evylegzhanin/agent-with-sandbox/excel

Excel Workbooks and Spreadsheets

Load python-scripts via read_skill if you have not already; use code in every write_python_file call, then run_python_file.

Reading and analyzing .xlsx / .xls

Before task-specific logic, profile the workbook once with staged Python:

• List sheet names (pd.ExcelFile(path).sheet_names).
• Load each relevant sheet.
• Print shape, column names, dtypes, missing-value counts, df.info(), and head/sample rows.

Use those observations for the rest of the analysis. If profiling invalidates an assumption, revise via revise_todo_list when a plan exists.

Save large tables under /workspace/out as .csv (or .xlsx when formatting matters) when follow-on scripts need them. Parquet is not available (no pyarrow).

Creating workbooks

• Prefer xlsxwriter for new multi-sheet reports with formatting-heavy output; use openpyxl when you must read and write the same file or apply fine-grained cell edits.
• Apply sensible column widths, header bolding, number/date formats, and freeze panes on wide sheets when helpful.
• For pivot-style summaries, prefer explicit pivot_table / groupby outputs written as clearly labeled tables rather than relying on Excel pivot cache features unless the task requires native pivots.

Quality

After writes, optionally reload with openpyxl or pandas in a short follow-on staged script to assert sheet names and row counts when the task is correctness-sensitive.