pandas-pro
Verifiedby Jeffallan · Updated Apr 16, 2026
Performs pandas DataFrame operations for data analysis, manipulation, and transformation. Use when working with pandas DataFrames, data cleaning, aggregation, merging, or time series analysis. Invoke for data manipulation tasks such as joining DataFrames on multiple keys, pivoting tables, resampling time series, handling NaN values with interpolation or forward-fill, groupby aggregations, type conversion, or performance optimization of large datasets.
See It In Action
Let me explore the data files first.
I can see three data quality issues that would cause conflicting CACs:
- Inconsistent channel names — "Facebook Ads" vs "Facebook" vs "facebook", "Google Ads" vs "google", etc.
- Missing values — LinkedIn W16 spend is blank; lead L1016 has no channel
- No join key alignment — channel names don't match across files
Let me build the reconciliation script.
The CRM table doesn't have a week column — need to get it from the leads join first. Let me fix that.
Reconciliation complete. Here's what was wrong and what I fixed:
Data Issues Resolved
| # | Problem | Fix |
|---|---|---|
| 1 | Channel names inconsistent across all 3 files (5 variants of "Facebook" alone) | Normalized to 4 canonical names via lowercase mapping |
| 2 | LinkedIn W16 spend missing | Interpolated to $6,750 (midpoint of W15 $6,500 and W17 $7,000) |
| 3 | Lead L1016 has no channel | Excluded from channel metrics (flagged in notes) |
| 4 | Pending/unqualified leads counted in some views | Only "qualified" leads feed CAC/CPL |
Key Numbers for the CFO
| Channel | 5-Week Spend | Customers | Blended CAC | Avg Deal | LTV:CAC |
|---|---|---|---|---|---|
| $6,250 | 4 | $1,563 | $6,650 | 4.26x | |
| Google Ads | $68,600 | 6 | $11,433 | $14,750 | 1.29x |
| $32,250 | 3 | $10,750 | $20,833 | 1.94x | |
| Facebook Ads | $46,400 | 4 | $11,600 | $9,425 | 0.81x |
| TOTAL | $153,500 | 17 | $9,029 | $12,665 | 1.40x |
The conflicting CACs were caused by the channel name mismatches — "google" leads weren't joining to "Google Ads" spend, so each file produced different denominators. Now reconciled.
Watch items: Facebook Ads is below 1.0x LTV:CAC (spending more to acquire than the deal is worth). W17 has zero closed deals across all channels (likely pipeline lag, not alarm).
Output files saved:
marketing_snapshot_weekly.csv— per-channel, per-week detailmarketing_snapshot_rollup.csv— channel-level summary with grand total
External Tools
| Tool | Type |
|---|---|
| python3 | binary |
Permissions
| Scope | Description |
|---|---|
| filesystem:read | |
| filesystem:write | |
| process:spawn |
SKILL.md
Pandas Pro
Expert pandas developer specializing in efficient data manipulation, analysis, and transformation workflows with production-grade performance patterns.
Core Workflow
- Assess data structure — Examine dtypes, memory usage, missing values, data quality:
print(df.dtypes) print(df.memory_usage(deep=True).sum() / 1e6, "MB") print(df.isna().sum()) print(df.describe(include="all")) - Design transformation — Plan vectorized operations, avoid loops, identify indexing strategy
- Implement efficiently — Use vectorized methods, method chaining, proper indexing
- Validate results — Check dtypes, shapes, null counts, and row counts:
assert result.shape[0] == expected_rows, f"Row count mismatch: {result.shape[0]}" assert result.isna().sum().sum() == 0, "Unexpected nulls after transform" assert set(result.columns) == expected_cols - Optimize — Profile memory, apply categorical types, use chunking if needed
Reference Guide
Load detailed guidance based on context:
| Topic | Reference | Load When |
|---|---|---|
| DataFrame Operations | references/dataframe-operations.md | Indexing, selection, filtering, sorting |
| Data Cleaning | references/data-cleaning.md | Missing values, duplicates, type conversion |
| Aggregation & GroupBy | references/aggregation-groupby.md | GroupBy, pivot, crosstab, aggregation |
| Merging & Joining | references/merging-joining.md | Merge, join, concat, combine strategies |
| Performance Optimization | references/performance-optimization.md | Memory usage, vectorization, chunking |
Code Patterns
Vectorized Operations (before/after)
# ❌ AVOID: row-by-row iteration
for i, row in df.iterrows():
df.at[i, 'tax'] = row['price'] * 0.2
# ✅ USE: vectorized assignment
df['tax'] = df['price'] * 0.2
Safe Subsetting with .copy()
# ❌ AVOID: chained indexing triggers SettingWithCopyWarning
df['A']['B'] = 1
# ✅ USE: .loc[] with explicit copy when mutating a subset
subset = df.loc[df['status'] == 'active', :].copy()
subset['score'] = subset['score'].fillna(0)
