Agent Proof Transcripts

June 3, 2026 ยท View on GitHub

Use these pages when a reviewer asks what a successful Bilig session actually looks like. They are terminal transcripts, not screenshots and not UI recordings. The shared proof comes from a public no-key run:

npm exec --yes --package @bilig/workpaper@latest -- bilig-evaluate --door agent-mcp --json

Observed package versions in the transcript:

{
  "@bilig/workpaper": "0.157.0",
  "xlsx-formula-recalc": "0.157.0"
}

The important proof shape is the same for every agent host:

StepEvidence
PromptAsk the agent to prove a workbook-shaped formula edit before driving Excel, LibreOffice, Google Sheets, or a browser grid.
Tool callRun bilig-evaluate --door agent-mcp --json or the local MCP equivalent.
ResultReturn schemaVersion: "bilig-evaluator.v1", door: "agent-mcp", and verified: true.
Workbook state changeInputs!B3 changes the dependent Summary!B3 value from 60000 to 96000.
Formula readbackafter, afterRestore, and afterRestart all read 96000.
JSON exportpersistedDocumentBytes is 1162, and the exported WorkPaper document is restored.
Restart readback verificationrestartReadbackMatchesAfter: true.

Pick the host transcript that matches the agent you are using:

Shared Verified Output

This is the compact output every transcript checks:

{
  "schemaVersion": "bilig-evaluator.v1",
  "door": "agent-mcp",
  "packageVersions": {
    "@bilig/workpaper": "0.157.0",
    "xlsx-formula-recalc": "0.157.0"
  },
  "evidence": {
    "editedCell": "Inputs!B3",
    "dependentCell": "Summary!B3",
    "before": 60000,
    "after": 96000,
    "afterRestore": 96000,
    "afterRestart": 96000,
    "persistedDocumentBytes": 1162,
    "toolCount": 8,
    "checks": {
      "listedFileBackedTools": true,
      "listedResourcesAndPrompts": true,
      "formulaValidationPassed": true,
      "dependentCellChanged": true,
      "persistedToDisk": true,
      "exportContainsWorkPaperDocument": true,
      "restartReadbackMatchesAfter": true,
      "displayValueRead": true
    }
  },
  "verified": true
}

What This Proves

The transcript proves that a coding agent can discover the WorkPaper MCP tool surface, write a cell, read a dependent formula, export JSON, restore the document, and verify restart readback.

It does not prove Excel desktop UI automation, macro execution, pivot refresh, chart layout, private workbook compatibility, or hosted multi-user storage.