Field notes · AI agents, Hermes & Claude Code By Julian Goldie

Hermes Agent Microsoft Office: Set Up The Office Skill For Word, PowerPoint And Excel In 7 Steps (2026)

One bundled skill, three file formats, zero Office installs. Here is the exact setup order, the helper map and the gotchas that corrupt files.

Get The AI Profit Stack Join AIPB →
  • 01 1,000+ done-for-you AI agent workflows
  • 02 5 live coaching calls / week with me
  • 03 7-day refund, cancel anytime
  • 04 3,400+ AI operators inside

Hermes agent microsoft office work runs through one bundled skill that writes real .docx, .pptx and .xlsx files from JSON specs, with no copy of Word, PowerPoint or Excel anywhere on the box.

Here is how I would set it up on a fresh machine, in seven steps, using only the commands and file names from the official Microsoft Office skill page.

The example running through every step is a small bookkeeping firm that sends a monthly client letter, a quarterly review deck and a cash-flow workbook.

If you want my team to look over your own document setup before you build it, book a free SEO strategy session.

Why This Skill Works On Files, Not On The Office Apps

The skill sits at skills/productivity/microsoft-office and ships installed by default.

Its job is to create, read, edit and template Word, PowerPoint and Excel files using three Python libraries: python-docx, python-pptx and openpyxl.

The page states it plainly: Office itself is never needed.

That changes how you plan the setup.

You are not wiring Hermes into a desktop app or into Microsoft 365 online; you are giving it a set of small command-line helpers that build and check files on disk.

So it runs the same way on a Linux server, a Mac or a Windows host, which are the three platforms the skill metadata lists.

The metadata also shows version 2.0.0, Nous Research as author, an MIT licence and pdf as the related skill.

Step 1: Get On v0.23.0 And Clear Out The Old Format Skills

The v0.23.0 release notes (5 November 2026) merged the separate Word, PowerPoint and Excel skills into the single microsoft-office skill.

The same release trimmed a fresh install from 58 built-in skills to 44 and moved twelve niche skills to optional installs.

If you upgraded an older install, the old per-format copies are still sitting on disk until you remove them yourself.

Delete them before the first real job, so the agent never picks an older copy for the bookkeeping letter.

Then open the bundled skills catalog and confirm microsoft-office is listed under productivity.

A patch release, v0.23.1, followed on 9 November 2026 and rolls up roughly 618 merged PRs, so update to that if you can.

Step 2: Lock The Skill Once You Have Tuned It

The same release added hermes curator lock.

It stops the agent from rewriting a skill you have curated.

Once you have settled how the firm wants its letters and decks built, lock the skill so the house rules stay put:

hermes curator lock microsoft-office

Skip this on day one while you are still experimenting, then run it once the outputs look right.

Step 3: Build The Documents Environment

Python 3.10+ is the floor, plus python-docx, python-pptx and openpyxl (imported as docx, pptx and openpyxl).

You do not install them by hand into Hermes.

When the hermes command is on PATH, the skill builds a shared, pinned documents environment with this command:

hermes pm environment documents --extra documents --extra documents-pdf

The docs say that environment lives under the Hermes home, survives container restarts, leaves the Hermes install alone and is safe for several agents to use at once.

If the terminal runs inside a Docker sandbox with no Hermes on PATH, the skill builds its own venv in the sandbox home with the same pinned versions.

And if plain python already imports the libraries, as it does in the Hermes Docker image, the skill just uses that.

One rule to tape to the monitor: never pip install into the interpreter that runs Hermes or into the documents environment, because Hermes owns both.

🔥 Want this set up the way I run my own agents? Inside the AI Profit Boardroom you get step-by-step video trainings, the installable Agent OS, four coaching calls a week and 3,400+ members building the same kind of document workflows. → Join the AI Profit Boardroom

Step 4: Add LibreOffice And Poppler Only If You Need Them

LibreOffice (soffice) is optional.

It covers three jobs: PDF export, Excel recalculation and slide rendering.

Poppler (pdftoppm or pdftocairo) adds slide PNGs on top.

For the bookkeeping firm, the client letter goes out as a PDF and the review deck gets checked as images, so both get installed.

Check what is already there before you install anything:

command -v soffice || command -v libreoffice

The helpers that need these tools report what is missing and exit 0, so a missing install shows up as a clear message rather than a crash.

