Description
I maintain Bilig WorkPaper, so stating that upfront.
Would the Agent Framework maintainers want a small sample showing an agent using a formula workbook as deterministic business logic?
The problem: teams often keep pricing, payout, quote, or finance rules in spreadsheet formulas. Agents can call Excel/Sheets UI or read stale XLSX cached values, but neither is a great production workflow. A small example could show the safer pattern: write input cells, recalculate formulas, read verified outputs, and persist/export workbook state before the agent continues.
Proposed scope:
- use the published Bilig WorkPaper package / MCP tool surface;
- write one input cell in a pricing or payout workbook;
- read a recalculated formula output;
- return a compact proof object with JSON persistence/readback;
- keep it as a sample/cookbook only, not a first-party dependency or API change.
If this is too vendor-specific for the Agent Framework repo, I can keep it in Bilig-owned docs and link to Agent Framework only from there.
Code Sample
# Sketch only:
# agent calls a workbook tool with units/price/discount,
# tool writes inputs, recalculates, reads formula outputs,
# and returns net revenue plus persistence proof.
result = await workbook_tool.run({
"units": 80,
"unit_price": 1500,
"discount_rate": 0.2,
})
assert result["verified"] is True
Current package smoke proof:
npm exec --yes --package @bilig/workpaper@0.90.0 -- bilig-mcp-challenge --json
That command verifies MCP tool discovery, cell write, formula readback, JSON export, and restart readback. In the current package run, it edited Inputs!B3, changed dependent Summary!B3 from 60000 to 96000, persisted the WorkPaper JSON, and verified the restart readback.
Language/SDK
Both
Description
I maintain Bilig WorkPaper, so stating that upfront.
Would the Agent Framework maintainers want a small sample showing an agent using a formula workbook as deterministic business logic?
The problem: teams often keep pricing, payout, quote, or finance rules in spreadsheet formulas. Agents can call Excel/Sheets UI or read stale XLSX cached values, but neither is a great production workflow. A small example could show the safer pattern: write input cells, recalculate formulas, read verified outputs, and persist/export workbook state before the agent continues.
Proposed scope:
If this is too vendor-specific for the Agent Framework repo, I can keep it in Bilig-owned docs and link to Agent Framework only from there.
Code Sample
Current package smoke proof:
npm exec --yes --package @bilig/workpaper@0.90.0 -- bilig-mcp-challenge --jsonThat command verifies MCP tool discovery, cell write, formula readback, JSON export, and restart readback. In the current package run, it edited
Inputs!B3, changed dependentSummary!B3from60000to96000, persisted the WorkPaper JSON, and verified the restart readback.Language/SDK
Both