GroupBy Aggregation
summary = (
df.groupby(['region', 'category'], observed=True)
.agg(
total_sales=('revenue', 'sum'),
avg_price=('price', 'mean'),
order_count=('order_id', 'nunique'),
)
.reset_index()
)
Merge with Validation
merged = pd.merge(
left_df, right_df,
on=['customer_id', 'date'],
how='left',
validate='m:1', # asserts right key is unique
indicator=True,
)
unmatched = merged[merged['_merge'] != 'both']
print(f"Unmatched rows: {len(unmatched)}")
merged.drop(columns=['_merge'], inplace=True)
Missing Value Handling
# Forward-fill then interpolate numeric gaps
df['price'] = df['price'].ffill().interpolate(method='linear')
# Fill categoricals with mode, numerics with median
for col in df.select_dtypes(include='object'):
df[col] = df[col].fillna(df[col].mode()[0])
for col in df.select_dtypes(include='number'):
df[col] = df[col].fillna(df[col].median())
Time Series Resampling
daily = (
df.set_index('timestamp')
.resample('D')
.agg({'revenue': 'sum', 'sessions': 'count'})
.fillna(0)
)
Pivot Table
pivot = df.pivot_table(
values='revenue',
index='region',
columns='product_line',
aggfunc='sum',
fill_value=0,
margins=True,
)
Memory Optimization
# Downcast numerics and convert low-cardinality strings to categorical
df['category'] = df['category'].astype('category')
df['count'] = pd.to_numeric(df['count'], downcast='integer')
df['score'] = pd.to_numeric(df['score'], downcast='float')
print(df.memory_usage(deep=True).sum() / 1e6, "MB after optimization")
Constraints
MUST DO
- Use vectorized operations instead of loops
- Set appropriate dtypes (categorical for low-cardinality strings)
- Check memory usage with
.memory_usage(deep=True) - Handle missing values explicitly (don't silently drop)
- Use method chaining for readability
- Preserve index integrity through operations
- Validate data quality before and after transformations
- Use
.copy()when modifying subsets to avoid SettingWithCopyWarning
MUST NOT DO
- Iterate over DataFrame rows with
.iterrows()unless absolutely necessary - Use chained indexing (
df['A']['B']) — use.loc[]or.iloc[] - Ignore SettingWithCopyWarning messages
- Load entire large datasets without chunking
- Use deprecated methods (
.ix,.append()— usepd.concat()) - Convert to Python lists for operations possible in pandas
- Assume data is clean without validation
Output Templates
When implementing pandas solutions, provide:
- Code with vectorized operations and proper indexing
- Comments explaining complex transformations
- Memory/performance considerations if dataset is large
- Data validation checks (dtypes, nulls, shapes)
FAQ
What does pandas-pro do?
Performs pandas DataFrame operations for data analysis, manipulation, and transformation. Use when working with pandas DataFrames, data cleaning, aggregation, merging, or time series analysis. Invoke for data manipulation tasks such as joining DataFrames on multiple keys, pivoting tables, resampling time series, handling NaN values with interpolation or forward-fill, groupby aggregations, type conversion, or performance optimization of large datasets.
When should I use pandas-pro?
Use it when you need a repeatable workflow that produces source code, structured table.
What does pandas-pro output?
In the evaluated run it produced source code, structured table.
How do I install or invoke pandas-pro?
npx skills add https://github.com/jeffallan/claude-skills --skill pandas-pro
Which agents does pandas-pro support?
Claude Code
What tools, channels, or permissions does pandas-pro need?
It uses python3; channels commonly include code, table; permissions include filesystem:read, filesystem:write, process:spawn.
Is pandas-pro safe to install?
Static analysis marked this skill as medium risk; review side effects and permissions before enabling it.
How is pandas-pro different from an MCP or plugin?
A skill packages instructions and workflow conventions; tools, MCP servers, and plugins are dependencies the skill may call during execution.
Does pandas-pro outperform not using a skill?
About pandas-pro
When to use pandas-pro
You need to write or revise pandas code for DataFrame manipulation and cleaning. You are optimizing DataFrame operations for performance and memory usage. You need reliable groupby, merge, pivot, or resampling patterns with validation checks.
When pandas-pro is not the right choice
You are not working in Python/pandas. You need external database or SaaS integrations rather than local DataFrame processing.
What it produces
Produces source code and structured table.
Install
npx skills add https://github.com/jeffallan/claude-skills --skill pandas-proInvoke: Ask Claude Code to use pandas-pro for the task.