There is no pure-Python route to PDF for any of the three formats, and the skill is written to say so rather than fake one.

Step 5: Write The First Word Document From A Spec

Creation is spec-driven: the agent writes a JSON spec with write_file, then runs the create helper from the skill directory.

The docs run every helper through the interpreter the setup step picked, stored in a variable called PY.

python scripts/docx_create.py spec.json out.docx
python scripts/docx_read.py --text out.docx

Before it writes the spec, the agent loads references/word.md with skill_view; that reference covers reports, letters, contracts, token templates, tracked changes and comments.

For a letter the firm reuses every month, a template is the better route: docx_template.py fills token placeholders in an existing Word file.

Always pass an output path where the helper supports one, so the original template is never overwritten.

Step 6: Build The Review Deck And Render It

Decks follow the same pattern with pptx_create.py, or pptx_from_template.py when the firm has a brand template.

The PowerPoint reference lists pitch decks, brand-template decks, chart refreshes, speaker notes and slide PNGs as typical jobs.

python scripts/pptx_create.py deck.json out.pptx
python scripts/pptx_read.py --outline out.pptx
python scripts/pptx_render.py out.pptx

The verification step in the docs is to render the deck and check each PNG with vision_analyze when the render tools exist.

Do that every time, because fonts missing on the host get substituted and can push text out of place.

Step 7: Build The Workbook And Recalculate

The Excel reference covers styled multi-sheet workbooks, formulas, charts, CSV conversion and row or column restructures.

The cash-flow workbook starts from a spec, or from an export the firm already has via csv_to_xlsx.py.

python scripts/xlsx_create.py spec.json report.xlsx
python scripts/xlsx_read.py --formulas report.xlsx
python scripts/xlsx_recalc.py report.xlsx

Here is the gotcha most people hit: nothing in the skill computes formulas.

Excel fills cached values when the file opens, or xlsx_recalc.py does it, and that helper needs LibreOffice.

If the firm emails the workbook straight to a client who previews it in a browser, run the recalc first.

The Helper Map I Keep Open

JobWordPowerPointExcel
Create from JSON specdocx_create.pypptx_create.pyxlsx_create.py, csv_to_xlsx.py
Read backdocx_read.py --text/--structurepptx_read.py --outlinexlsx_read.py --sheets/--json
Editdocx_edit.pypptx_edit.pyxlsx_edit.py, xlsx_restructure.py
Fill a templatedocx_template.pypptx_from_template.pynone listed
Review or checkdocx_revisions.py, docx_comments.py, docx_validate.pypptx_render.pyxlsx_recalc.py, xlsx_to_csv.py

Every helper is an argparse CLI with --help, prints JSON to stdout and exits non-zero on failure.

That last part is what lets the agent catch its own mistakes without you reading every file.

Converting Old .doc, .ppt And .xls Files

The skill does not open legacy binary formats.

If the firm still has an old .doc letterhead, convert it first, then point the agent at the new copy:

soffice --headless --convert-to docx old-letterhead.doc

Swap docx for pptx or xlsx for the other two formats.

OpenDocument files (.odt, .odp, .ods) and WYSIWYG layout work are outside the skill altogether.

For PDF export the docs give soffice --headless --convert-to pdf --outdir out/ file.docx, which also takes .pptx and .xlsx.

📞 Want a second pair of eyes on your document pipeline? Book a free SEO strategy session and my team will walk through where an agent like this fits around the rest of your work. → Book the free call

Gotchas That Corrupt Files Or Waste A Run

  • Never unzip and edit the XML. Office files are zip packages and raw text swaps break them; use the helpers.
  • Text is split into runs. A replace that crosses differently formatted text takes the first run's formatting.
  • Word fields stay empty. Tables of contents and page fields fill in when Word opens the file, not before.
  • Specs must be UTF-8. The helpers read UTF-8 explicitly.
  • Validate after surgery. docx_validate.py should exit 0 after any revision, comment or field change.

If the agent edits a client contract with tracked changes, the deeper notes live in references/revisions-and-comments.md.

For workbooks where rows move around, references/restructuring.md covers how references shift.

