| name | excel-xray |
| description | Inspect large or complex XLSX workbooks with this repository's JBang and Apache POI analyzer before answering. Use for workbook structure, formulas, bounded cell evidence, the UCI Online Retail II audit, or embedded images when the binary must stay outside the agent context.
|
Excel X-Ray
Keep the workbook on disk. Bring only a bounded inventory, aggregate, cell range, or selected image into context.
Required Workflow
-
Confirm the input is an .xlsx file inside the workspace. In the command patterns below, replace $WORKBOOK with that exact path before running the command; do not run with an unset variable. Do not modify the input, encode it as text, or dump its ZIP/XML contents into the conversation.
-
From the repository root, inventory it before running any analysis command:
jbang --offline scripts/ExcelXray.java inventory "$WORKBOOK" \
--sample-rows 0 \
--max-output-chars 12000
-
If the column schema is unknown, use slice for the header and at most one representative data row after the inventory has identified the sheet and dimensions. Do not dump sample rows from every sheet by default.
-
Turn the user's request into one narrow question. State the metric, filters, grouping, and treatment of missing values before calculating.
-
Use the smallest command that can answer it:
- For the tutorial's retail question, read the retail audit contract, then run
audit-retail.
- For exact cell evidence, use
slice with one sheet and one explicit range.
- For visuals, run
images first. Use extract-image for one selected index only.
-
If a command reaches --max-output-chars, reduce the sample or range. Do not remove the cap just to make the command succeed.
-
Cite workbook evidence as Sheet!A1:H20 ranges. Keep conclusions separate from assumptions and data-quality rules.
For data/online_retail_II.xlsx, the required order is exact:
- Run the zero-sample
inventory command above.
- Read the retail audit contract.
- Run
audit-retail.
Do not skip or reorder these steps even when the user prompt only asks for the answer.
Command Patterns
jbang --offline scripts/ExcelXray.java audit-retail "$WORKBOOK" \
--evidence-lines 3 \
--max-output-chars 12000
jbang --offline scripts/ExcelXray.java slice "$WORKBOOK" \
--sheet "Year 2010-2011" \
--range "A540424:H540424" \
--include-formulas \
--max-output-chars 12000
jbang --offline scripts/ExcelXray.java images "$WORKBOOK"
jbang --offline scripts/ExcelXray.java extract-image "$WORKBOOK" \
--index 1 \
--output build/selected-image.png
Accuracy Rules
- Treat formula results as cached workbook values unless a spreadsheet engine has recalculated them. Return both formula and cached value when they matter.
- Call a number a character count when the tool reports characters. Do not convert it to tokens without a named tokenizer and model.
- The analyzer reads the workbook from disk. Say that the binary stayed outside the model context; do not say that the workbook was never read.
- Count inventory samples, slices, audit evidence, and extracted images when describing what entered the model context. A zero-sample inventory does not add cell rows.
- Do not call every stock code a product. Apply documented business rules before ranking.
- Do not hide excluded or missing rows. Report the applied policy.
- Do not speculate about the identities or business meaning behind missing values.
- Do not claim that a workbook image proves a numeric result unless its visible content agrees with cell-level evidence.
Final Answer Shape
Keep the final answer under 250 words unless the user asks for more detail. Use short paragraphs or bullets, not tables. Do not repeat the complete JSON response, scan statistics, cancellation counts, or unrelated workbook totals.
Return:
- The direct answer.
- The calculation scope and business rules.
- A small set of sheet-and-range evidence.
- Any formula-cache, missing-data, image, or file-format limitation that could change the conclusion.
For the retail audit, finish with one accurate context statement: the zero-sample inventory contributed workbook metadata but no cell rows; the audit contributed its returned evidence rows and aggregates; the analyzer read the binary from disk, but the binary did not enter model context.
