| name | sheets |
| description | Handle spreadsheets and tabular files safely; use text reads only for text tables and avoid treating Excel or Parquet binaries as UTF-8 files. |
| tools | ["data_agent_tools","request_clarification"] |
sheets
Overview
Use this skill for spreadsheet and tabular file requests, including .csv, .tsv, .xlsx, .xls, .ods, and .parquet.
Rules
- Do not call
read_file on .xlsx, .xls, .ods, or .parquet.
- Only treat
.csv and .tsv as text files.
- Use the data tools for structured analysis when the task is analytical rather than purely editorial.
- Be explicit when a binary spreadsheet format needs conversion before the agent can inspect it reliably.
Workflow
-
Identify the actual table format.
- Text tabular formats:
.csv, .tsv
- Binary spreadsheet formats:
.xlsx, .xls, .ods
- Columnar binary data:
.parquet
-
Choose the safest path.
- For
.csv and .tsv:
- You may use normal text-file workflows.
- If the user wants analysis, prefer the data toolchain.
- For
.parquet:
- Prefer the data toolchain.
- Do not attempt UTF-8 decoding.
- For
.xlsx:
- If the deliverable is an Excel workbook or the user asks to create/edit/recalculate/validate a workbook, load the
xlsx skill.
- Prefer the platform's derived Markdown artifact when the workbook was attached through chat or uploaded after workbook understanding support was enabled.
- Ask for a CSV export of the relevant sheet when exact formulas, formatting, hidden rows, or full-fidelity workbook behavior is required.
- For
.xls and .ods:
- Do not pretend the workbook is directly readable if no converter is available.
- Ask for a CSV or
.xlsx export of the relevant sheet when exact cell inspection is required.
-
Set expectations for workbook files.
.xlsx content is converted to a readable Markdown artifact with one CSV block per sheet; this is good for summarization and lightweight Q&A.
- If the user asks to inspect formulas, hidden tabs, formatting, pivots, charts, or exact workbook behavior, say the readable artifact may not preserve those details.
- Ask for the specific sheet exported as CSV if exact tabular inspection is needed.
-
Stay honest about scope.
- Do not fabricate sheet names, formulas, or hidden-tab content.
- If only partial tabular data is accessible, say exactly what was inspected.
Good uses
- Analyze CSV exports
- Work with Parquet-backed data through the data toolchain
- Help the user decide how to convert a workbook for agent analysis
Avoid
- Calling
read_file on Excel binaries
- Claiming workbook-level insight without a readable export