Data Analyst - Automated Data Analysis & Report Generator
Upload any data file (CSV, Excel, JSON, SQL export) and get a complete analysis report with insights, anomalies, and actionable recommendations — no code required.
📖 如何使用 / How to Use
安装 / Install
openclaw skills install smart-data-analyst
支持格式 / Supported Formats
.csv · .xlsx · .xls · .json · .tsv · 直接粘贴表格内容
使用步骤 / Steps
- 打开 OpenClaw，上传你的数据文件（拖拽或附件按钮）
- 说一句话触发：
帮我分析这个数据
analyze this CSV and find key insights
示例场景 / Example Use Cases
| 场景 | 触发语句 |
|---|---|
| 电商销售分析 | "分析这个月销售数据，找出下滑原因" |
| 财务报表 | "帮我分析这个 Excel，重点看支出异常" |
| 用户行为数据 | "找出用户流失的规律" |
| 库存数据 | "哪些 SKU 是滞销品？" |
报告包含 / Report Includes
✅ 数据概览（行列数、数据类型、缺失值统计） ✅ 描述性统计（均值、中位数、异常值） ✅ 相关性分析 ✅ 时序趋势（有日期列时自动检测） ✅ 异常检测 + 数据质量评分 ✅ 中文数据 → 全程中文输出 ✅ 可执行的改进建议
追问示例 / Follow-up Prompts
把第3个发现展开分析一下
帮我生成一份可以发给老板的摘要，不超过200字
对比今年和去年同期的差异
常见问题 / FAQ
- 数据安全吗？ 数据只在本地 OpenClaw 和你配置的 AI 之间流转，不存储到第三方
- 文件多大？ 建议不超过 10MB，超大数据集自动采样
- 需要额外配置吗？ OpenClaw 配置好 AI 即可使用，支持阿里云百炼、DeepSeek 等，录入 Key 就行
- 有问题找谁？ ClawHub 页面留言或联系 @ShuaigeSkillBot
- 需要帮你配置好直接用？ Telegram 私信 @ShuaigeSkillBot，配置服务 ¥99，配好即用
Trigger
When the user says any of: "analyze this data", "data analysis", "analyze CSV", "analyze Excel", "what does this data show", "find insights", "data report", "分析数据", "分析这个表", "数据报告", "看看这个数据", "帮我分析"
Or when the user uploads/provides a .csv, .xlsx, .json, .tsv file and asks about its contents.
Workflow
Step 1: Data Ingestion
Read the uploaded file. Determine:
- File format and encoding
- Number of rows and columns
- Column names and data types
- Missing values per column
- Sample of first 5 and last 5 rows
Output a Data Overview table immediately so the user knows you understood the file.
Step 2: Automated Analysis
Perform ALL applicable analyses based on the data type:
For Numeric Data:
- Descriptive statistics (mean, median, mode, std dev, min, max, quartiles)
- Distribution shape (normal, skewed, bimodal)
- Outlier detection (values beyond 2 standard deviations)
- Correlation matrix between numeric columns
- Trend analysis if time-series data is detected
For Categorical Data:
- Value counts and frequency distribution
- Top N most common values
- Cardinality (number of unique values)
- Cross-tabulation with other categorical columns
For Time-Series Data:
- Trend direction (increasing / decreasing / stable)
- Seasonality patterns
- Period-over-period changes (day/week/month/year)
- Notable spikes or drops with dates
For Mixed Data:
- Group-by analysis (numeric stats per category)
- Segment comparison
- Pivot table summaries
Step 3: Anomaly Detection
Flag anything unusual:
- Sudden value changes (> 2x standard deviation)
- Missing data patterns (random vs systematic)
- Duplicate rows
- Data quality issues (mixed types, encoding errors, impossible values)
Step 4: Generate Report
Output a structured report in markdown:
## Data Analysis Report
### 1. Overview
- File: [name]
- Records: [count]
- Time Range: [if applicable]
- Data Quality Score: [percentage of clean/complete data]
### 2. Key Findings
- Finding 1: [most important insight]
- Finding 2: [second most important]
- Finding 3: [third]
### 3. Detailed Analysis
[Full analysis results organized by section]
### 4. Anomalies & Warnings
[Flagged issues with severity: HIGH / MEDIUM / LOW]
### 5. Recommendations
- Actionable recommendation 1
- Actionable recommendation 2
- Suggested next steps for deeper analysis
### 6. Data Quality Notes
[Missing values, duplicates, type issues]
Step 5: Interactive Follow-up
After delivering the report, offer:
- "Want me to drill deeper into any specific finding?"
- "Should I compare specific columns or segments?"
- "Want me to generate a summary for your team/boss?"
Rules
- Always show the Data Overview first before diving into analysis — the user needs to confirm you read the file correctly.
- Use plain language for insights. Say "Sales dropped 23% in March" not "The dependent variable exhibited a negative coefficient of -0.23 in period 3."
- If the dataset has > 100 columns, ask the user which columns to focus on.
- For Chinese data (Chinese column names, Chinese values), output the entire report in Chinese.
- Never fabricate data points. If a calculation can't be performed, explain why.
- For large datasets (> 100K rows), note that analysis is based on statistical sampling and state the sample size.
- Always end with actionable recommendations, not just observations.
📢 每日市场数据播报，关注 Telegram 频道：https://t.me/shuaigeclaw