MaybeAI Sheet

MaybeAI Sheet

Use this skill for MaybeAI spreadsheet lifecycle work: upload or import files, profile workbook contents, inspect worksheets, read and write data, manage worksheets, run formulas, trace formula lineage, build SQL result sheets, apply lightweight formatting, share, and export.

Do not use this skill as the primary workflow for chart authoring or dashboard composition. For dashboard-first work, use sheet-dashboard.

Quick Start

Required environment:

• MAYBEAI_API_TOKEN

Common script prerequisites:

• curl
• jq

Base URL:

https://play-be.omnimcp.ai

Authorization header:

Authorization: Bearer 

Ready-to-run examples live in scripts/:

bash scripts/01-file-management.sh
bash scripts/02-read-data.sh
bash scripts/03-write-data.sh
bash scripts/04-rows-columns.sh
bash scripts/05-worksheets.sh
bash scripts/06-formulas.sh
bash scripts/07-charts-pictures.sh
bash scripts/08-formatting.sh
bash scripts/09-end-to-end.sh
bash scripts/10-workbook-profile.sh
bash scripts/11-lineage-trace.sh

When To Use Which Path

Core Rules

1. Choose the worksheet before you read or write

Do not rely on defaults for non-first worksheets.

• Prefer worksheet_name when the endpoint supports it
• Some endpoints require ?gid= in uri
• If you pass neither, many calls will land on the first worksheet, which causes classic “wrote to the wrong sheet” failures

See references/read-write.md for details.

2. Choose the right engine before importing large files

MaybeAI Sheet routes /api/v1/excel/* through Playground, then into either excelize-mcp or SheetTable. Engine choice is a Playground routing decision.

• Use SheetTable/PG for table-like workbooks when any worksheet has more than 10,000 rows, the workbook has more than 100,000 populated cells, the data is mostly flat records, or the workflow is SQL-heavy / append-upsert-heavy.
• Use Excelize for workbook-style files where preserving Excel layout, styles, formulas, merged cells, and workbook semantics matters more than table scale.
• For large table-like .xlsx uploads, prefer POST /api/v1/excel/import with multipart engine=postgres; avoid /api/v1/excel/upload for those large-data cases.
• After import, verify list_worksheets reports engine: "pg" or worksheet data_engine: "pg".

See references/file-management.md for the exact request.

3. Prefer high-level write APIs before raw A1 writes

Priority order:

1. update_data_keep_headers
2. update_range_by_lookup
3. append_rows
4. update_range

Meaning:

• Use update_data_keep_headers for full-table replacement while preserving headers and column order
• Use update_range_by_lookup for key-based updates with automatic append of new rows
• Use append_rows for simple object-row appends
• Use update_range only when you must target an exact A1 range or non-tabular cells

update_range defaults to RAW: numeric-looking strings such as "5.53%" and "9,007,000" remain strings. Use value_input_option=USER_ENTERED only when the user intentionally wants Excel-like parsing of formulas, dates, numbers, and percentages. Check the write response message: parse_result=NOT_REQUESTED means RAW kept strings as text and lists them in preserved_values; parse_result=PASS means USER_ENTERED parsed values listed in parsed_values; parse_result=PARTIAL means values listed in parsed_values parsed, while preserved_text_values may stay text unless target cells are numeric-formatted.

4. Separate data writes from style writes

Do not assume write_new_worksheet, update_range, or sql/write_result will automatically apply formatting.

If the user wants a readable report or manager-facing table:

1. Write the data first
2. Call freeze_panes
3. Call batch_set_cell_style
4. Optionally set column widths, row heights, and autofilter

See references/charts-formatting.md for the style playbook.

5. Compile SQL before writing a result sheet

Default SQL result-table flow:

1. read_headers
2. Optionally read_sheet for sampling
3. sql/compile
4. sql/write_result
5. read_sheet to verify the result

Do not skip sql/compile.

6. Always read back after writing

Do at least one of the following:

• read_sheet
• list_worksheets
• read_headers
• Export the file and inspect it manually

Verification is especially required after:

• SQL result-table writes
• Writes to non-first worksheets
• Overwrite flows that preserve formulas or styles
• Chart, image, or style changes

7. Use workbook profile for broad workbook understanding

Use POST /api/v1/excel/workbook_profile when the user asks what a workbook contains, which sheets are relevant, or where to start an analysis.

• It returns a cached LLM-generated workbook summary plus worksheet sample rows
• It is read-only and requires viewer access
• Use it before manual read_sheet calls for unfamiliar multi-sheet workbooks
• Use force_refresh: true after worksheet structure changes or when cached semantics may be stale

See references/workbook-profile.md for request and response details.

8. Use lineage trace to explain computed cell dependencies

Use POST /api/v1/excel/lineage/trace when the user asks where a formula result comes from, why a computed value is wrong, or which source columns feed a target cell.

• It is read-only and requires viewer access
• It traces ordinary Excel formulas, cross-sheet references, column-like formulas, and supported =SQL(...) formula outputs
• Prefer format: "tree" when explaining to a user
• Prefer format: "node" when a UI or downstream tool needs graph nodes and edges

See references/lineage-trace.md for request and response details.

