# Microsoft Office Documents > Fill in and edit Word, Excel and PowerPoint files — letters, spreadsheets and slide decks — starting from an existing template. Read this before creating or changing any DOCX, XLSX or PPTX file. - Skill: `archmax-ai/microsoft-office-documents` (Agent Skill, multi-file: 4 files) - Install (CLI): `npx skillmds@latest add archmax-ai/microsoft-office-documents` - Raw SKILL.md: https://api.skillmd.com/api/skills/archmax-ai/microsoft-office-documents/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Data & Analytics - Author: archmax-ai (https://skillmd.com/u/archmax-ai) - Updated: 2026-09-22 - Page: https://skillmd.com/skills/archmax-ai/microsoft-office-documents --- # Microsoft Office documents An Office document is a zip container of XML parts. You never write the binary; you unpack the container, edit the parts as text, and pack it again. ## What you need This skill assumes the capabilities listed in its frontmatter. Check that you have each before you start; if one is missing, say which and stop rather than improvising. - **Unpack a container** — a tool that takes a file you hold and extracts every entry into a directory, preserving the entry paths inside the container (`word/document.xml`, `[Content_Types].xml`). Office files are zips, so any tool that unpacks a `.zip` unpacks a `.docx`. - **Pack a directory** — a tool that zips a directory back into one file, keeping paths relative to the directory and the file extension you give the result. This is what turns the edited directory into a valid `.docx`, `.xlsx` or `.pptx`. - **Read and edit files as text** — list a directory, read a file, search files for a string, and replace one exact string with another in a file. Writing a whole new file is needed only when adding a part, such as a new sheet. - **Move or copy a file** — to copy a template to the name your document should have before you unpack it, and for images: placing a picture you hold into the unpacked directory as a media part, or swapping one media part for another. A text editor cannot do the latter; image bytes are never read or written as text. - **Render a document as text** — a tool that shows a document's prose and tables as markdown. Use it to read; it never edits. - **Deliver the finished file** — a tool that sends, stores or forwards a file you hold, or turns it into a download link. Without one the document never leaves your working directory. Files are referred to by the paths your file tools report, relative to your working directory. Never paste a file's bytes into a tool argument. ## Core rules 1. **Never write a `.docx`, `.xlsx` or `.pptx` from scratch.** Derive it from a template: a document attached to this conversation, one in a directory you were given, one you fetched from a link, or — when nobody supplies one — a copy of the blank template that ships beside this file (see "Blank templates"). List and search your working directory before asking. Raw OOXML written by hand yields missing styles and files Office refuses to open. A supplied template beats a blank one whenever the document needs a house style, letterhead or logo, so if the request implies one and none is there, ask — and if nobody is watching this session, ask through the channel the request came from. 2. **Read as text, edit as parts.** Render the whole document to see what it says. Change it by unpacking, editing parts, packing. 3. **Nothing checks your XML.** A part that is no longer well-formed produces a document Office rejects, and no tool warns you. Keep every edit inside whole elements and escape text (see "XML must stay well-formed"). 4. **Deliver with a tool.** A packed document in your working directory reaches nobody by itself: attach it with the tool that sends messages, hand it to the tool that stores or forwards files, or create a download link when a person or system needs one. ## Blank templates Three empty but valid documents sit next to this file, for when the request supplies no template of its own: | File | What is in it | | --- | --- | | `template.docx` | one empty paragraph, A4 portrait, default Word styles | | `template.xlsx` | one empty sheet named `Tabelle1` | | `template.pptx` | one title slide (title and subtitle placeholders), 16:9, Office theme | They hold no content and no author metadata, so they are a clean start rather than someone else's document with the text taken out. **Copy one, never work on it in place.