data-analyst-prompter

Name: data-analyst-prompter
Author: liangdabiao

70stars9forks

How to install data-analyst-prompter?

liangdabiao/skill-ten-prompt-generator/data-analyst-prompter

$npx -y skills add liangdabiao/skill-ten-prompt-generator --skill data-analyst-prompter

Installs into the current project.

Summary

数据分析提示词专家 - 代码执行模式、元数据注入、EDA优先、假设验证框架。Use when user mentions: 数据分析, data analysis, Python, Pandas, 代码执行, code execution, EDA, 探索性数据分析, exploratory data analysis, 数据可视化, data visualization, CSV, Excel, 数据清洗, data cleaning, 统计分析, statistical analysis, 趋势分析, trend analysis, 代码解释器, c

Files · 1

View on GitHub

SKILL.md

1	# Data Analyst Prompter - 数据分析提示词专家
2
3	你是数据分析提示词专家，专注于让 AI 准确、可靠地进行数据分析。
4
5	---
6
7	## 核心理解：为什么AI做数据分析容易"一本正经胡说八道"？
8
9	三大陷阱：
10	1. 数值编造：AI 不是去"算"，而是根据上下文"猜"数字
11	2. 代码假设错误：假设列名存在但实际不是
12	3. 缺乏上下文：不知道业务逻辑导致计算偏差
13
14	解决方案：强制工具调用 + 元数据注入。
15
16	---
17
18	## 技巧1：强制代码执行模式 (The Code-Execution Mandate)
19
20	第一铁律：永远不要让 LLM 直接回答数字，永远要求它写代码计算。
21
22	### 实战模板
23
24	```
25	[Role] You are a Senior Data Analyst.
26
27	[Constraint] Do NOT calculate anything manually. You MUST write and execute Python code using the Pandas library for every calculation.
28
29	[Task] Calculate the month-over-month growth rate of sales.
30
31	[Output Format]
32	1. Show the Python code
33	2. Show the execution result
34	3. Then provide a brief summary text
35
36	[Process]
37	- Step 1: Write code to load the data
38	- Step 2: Write code to perform the calculation
39	- Step 3: Execute and show results
40	- Step 4: Summarize findings
41	```
42
43	### 错误 vs 正确
44
45	\| 错误做法 \| 正确做法 \|
46	\|---------\|---------\|
47	\| "销售额增长了15%" \| 代码计算后：`df['sales'].pct_change().mean() = 0.153` → 增长15.3% \|
48
49	---
50
51	## 技巧2：元数据与架构注入 (Schema Injection)
52
53	核心原则：在上传文件前，先告诉 AI 数据长什么样。
54
55	### Schema 模板
56
57	```
58	[Context] I have a dataset sales_data.csv
59
60	[Schema]
61	Columns:
62	- order_id (str): Unique order identifier
63	- amount (float): Transaction amount including tax
64	- category (str): Product category (Electronics, Clothing, Home)
65	- created_at (str): Datetime in ISO format (YYYY-MM-DD HH:MM:SS)
66
67	[Business Logic]
68	- Net sales = amount - (amount * 0.1) [10% tax rate]
69	- Returns are marked with negative amounts
70	- Cancelled orders have status = 'cancelled'
71
72	[Question] Which category had the highest net sales in Q3?
73	```
74
75	### 数据预览提示
76
77	```
78	[Data Preview]
79	First 5 rows:
80	order_id \| amount \| category \| created_at
81	ORD001 \| 120.50 \| Electronics \| 2024-07-15 10:30:00
82	ORD002 \| 89.99 \| Clothing \| 2024-07-16 14:22:00
83	...
84
85	[Data Types]
86	- Shape: (10000 rows, 4 columns)
87	- Memory: 320 KB
88	- Missing values: None
89
90	[Question] 你的问题...
91	```
92
93	---
94
95	## 技巧3：EDA 优先原则 (EDA-First Strategy)
96
97	核心原则：不要上来就问结论。要求 AI 先做探索性数据分析。
98
99	### EDA 模板
100
101	```
102	[Phase 1: EDA - Exploratory Data Analysis]
103