9. Use clickable references only for confirmed workbook locations

When the final answer references real cells, ranges, or worksheets from the current Maybe Sheet workbook, follow references/clickable-refs.md so the frontend can make those references clickable.

• Use clickable references only for confirmed current-workbook locations
• Include the current workbook document ID in the docId attribute
• Include the target worksheet gid in the gid attribute
• Do not use clickable references for examples, guesses, inferred locations, or uncertain references
• When referencing workbook cells, ranges, or table names, always use ... tags. Do not use conversational location descriptions such as "above", "the cell above", "D column row fourteen", or "D 列第十四行".
• Always use paired sheet-ref tags with visible text, never self-closing tags. Use Sheet1!A1, not ``.
• Do not place clickable references inside code blocks, formulas, SQL, JSON, or shell examples

10. Use the share API family for visibility changes

When the user asks to make a Maybe Sheet public or private, use:

• POST /api/v1/share/sheet/visibility

Do not use:

• /api/v1/excel/spreadsheets/d/{id}/share

That spreadsheet-path share route is the wrong endpoint for MaybeAI Sheet visibility changes and returns 404. Use the share/sheet/* API family for visibility, permission updates, access removal, and share listing.

Agent-Safe Playbooks

Upload and inspect a file

1. Upload or import the file
2. Capture document_id and uri
3. list_worksheets
4. read_headers
5. Optionally read_sheet for a small sample

References:

• references/file-management.md
• references/read-write.md
• scripts/01-file-management.sh
• scripts/02-read-data.sh

Profile a workbook before analysis

1. Capture document_id or uri
2. Call workbook_profile
3. Use profile.summary to explain the workbook at a high level
4. Use profile.worksheets[].sample_rows to choose targeted read_headers or read_sheet calls
5. If the workbook changed materially, call again with force_refresh: true

References:

• references/workbook-profile.md
• scripts/10-workbook-profile.sh

Refresh a table while keeping headers and formulas

1. Identify the target worksheet
2. Use update_data_keep_headers
3. If the sheet has computed columns, set preserve_formulas: true
4. read_sheet to verify

References:

• references/read-write.md
• scripts/03-write-data.sh

Sync business records by key

1. Identify the key column, such as Order ID or SKU
2. Use update_range_by_lookup
3. If you need workbook-wide downstream recalculation, call recalculate_formulas
4. read_sheet to verify

References:

• references/read-write.md
• references/formulas-sql.md

Build an SQL result sheet

1. Identify the worksheet name or gid
2. read_headers
3. Write PostgreSQL-compatible worksheet SQL
4. sql/compile
5. sql/write_result
6. read_sheet to verify the output

References:

• references/formulas-sql.md
• scripts/06-formulas.sh

Persist many report formulas efficiently

1. Scaffold the destination worksheet first
2. Group derived cells into rectangular blocks
3. Use formula/batch_set for those blocks
4. Prefer recalculate_mode=workbook when later blocks depend on earlier blocks
5. Use formula/set only for sparse one-off fixes
6. read_sheet to verify representative cells

References:

• references/formulas-sql.md
• scripts/06-formulas.sh

Trace formula lineage for a cell

1. Capture document_id or uri
2. Identify the target worksheet by worksheet_name or gid
3. Call lineage/trace with the exact A1 target cell
4. Use tree output to explain the dependency chain, or node output to render a graph
5. If lineage is incomplete, read the referenced sheets and headers to verify unsupported or unresolved references

References:

• references/lineage-trace.md
• scripts/11-lineage-trace.sh

Produce a readable report worksheet

1. Create the table with write_new_worksheet or a data-write API
2. freeze_panes
3. batch_set_cell_style
4. Optionally set widths, heights, and autofilter
5. If the response includes source_info.styles_ignored=true, explicitly tell the user the engine did not apply styles

References:

• references/charts-formatting.md
• references/errors-recovery.md
• scripts/08-formatting.sh

Reference Map

• references/file-management.md Best for upload, import, search, copy, sharing, export, and file-entry issues.
• references/workbook-profile.md Best for workbook semantic overview, sheet discovery before analysis, cached profile behavior, and workbook_profile request/response details.
• references/read-write.md Best for reading sheets, worksheet targeting, choosing the right write API, row and column operations, and worksheet management.
• references/formulas-sql.md Best for formulas, recalculation, SQL compile, and SQL result tables.
• references/sql-formula-showcase.md Best when you need a single live =SQL(...) formula that demonstrates joins, aggregation, Top N, and spill behavior.
• references/lineage-trace.md Best for tracing a target cell's formula dependencies, explaining source columns behind computed values, or producing lineage graph nodes and edges.
• references/charts-formatting.md Best for low-level chart APIs, pictures, freezing panes, styles, autofilter, and conditional formatting.
• references/errors-recovery.md Best for troubleshooting, limitations, and recovery paths.

What This Skill Intentionally Excludes

• It does not handle full dashboard composition, chart layout strategy, or infographic design; use sheet-dashboard for those.
• It does not turn the main SKILL.md into a full API manual; long request bodies and examples belong in references/ and scripts/.
• It does not default to large visual redesign or chart rearrangement unless the user explicitly asks for that.