** Copy the file into your working directory under the name the finished document should have, then unpack that copy and edit it. Unpacking or packing over the template itself leaves the next document starting from your leftovers. A blank template carries no letterhead, logo or corporate styling. When the request wants those, it needs a real template — the blank one cannot stand in for it. ### Language The blank templates are set to German (`de-DE`). That setting does not translate anything; it decides the proofing language and the locale Word and PowerPoint assume, so leave it alone for a German document and change it for any other. Each format keeps it elsewhere: - **DOCX** — `w:val` on `` in `word/styles.xml` (the `docDefaults` run properties), and `` in `word/settings.xml`. Runs that carry their own `` override the default, so search the unpacked directory for `w:lang` rather than trusting the two above. - **PPTX** — a `lang="de-DE"` attribute on every ``, `` and ``, spread over `ppt/slides/`, `ppt/slideLayouts/` and `ppt/slideMasters/`. This is one of the rare edits where replacing all occurrences of `lang="de-DE"` in each part really is the same edit everywhere. - **XLSX** — cells have no language. What is German is the sheet name `Tabelle1`; rename it in `` in `xl/workbook.xml`, and change the matching `` in `docProps/app.xml` so the two agree. The same applies to a supplied template: it is written in some language already, and the document you produce should keep it unless the request says otherwise. ## Workflow: unpack → edit → pack 1. Unpack the template into a directory of its own on local disk, one per document, so two documents never mix parts (see "Work on local disk"). 2. List that directory to see the parts; search it for your placeholders (for example `[[`) to learn which part holds each one — a date may sit in the body and in a header. 3. Read the part you will change to see its actual XML around the placeholder. 4. Replace strings, one region per edit. 5. Pack the directory into a file whose name carries the right extension, then deliver it. Notes: - **Work on local disk.** Unpacking spreads one document into dozens or hundreds of small parts, and every edit and the final pack touches them again. A directory on remote storage — a synced folder such as OneDrive, SharePoint or Dropbox, or a mounted network share — makes that slow, and a sync that runs mid-edit can hand you a half-written part or pack a file that is still uploading. So unpack into a local scratch directory, do the work there, and copy only the finished document to wherever it belongs. If the only directory you can write to is remote, accept it and carry on there — it works, it is just slower and worth one extra read-back of the parts you changed before you pack. - Do not add or remove files in the unpacked directory unless you also update `[Content_Types].xml` and the relationship parts (see "Add a new sheet"). - Address `[Content_Types].xml` by its exact path with read and edit. Do not use a glob pattern for it: the brackets are a character class in glob syntax. - Word and Excel emit their XML as one very long line. Search first to locate the region, then replace with an `old` string that is unique and short. Replace one region per edit rather than rewriting a whole `document.xml` or `sheetN.xml`; a rewritten part is a part you had to reproduce perfectly. ## Reading a document 1. Render the container as markdown to confirm it is the right file and see its shape. 2. Unpack it and list the unpacked directory to see the parts. 3. Read the part you will change to see the raw XML. Rendering works on the container; on an unpacked XML part use the plain file reader. ## Editing Edit by exact-string replacement: an `old` string that occurs once, a `new` string that replaces it. Replace all occurrences only when every occurrence really is the same edit. ### XML must stay well-formed The usual ways an edit breaks a part: - literal `&`, `<`, `>` in text → write `&`, `<`, `>`; - HTML entities like ` ` → XML has no DTD; use numeric references (` `) or a space; - an `old` string that cuts through a tag (ends inside ``) while `new` does not restore it → extend `old` to whole `…` or `…` elements; - a `new` string that opens ``, ``, `` (or ``, ``) without closing it. After a substantial edit, read the changed region back and check that every element you opened is closed before you pack. A `.docx` has several content parts (body, headers, footers, footnotes); an `.xlsx` has one `xl/worksheets/sheetN.xml` per sheet plus `xl/sharedStrings.xml`. Edits are scoped to the part you address, so a placeholder that recurs is edited in each part; a search over the unpacked directory shows every part containing it. ## DOCX ### How replacements work Use the **literal placeholder text** (`[[Datum]]`, `[[Name]]`) as the `old` string and target the body part, `word/document.xml`. Word may split a placeholder across several ``/`` runs (spell-check and revision tracking do this), and then the literal will not match. When a replacement reports that the string was not found, search for a distinctive fragment (`[[Na`) to see the run-split XML, then use that XML as the `old` string — usually replacing the whole `…` sequence with a single run. **Fill a template:** unpack the letterhead template into its own directory; search that directory for `[[`; in its `word/document.xml` replace `[[Datum]]` with `06. März 2026` and `[[Name]]` with `Max Mustermann`; in its `word/header1.xml` replace `[[Date]]` with `06.03.2026`; pack the directory into a `.docx` with the name the request asks for. ### Paragraph-level replacement To insert paragraphs with specific formatting (bold, centred, a heading, a clause), make the `old` string cover the whole `…` you are replacing and the `new` string the new paragraph XML, for example: ```xml § 20 Dienstreisen(1) Flugreisen in der Business Class. ``` Read how the template formats its existing paragraphs first and model your XML on those patterns; a style referenced by `w:pStyle` must exist in `word/styles.xml`. ## XLSX ### Modify an existing sheet Unpack the workbook template into its own directory. Text cells usually point into `xl/sharedStrings.xml` (`t="s"` with an index), so replace `[[Title]]` there; numbers sit in the sheet part as `` values, so replace `0` in `xl/worksheets/sheet1.xml` — all occurrences only if every zero really becomes the same number. Pack the directory into an `.xlsx`. ### Add a new sheet A sheet is a new part and needs four coordinated changes inside the unpacked directory: 1. Read `xl/worksheets/sheet1.xml` to copy its shape, and write the new sheet as `xl/worksheets/sheet2.xml`. 2. In `xl/_rels/workbook.xml.rels`, add a `Relationship` for it with a new, unique `rId`. 3. In `xl/workbook.xml`, add a `` element whose `r:id` is that `rId` and whose `sheetId` is unique. 4. In `[Content_Types].xml`, replace `` with ``. ## Images Pictures are **media parts**: binary files under `word/media/` (Word), `xl/media/` (Excel) or `ppt/media/` (PowerPoint), reached from a content part through a relationship. Three things tie one together: - the media file itself, for example `word/media/image1.png`; - a `Relationship` in the content part's `.rels` file (`word/_rels/document.xml.rels`, `ppt/slides/_rels/slide1.xml.rels`) with `Type=".../relationships/image"`, a `Target` pointing at the media file, and an `Id` such as `rId5`; - a reference to that `Id` inside the content XML: `` in a `` (Word) or `` (PowerPoint). The file's extension must also be declared once in `[Content_Types].xml` as a `Default` (``); templates that already contain a picture of that type have it, templates that do not need it added. Never open a media file with a text tool: reading it shows noise and writing it corrupts it. Move or copy the file as bytes. ### Replace a picture, keep its place and size The simplest and most reliable operation. Find which media file the picture is: search the content part for `r:embed`, take the `rId`, look it up in the `.rels` file to get the `Target`. Then copy your new image over that media file **under the same name**. Nothing else changes: the relationship, the placement and the declared size stay as the template had them. A logo swap, a signature, a product photo in a fixed frame all work this way. The picture is drawn at the size the XML declares, not the image's own size, so a replacement with a different aspect ratio is stretched. Match the original's proportions, or adjust the extent (see below). ### Insert a new picture Adding a picture is a new part with the same four-part discipline as a new sheet: 1. Copy the image into the media directory under a new, unused name (`image7.png`). 2. In the content part's `.rels`, add `` with an `Id` unused in that file. 3. Ensure `[Content_Types].xml` has a `Default` for the extension. 4. In the content XML, insert a drawing that references the `Id`. Do not write one from memory: copy an existing `…` (or `…`) from the template, change its `r:embed` to the new `Id`, give its `wp:docPr` a new unique `id`, and set its size. If the template contains no picture at all, prefer asking for a template that does, or replacing a picture in a template that has one, over composing drawing XML by hand. ### Size and position Sizes are in EMUs: 914400 per inch, 360000 per centimetre. A picture 6 cm wide is `cx="2160000"`. Set the same values in both places Word keeps them — `` and `` inside the drawing — and keep the aspect ratio of the image. Inline pictures (``) flow with the text and are the easy case; anchored pictures (``) carry positioning that is best left as the template had it. ### Excel and PowerPoint In Excel a picture belongs to a drawing part (`xl/drawings/drawing1.xml`) that the sheet references through its own `.rels`; the media relationship lives in `xl/drawings/_rels/drawing1.xml.rels`. Replacing the media file by name works exactly as in Word. In PowerPoint each slide's `.rels` holds its own image relationships, so the same `rId` may mean different pictures on different slides — always read the `.rels` of the slide you are editing. ## PPTX Slides are `ppt/slides/slideN.xml`, each with `ppt/slides/_rels/slideN.xml.rels`, listed in `ppt/presentation.xml` and `ppt/_rels/presentation.xml.rels`. Replace text inside existing `` elements; adding a slide follows the same four-part recipe as a new sheet (part, relationships, presentation list, `[Content_Types].xml`).