confirm-form – OpenClaw Skill

---

name: confirm-form description: Generate structured confirmation forms to collect user feedback on multiple questions. Use when completing work that needs user review, when multiple issues need batch confirmation, or when the user needs to choose between options with detailed context. Triggers include review, confirm, batch questions, multiple choices, need user input on several items.

Confirm Form

Generate HTML forms for structured user confirmation, upload to GitHub Gist, and parse responses.

Quick Start

1. Prepare Questions JSON

[
  {
    "title": "Question title",
    "context": "Background: what I was working on",
    "uncertainty": "What specifically I cannot decide alone",
    "findings": [
      { "content": "Finding 1", "source": "Source A" },
      { "content": "Finding 2", "source": "Source B" }
    ],
    "judgment": "My recommendation and reasoning",
    "options": [
      { "label": "Option A", "basis": "Rationale for A" },
      { "label": "Option B", "basis": "Rationale for B【我的推荐】" },
      { "label": "Need more info", "basis": "If none of the above fits" }
    ]
  }
]

2. Generate Form

node scripts/generate.js questions.json

Output includes:

• Local HTML file
• GitHub Gist URL
• Preview link (htmlpreview.github.io)

3. Send Link to User

Send the preview URL. User fills the form and copies the result JSON back.

4. Parse Response

User response structure:

{
  "formId": "form-20260130-180000",
  "timestamp": "...",
  "globalFeedback": "all_ok|need_more_info|discuss|null",
  "globalComment": "Overall feedback",
  "summary": { "total": 3, "answered": 3, "agreedWithAI": 2 },
  "answers": [
    {
      "question": "Question title",
      "selectedLabel": "Option B",
      "customAnswer": "User's custom input if any",
      "agreedWithMyJudgment": true
    }
  ]
}

Question Design Guidelines

Required Fields

• title: Clear, concise question title
• options: At least 2 options, each with label

Recommended Fields

• context: Background - what was I working on
• uncertainty: Why I need to ask - what blocks my decision
• findings: Evidence with sources - show original text, not just summaries
• judgment: My recommendation with reasoning

Option Design

• Add basis (rationale) to each option
• Mark recommended option with 【我的推荐】 in basis
• Include "Need more information" as fallback option

Quality Principles

1. Give full context - Show original text, not just extracted numbers
2. Explain uncertainty - Why can't I decide alone?
3. Cite sources - Where did each finding come from?
4. Justify recommendations - Why do I lean this way?

Workflow Integration

CC Task Integration

When CC completes a task that needs user confirmation, include in CC prompt:

如果有需要主人确认的问题，在任务最后生成 questions.json 文件，格式：
[{"title":"问题","context":"背景","uncertainty":"不确定点","findings":[...],"judgment":"判断","options":[...]}]

After CC completes, check for questions.json and generate form if exists.

Response Archiving

After receiving user response, save to archive:

~/clawd/records/confirm-form/YYYY-MM-DD_<formId>.json

Archive format:

{
  "formId": "...",
  "createdAt": "...",
  "respondedAt": "...",
  "questions": [...],
  "response": {...}
}

Use for: reviewing past decisions, tracking confirmation patterns.

Files

• scripts/generate.js - Form generator script
• assets/template.html - HTML form template (copy-paste mode)
• assets/template-v2.html - HTML form template (auto-notify mode, experimental)
• assets/examples/sample.json - Example questions

Confirm Form

让 AI 助手批量收集决策，人类一次性确认，效率翻倍

痛点

AI 做完任务后，经常遇到需要人类决策的问题：

• 数据有出入，用哪个？
• 方案有几种，选哪个？
• 格式不确定，要改吗？

传统方式：AI 问一个 → 人答一个 → AI 再问 → 人再答... 来回聊天，效率低下。

解决方案

AI 完成任务
    ↓
遇到 5 个需要决策的问题
    ↓
生成 1 个确认表单（含完整上下文）
    ↓
发链接给人类
    ↓
人类一次看完、一次确认
    ↓
AI 拿到结果继续执行

效果：本来要来回 10 条消息的沟通，变成 1 个表单 + 1 次确认。

表单长什么样

每个问题包含：

• 📋 问题标题：清晰的一句话
• 📖 背景：我在做什么任务
• ❓ 不确定点：为什么需要你决定
• 📎 发现：原始数据 + 来源引用
• 💡 我的判断：AI 的推荐 + 理由
• ✅ 选项：带理由的多选项

人类只需要勾选，不用重新看原始材料。

快速使用

# 1. 准备问题 JSON
cat questions.json

# 2. 生成表单
node scripts/generate.js questions.json

# 3. 发送链接给人类，等待确认

# 4. 解析返回的 JSON 结果

适用场景

• ✅ CC (Claude Code) 完成文档整理，发现数据不一致
• ✅ AI 做调研，有多个方案需要选择
• ✅ 批量审核任务，需要人类抽查确认
• ✅ 任何"AI 做完了但有几个点拿不准"的情况

安装

这是一个 Clawdbot / Agent Skill。

# Clawdbot 用户
clawdhub install confirm-form

# 或直接克隆
git clone https://github.com/xiaozhuang0127/confirm-form-skill.git

详细文档

见 SKILL.md

---

Made with 💁‍♀️ by XiaoZhuang