Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use formula-backed WorkPaper JSON and MCP tools for agent spreadsheet tasks without driving Excel or a browser UI.
.claude/skills/bilig-workpaper/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-19 | ✗→✓ | ▲ Improved | — | — |
| case-04 | ✗→✓ | ▲ Improved | — | — |
| case-02 | ✗→✓ | ▲ Improved | — | — |
| case-05 | ✗→✓ | ▲ Improved | — | — |
| case-14 | ✗→✓ | ▲ Improved | — | — |
Bilig WorkPaper gives agents a code-first workbook runtime for spreadsheet-style business logic. Use it when the task is easier to model as sheets and formulas, but the reliable path is to edit cells through an API, recalculate, read computed values back, and persist a JSON workbook document.
The main use case is replacing fragile spreadsheet UI automation with deterministic tool calls. It is useful for quote calculators, payout models, budget checks, import validation, and reduced XLSX formula bug reports.
Use this skill when the user needs to:
Do not use it for manual spreadsheet editing, VBA/macros, pivots, charts, COM automation, or exact desktop Excel behavior unless the user explicitly asks to compare against Excel as an oracle.
Prefer argument arrays in MCP/client configuration. Do not shell-concatenate user-provided paths, sheet names, formulas, or cell addresses. Reject path or cell input containing newlines, backticks, $(, ;, &, |, <, or > before using it in a command.
The MCP examples execute the public @bilig/workpaper npm package. Treat that as third-party code execution: pin the package version you reviewed, run it only in a trusted project, and get explicit user approval before starting a writable MCP server.
First prove the package-owned challenge works:
json{ "command": "npm", "args": ["exec", "--package", "@bilig/workpaper@<reviewed-version>", "--", "bilig-mcp-challenge"] }
Then run a writable file-backed MCP server:
json{ "command": "npm", "args": [ "exec", "--package", "@bilig/workpaper@<reviewed-version>", "--", "bilig-workpaper-mcp", "--workpaper", "./pricing.workpaper.json", "--init-demo-workpaper", "--writable" ] }
Useful tools exposed by the MCP server:
list_sheetsread_rangeread_cellset_cell_contentsget_cell_display_valueexport_workpaper_documentvalidate_formulaAfter every write, read the dependent output cell and export the WorkPaper document. Do not claim success from the write call alone.
Use the package directly when workbook logic belongs inside application code:
tsimport { WorkPaper, exportWorkPaperDocument, serializeWorkPaperDocument, } from "@bilig/workpaper"; const workbook = WorkPaper.buildFromSheets({ Inputs: [ ["Metric", "Value"], ["Customers", 20], ["Average revenue", 1200], ], Summary: [ ["Metric", "Value"], ["Revenue", "=Inputs!B2*Inputs!B3"], ], }); const inputs = workbook.getSheetId("Inputs"); const summary = workbook.getSheetId("Summary"); if (inputs === undefined || summary === undefined) { throw new Error("Workbook is missing required sheets"); } workbook.setCellContents({ sheet: inputs, row: 1, col: 1 }, 32); const revenue = workbook.getCellDisplayValue({ sheet: summary, row: 1, col: 1 }); const saved = serializeWorkPaperDocument( exportWorkPaperDocument(workbook, { includeConfig: true }), ); console.log({ revenue, savedBytes: saved.length });
A good agent response should include:
If any proof step fails, report the blocker instead of saying the workbook was updated.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-19 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-04 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-02 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-21 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-05 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-14 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-17 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-10 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-22 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-13 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-03 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-11 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-06 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-16 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-18 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-15 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-08 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-20 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-01 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-07 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-09 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-12 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 20 counted toward the lift figure. The other 2 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +77 percentage points is the difference between those two pass rates over the 20 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
The per-case answers from this run were removed by the retention sweep, so the case table below shows the verdicts without the text either arm produced. The counts above were recorded at the time and are unaffected. Answers are now kept for 180 days.
Other measured skills in the registry, with their headline benchmark lift.