Sending The Finished Files Back Through Chat

If you brief Hermes from Slack, Telegram, Discord, WhatsApp or Signal, you do not need to fetch files by hand.

The Deliverable Mode page says the gateway picks file paths out of the reply and uploads the files as native attachments, and it names this skill as the source of Word docs, spreadsheets and decks.

So the bookkeeper can ask for the cash-flow workbook from a phone and get the .xlsx back in the same thread.

If you want more bundled skills to pair with this one, the best Hermes Agent skills list is the next stop.

One Monthly Run, Start To Finish

Put the steps together and a month at the bookkeeping firm looks like this.

On the first working day, someone messages the agent with the client list and asks for this month's letters.

The agent loads the Word reference, fills the locked letter template once per client with docx_template.py, and re-reads each file with docx_read.py --text to confirm the client name and dates landed.

It converts each letter to PDF with soffice and lists the output paths in its reply, so the gateway attaches them.

At quarter end, it builds the review deck from the brand template, renders every slide and checks the images before anyone sees them.

The workbook gets a recalc pass so the totals show as values, not empty cells.

Nobody opens Word, PowerPoint or Excel at any point in that run.

A person still reads the final files before they go to clients, which is where the approval belongs.

My Verdict On The Setup

Seven steps, and most of them take minutes.

The two that matter most are clearing the old per-format skills after an upgrade and installing LibreOffice if anything you send needs to be a PDF or have live formula values.

Get those right and the skill does what it promises: real Office files, built and checked offline.

If you would rather have a coach walk you through it live, the AI Profit Boardroom runs four coaching calls a week.

And to talk through a bigger rollout with my team, book a free SEO strategy session.

Also On Our Network

FAQ: Hermes Agent Microsoft Office

Is the Hermes Agent Microsoft Office skill installed by default?

Yes. It is a bundled skill at skills/productivity/microsoft-office, so a normal install already has it.

Do I need Word, PowerPoint or Excel on the machine?

No. The skill uses python-docx, python-pptx and openpyxl offline. LibreOffice is only needed for PDF export, recalculation and slide rendering.

What happened to the separate Word, PowerPoint and Excel skills?

v0.23.0, released 5 November 2026, merged them into one microsoft-office skill. Old copies stay on disk after an upgrade until you remove them.

How do I open an old .doc or .xls file?

Convert it first with soffice --headless --convert-to docx, pptx or xlsx, then let the skill work on the converted copy.

How do I check a deck the agent built?

Render it with pptx_render.py to get slide PNGs, then have the agent check each image with vision_analyze. Poppler is needed for the PNGs.

Can it fill my existing Word letterhead?

Yes. docx_template.py fills token placeholders in a Word template, and pptx_from_template.py does the same job for brand-template decks.

📺 Video notes + links to the tools 👉

🎥 Learn how I make these videos 👉

🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉

About Julian

I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.

I help business owners scale with AI agents, automation and SEO.

  • I've built a 7-figure SEO and link building agency, Goldie Agency, with a team of 70+ SEO experts.
  • I've grown a YouTube channel to 400,000+ subscribers.
  • I wrote "Link Building Mastery".
  • My Udemy courses have taught over 29,000 students.

→ Get my best AI training inside the AI Profit Boardroom

Clear the old skills, build the documents environment, add LibreOffice if you need PDFs, and if you want my team to check the plan, book a free SEO strategy session and bring your hermes agent microsoft office setup.

§ Field reports

Real wins from inside the AI Profit Boardroom

See all 3,400+ members →
AIPB member win screenshot AIPB member win screenshot AIPB member win screenshot AIPB member win screenshot AIPB member win screenshot AIPB member win screenshot AIPB member win screenshot AIPB member win screenshot AIPB member win screenshot AIPB member win screenshot AIPB member win screenshot AIPB member win screenshot

Scroll sideways →

⚑ Next step

Ready to run AI agents that actually do the work?

Join 3,400+ operators inside the AI Profit Boardroom. Get 1,000+ plug-and-play AI agent workflows, 5 coaching calls a week, and a community that holds you accountable.

Join The AI Agent Community →

7-day no-questions refund · Cancel anytime

Found this useful? Get more of this site in your Google.

One tap adds us as a preferred source — Google then shows you our posts more often in Search, Top Stories, and AI Overviews.

← Back to all posts