| name | csv-insights |
| description | Analyze CSV data by computing aggregations, filtering rows, and generating readable summaries. Use when working with tabular data files in a container environment where the agent creates and manipulates files directly.
|
CSV Insights
Analyze CSV files to extract statistics, filter data, and produce human-readable summaries of findings.
Approach
1. Read and Validate the Data
Before computing anything, understand the dataset:
- Load the CSV and inspect the first few rows to confirm column names and types
- Check for missing values, inconsistent formatting, or mixed types in columns
- Note the row count and column count as baseline context
- If the CSV does not exist yet, create it from the provided data with clean headers
2. Compute Column Statistics
For each numeric column, compute:
- Count — number of non-null values
- Sum — total of all values
- Average — arithmetic mean
- Min / Max — range boundaries
- Median — middle value for skew detection
For categorical columns, compute:
- Distinct values — unique entries and their counts
- Most frequent — mode and its percentage of total rows
- Missing rate — percentage of null or empty entries
3. Filter and Group
Apply filters and groupings based on the analysis goal:
- Filter rows by column value (exact match, range, or pattern)
- Group by one or more categorical columns
- Compute aggregations within each group (sum, average, count)
- Sort results by the aggregated metric to surface top/bottom segments
- Limit output to the most relevant rows when datasets are large
4. Generate a Summary
Produce a readable narrative from the computed results:
- Lead with the key finding — the single most important insight
- Support with 2-3 secondary observations backed by specific numbers
- Flag anomalies — outliers, unexpected nulls, or skewed distributions
- State the scope — how many rows and columns were analyzed, any filters applied
- Use plain language — avoid jargon, explain what the numbers mean in context
Output Format
Structure results as:
- Dataset overview — row count, column inventory, data quality notes
- Statistics table — per-column metrics in a readable table
- Grouped analysis — aggregations by category if grouping was applied
- Key findings — narrative summary of the most important insights
- Data quality notes — any issues found (missing values, type mismatches, duplicates)
Guidelines
- Prefer exact numbers over vague descriptions ("revenue is 42,350" not "revenue is high")
- When comparing groups, include both absolute values and relative differences
- If the dataset is too large to display fully, show representative samples and totals
- Preserve the original data — create new files for transformed outputs rather than modifying the source
