete change.
• If nothing changes, the elicitation is incomplete.

License

MIT. See LICENSE.