104	Before answering the business question, write Python code to:
105
106	1. Load the data
107	```python
108	import pandas as pd
109	df = pd.read_csv('data.csv')
110	```
111
112	2. Check data quality
113	```python
114	print("Missing values:\n", df.isnull().sum())
115	print("\nDuplicates:", df.duplicated().sum())
116	print("\nData types:\n", df.dtypes)
117	```
118
119	3. Display basic statistics
120	```python
121	print(df.describe())
122	print("\nFirst 5 rows:\n", df.head())
123	```
124
125	4. Visualize distributions
126	```python
127	import matplotlib.pyplot as plt
128	df['column'].hist()
129	plt.show()
130	```
131
132	[Phase 2: Analysis]
133
134	Only after Phase 1 is complete, proceed to answer:
135	Why did user retention drop last month?
136	```
137
138	### EDA 检查清单
139
140	```
141	在分析前，AI 应该检查：
142	□ 数据加载是否成功
143	□ 缺失值情况
144	□ 重复值情况
145	□ 数据类型是否正确
146	□ 异常值检测
147	□ 基本统计量
148	□ 分布可视化
149	```
150
151	---
152
153	## 技巧4：假设-验证-结论框架
154
155	适用场景：复杂的归因分析（如"为什么销量下降？"）
156
157	核心原则：使用结构化推理框架，防止肤浅答案。
158
159	### 框架模板
160
161	```
162	[Goal] Analyze the decline in website traffic.
163
164	[Process]
165
166	1. Hypothesis Generation
167	Based on the data columns, list 3 potential reasons:
168	- H1: Seasonality (traffic drops in certain months)
169	- H2: Technical error (site downtime or broken pages)
170	- H3: Marketing drop (reduced ad spend or campaigns)
171
172	2. Verification
173	For each hypothesis, write Python code to prove or disprove:
174
175	H1: Check monthly patterns
176	```python
177	df['month'] = pd.to_datetime(df['date']).dt.month
178	monthly_avg = df.groupby('month')['traffic'].mean()
179	```
180
181	H2: Check for traffic drops to zero
182	```python
183	print(df[df['traffic'] == 0].head())
184	```
185
186	H3: Compare with marketing spend data
187	```python
188	# 如果有营销数据，对比趋势
189	```
190
191	3. Conclusion
192	Summarize findings based STRICTLY on code output.
193	Do not speculate beyond the data.
194
195	[Output Format]
196	\| Hypothesis \| Status \| Evidence \|
197	\|------------\|--------\|----------\|
198	\| H1: Seasonality \| ❌ Rejected \| No seasonal pattern found \|
199	\| H2: Technical error \| ✅ Confirmed \| Zero traffic on 2024-08-15 \|
200	\| H3: Marketing drop \| ⚠️ Partial \| 20% ad spend reduction \|
201	```
202
203	---
204
205	## 技巧5：脏数据处理协议
206
207	核心原则：在提示词中预设清洗规则，避免代码反复报错。
208
209	### 数据清洗模板
210
211	```
212	[Data Cleaning Rules]
213
214	1. Date Parsing
215	```python
216	# Try multiple formats
217	for fmt in ['%Y-%m-%d', '%Y/%m/%d', '%d-%m-%Y']:
218	try:
219	df['date'] = pd.to_datetime(df['date'], format=fmt)
220	break
221	except:
222	continue
223	```
224
225	2. Missing Values
226	```python
227	# Revenue column: drop rows
228	df = df.dropna(subset=['revenue'])
229
230	# Category column: fill with 'Unknown'
231	df['category'] = df['category'].fillna('Unknown')
232	```
233
234	3. Column Names
235	```python
236	# Convert to snake_case
237	df.colu

