DataToRAGDataToRAG
All skills

You run it out of a spreadsheet

“The spreadsheet is the system. Everything real about how we operate is in it, and it is getting away from me.”

Inventory, pipeline, roadmap, who-owes-what. It started as a tab and became the place the answer actually lives, which is fine right up until you want something else to read it.

A spreadsheet is a human interface. An agent reading one is doing document archaeology: merged headers arrive as repeated text, a colour means nothing, and a formula reads back as its result. Most "let the AI read our spreadsheet" attempts fail on that and not on the model.

The skills below are about making the sheet legible to something that is not you: a schema that survives being read, and the traps worth knowing before you find them the expensive way.

“Our notes are scattered and half of them are stale. I want one place my agent can read and update.”

A Google Sheet your agent reads and writes reliably, without the traps that make spreadsheets bad interfaces.

Google Workspace5 tools
Read the skill

“A 'share what you're building' thread has a hundred products buried in comments, and I want them as a directory I can sort, share, and keep current.”

A formatted, deduplicated Google Sheet directory built from a thread, safe to share publicly, with every entry summarized in your words rather than pasted from the pitch.

Google Workspace7 tools
Read the skill

“Every Friday someone asks what I got done this week, and I rebuild the answer from memory.”

A written weekly update in Google Docs, built from your own calendar and sent mail, with a Sheet that indexes every week.

Google Workspace8 tools
Read the skill

“Claude fills the spreadsheet with the right values, and every long cell is truncated to a sliver in a 100-pixel column.”

A Google Sheet with sensible column widths, wrapped text, a frozen header, section bands and your fonts, formatted through the Sheets API after the values are in.

Google Workspace6 tools
Read the skill

“The same report lands in my inbox every week and I file it by hand, or I forget to.”

Attachments filed into the right Drive folder, without the files ever passing through the conversation.

Google Workspace4 tools
Read the skill

“The summary already exists. Turning it into slides is an hour of work nobody wants to do.”

A structured deck built from the document you already wrote, ready to edit rather than ready to present.

Google Workspace4 tools
Read the skill

Questions from people in this situation

Why do assistants struggle with the spreadsheet we run on? #

Because a spreadsheet is a human interface. Merged headers arrive as repeated text with no indication of which columns they spanned, a colour means nothing, and a formula reads back as its result, so an agent reading one is doing document archaeology. Most attempts at this fail there rather than on the model.

Can it change the sheet, or only read it? #

Change it. Cell level updates, appended rows, added and renamed tabs, and a find-rows call that hands back the matching row numbers with a ready made range for each, so a lookup can be followed straight by an update. The Sheets docs list every action.

What makes a sheet legible to something that is not you? #

A schema that survives being read: one header row, no merged cells anywhere, no spacer columns, an id column on every table so rows are never matched on a display name, and one concern per tab. That is what the knowledge base skill on this page sets up, and it is the part worth doing before anything else.

If a value starts with an equals sign, does it become a formula? #

Not unless that call asked for it. Values written through the DataToRAG Sheets connector are stored as text by default, and evaluation is turned on per call with parse_formulas rather than per cell. Anything harvested from somewhere else belongs in a call with parsing off, which is why the Reddit directory skill writes each row in two calls rather than one.

Not quite you?

These are the ones that fit this situation, not all of them. Browse every skill.