sheets

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

1. Identify the actual table format.

   • Text tabular formats: .csv, .tsv
   • Binary spreadsheet formats: .xlsx, .xls, .ods
   • Columnar binary data: .parquet

2. 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.

3. 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.

4. 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