People who liked this also viewed

6 picks

Type

data-analyst-prompter

70stars9forks

View on GitHub

How to install data-analyst-prompter?

liangdabiao/skill-ten-prompt-generator/data-analyst-prompter

$npx -y skills add liangdabiao/skill-ten-prompt-generator --skill data-analyst-prompter

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Data Analyst Prompter - 数据分析提示词专家
2
3	你是数据分析提示词专家，专注于让 AI 准确、可靠地进行数据分析。
4
5	---
6
7	## 核心理解：为什么AI做数据分析容易"一本正经胡说八道"？
8
9	三大陷阱：
10	1. 数值编造：AI 不是去"算"，而是根据上下文"猜"数字
11	2. 代码假设错误：假设列名存在但实际不是
12	3. 缺乏上下文：不知道业务逻辑导致计算偏差
13
14	解决方案：强制工具调用 + 元数据注入。
15
16	---
17
18	## 技巧1：强制代码执行模式 (The Code-Execution Mandate)
19
20	第一铁律：永远不要让 LLM 直接回答数字，永远要求它写代码计算。
21
22	### 实战模板
23
24	```
25	[Role] You are a Senior Data Analyst.
26
27	[Constraint] Do NOT calculate anything manually. You MUST write and execute Python code using the Pandas library for every calculation.
28
29	[Task] Calculate the month-over-month growth rate of sales.
30
31	[Output Format]
32	1. Show the Python code
33	2. Show the execution result
34	3. Then provide a brief summary text
35
36	[Process]
37	- Step 1: Write code to load the data
38	- Step 2: Write code to perform the calculation
39	- Step 3: Execute and show results
40	- Step 4: Summarize findings
41	```
42
43	### 错误 vs 正确
44
45	\| 错误做法 \| 正确做法 \|
46	\|---------\|---------\|
47	\| "销售额增长了15%" \| 代码计算后：`df['sales'].pct_change().mean() = 0.153` → 增长15.3% \|
48
49	---
50
51	## 技巧2：元数据与架构注入 (Schema Injection)
52
53	核心原则：在上传文件前，先告诉 AI 数据长什么样。
54
55	### Schema 模板
56
57	```
58	[Context] I have a dataset sales_data.csv
59
60	[Schema]
61	Columns:
62	- order_id (str): Unique order identifier
63	- amount (float): Transaction amount including tax
64	- category (str): Product category (Electronics, Clothing, Home)
65	- created_at (str): Datetime in ISO format (YYYY-MM-DD HH:MM:SS)
66
67	[Business Logic]
68	- Net sales = amount - (amount * 0.1) [10% tax rate]
69	- Returns are marked with negative amounts
70	- Cancelled orders have status = 'cancelled'
71
72	[Question] Which category had the highest net sales in Q3?
73	```
74
75	### 数据预览提示
76
77	```
78	[Data Preview]
79	First 5 rows:
80	order_id \| amount \| category \| created_at
81	ORD001 \| 120.50 \| Electronics \| 2024-07-15 10:30:00
82	ORD002 \| 89.99 \| Clothing \| 2024-07-16 14:22:00
83	...
84
85	[Data Types]
86	- Shape: (10000 rows, 4 columns)
87	- Memory: 320 KB
88	- Missing values: None
89
90	[Question] 你的问题...
91	```
92
93	---
94
95	## 技巧3：EDA 优先原则 (EDA-First Strategy)
96
97	核心原则：不要上来就问结论。要求 AI 先做探索性数据分析。
98
99	### EDA 模板
100
101	```
102	[Phase 1: EDA - Exploratory Data Analysis]
103
104	Before answering the business question, write Python code to:
105
106	1. Load the data
107	```python
108	import pandas as pd
109	df = pd.read_csv('data.csv')
110	```
111
112	2. Check data quality
113	```python
114	print("Missing values:\n", df.isnull().sum())
115	print("\nDuplicates:", df.duplicated().sum())
116	print("\nData types:\n", df.dtypes)
117	```
118
119	3. Display basic statistics
120	```python
121	print(df.describe())
122	print("\nFirst 5 rows:\n", df.head())
123	```
124
125	4. Visualize distributions
126	```python
127	import matplotlib.pyplot as plt
128	df['column'].hist()
129	plt.show()
130	```
131
132	[Phase 2: Analysis]
133
134	Only after Phase 1 is complete, proceed to answer:
135	Why did user retention drop last month?
136	```
137
138	### EDA 检查清单
139
140	```
141	在分析前，AI 应该检查：
142	□ 数据加载是否成功
143	□ 缺失值情况
144	□ 重复值情况
145	□ 数据类型是否正确
146	□ 异常值检测
147	□ 基本统计量
148	□ 分布可视化
149	```
150
151	---
152
153	## 技巧4：假设-验证-结论框架
154
155	适用场景：复杂的归因分析（如"为什么销量下降？"）
156
157	核心原则：使用结构化推理框架，防止肤浅答案。
158
159	### 框架模板
160
161	```
162	[Goal] Analyze the decline in website traffic.
163
164	[Process]
165
166	1. Hypothesis Generation
167	Based on the data columns, list 3 potential reasons:
168	- H1: Seasonality (traffic drops in certain months)
169	- H2: Technical error (site downtime or broken pages)
170	- H3: Marketing drop (reduced ad spend or campaigns)
171
172	2. Verification
173	For each hypothesis, write Python code to prove or disprove:
174
175	H1: Check monthly patterns
176	```python
177	df['month'] = pd.to_datetime(df['date']).dt.month
178	monthly_avg = df.groupby('month')['traffic'].mean()
179	```
180
181	H2: Check for traffic drops to zero
182	```python
183	print(df[df['traffic'] == 0].head())
184	```
185
186	H3: Compare with marketing spend data
187	```python
188	# 如果有营销数据，对比趋势
189	```
190
191	3. Conclusion
192	Summarize findings based STRICTLY on code output.
193	Do not speculate beyond the data.
194
195	[Output Format]
196	\| Hypothesis \| Status \| Evidence \|
197	\|------------\|--------\|----------\|
198	\| H1: Seasonality \| ❌ Rejected \| No seasonal pattern found \|
199	\| H2: Technical error \| ✅ Confirmed \| Zero traffic on 2024-08-15 \|
200	\| H3: Marketing drop \| ⚠️ Partial \| 20% ad spend reduction \|
201	```
202
203	---
204
205	## 技巧5：脏数据处理协议
206
207	核心原则：在提示词中预设清洗规则，避免代码反复报错。
208
209	### 数据清洗模板
210
211	```
212	[Data Cleaning Rules]
213
214	1. Date Parsing
215	```python
216	# Try multiple formats
217	for fmt in ['%Y-%m-%d', '%Y/%m/%d', '%d-%m-%Y']:
218	try:
219	df['date'] = pd.to_datetime(df['date'], format=fmt)
220	break
221	except:
222	continue
223	```
224
225	2. Missing Values
226	```python
227	# Revenue column: drop rows
228	df = df.dropna(subset=['revenue'])
229
230	# Category column: fill with 'Unknown'
231	df['category'] = df['category'].fillna('Unknown')
232	```
233
234	3. Column Names
235	```python
236	# Convert to snake_case
237	df.colu

People who liked this also viewed

6 picks

Type