changhochien xlsx-author
Use this skill when running headless (managed-agent / CMA mode) and you need to deliver an Excel workbook as a file artifact rather than editing a live workbook via mcp__office__excel_*.
Output contract
- Write to
./out/<name>.xlsx. Create./out/if it does not exist. - Return the relative path in your final message so the orchestration layer can collect it.
How to build the workbook
Write a short Python script and run it with Bash. Use openpyxl:
from openpyxl import Workbook
from openpyxl.styles import Font, PatternFill
wb = Workbook()
ws = wb.active; ws.title = "Inputs"
ws["B2"] = "Revenue"; ws["C2"] = 1_250_000_000
ws["C2"].font = Font(color="0000FF") # blue = hardcoded input
calc = wb.create_sheet("DCF")
calc["C5"] = "=Inputs!C2*(1+Inputs!C3)" # black = formula
wb.save("./out/model.xlsx")
Conventions (mirror audit-xls)
- Blue / black / green. Blue = hardcoded input, black = formula, green = link to another sheet/file.
- No hardcodes in calc cells. Every calculation cell is a formula; every input lives on an Inputs tab.
- Named ranges for any value referenced from a deck or memo.
- Balance checks. Include a Checks tab that ties (BS balances, CF ties to cash, etc.) and surfaces TRUE/FALSE.
- One model per file. Do not append to an existing workbook unless explicitly asked.
When NOT to use
If mcp__office__excel_* tools are available (Cowork plugin mode), use those instead — they drive the user's live workbook with review checkpoints. This skill is the file-producing fallback for headless runs.
Hermes Profile Notes
This skill was packaged from pi-financial-services source path plugins/vertical-plugins/financial-analysis/skills/xlsx-author for the Hermes financial-services profile. Use institutional data connectors first when available, cite sources, and stage outputs for qualified human review.
Common Pitfalls
- Do not present drafts as investment, legal, tax, or accounting advice.
- Do not use web search as the primary source when an institutional MCP/data connector is available.
- Do not execute transactions, contact clients, post to a ledger, or approve onboarding. Stage outputs for review.
Verification Checklist
- Inputs, assumptions, dates, and currencies are explicit.
- Numbers tie across tables, models, decks, and memos.
- Sources are cited and institutional data is preferred where available.
- Output is labeled draft / for human review where appropriate.