Dev Diary -> MDX

Turns a Millennium Dawn dev diary .docx into a publish-ready .mdx for the docs site. The job is structure, not editing: scaffold the MDX and place the images, but leave the author's prose as written.

Core rule: preserve the author's voice (read this first)

Dev diaries are written in a deliberate, often in-character voice. The rule most likely to be broken is "improving" the prose. Do not.

• Do NOT proofread, normalize, sanitize, reword, reorder, or delete text.
• Keep casual spellings ("breakin", "gonna", "mayhaps"), run-on sentences, asides, mild profanity/jokes, smart quotes, and the author's own spelling/hyphenation ("self sufficiency", "Self Government Act").
• Keep the author's stylistic ellipses (…., ...., …..) exactly as written. These are voice, not AI ellipsis abuse. Do not collapse them.
• The only allowed change to source text is splitting a paragraph at a sentence boundary so an image can sit beside the passage it illustrates.

Anti-AI-tell exceptions (still apply, scoped narrowly)

The repo's no-em-dash / no-ellipsis-abuse style rules also keep the published page from looking AI-generated, so:

• Text you author (section headers, image alt text, the seam where you split a paragraph): never use an em dash (—) or ... ellipsis abuse, and never generic AI phrasing.
• Author's source prose: convert any em dash (—) per the loc rule (period, comma, or colon as fits) — em dashes are a strong AI tell and the author rarely intends one. But keep their stylistic ellipses (above) untouched.

If the user explicitly asks for proofreading, that is a separate, opt-in task: do it then, and list every change so they can revert.

Step 1 — Get the source .docx

If $ARGUMENTS (or an attached file) already points to a .docx, use it. Otherwise ask the user to attach or give the path to the .docx before doing anything else. If they have a Google Doc, tell them to use Download -> Microsoft Word (.docx) — that export is the same format. Reject non-.docx input.

Step 2 — Determine the next diary number

List docs/src/content/devDiaries/*.mdx, take the highest NNN- prefix, add 1, and zero-pad to 3 digits (e.g. 057 -> 058). This NNN is reused for both the .mdx filename and the image folder. Warn if docs/src/assets/images/dev-diaries/<NNN>/ already exists.

Step 3 — Extract text and images

Run the bundled helper (stdlib-only, no deps). From the repo root:

python .claude/skills/dev-diary-mdx/extract_docx.py \
  --docx "<path to .docx>" \
  --image-dir docs/src/assets/images/dev-diaries/<NNN>

It creates the image folder, copies every embedded image — in document order — to picture-1.png, picture-2.png, … (extensions preserved), and prints JSON:

{ "title": "…|null",
  "blocks": [ {"type":"heading","level":2|3,"text":"…"},
              {"type":"para","text":"…"},
              {"type":"image","index":N} ],
  "images": ["picture-1.png", …] }

The JSON text is verbatim UTF-8. Treat the blocks order as the source order.

Step 4 — Collect frontmatter (ask the user)

Prompt for the fields that can't be reliably derived, offering defaults:

• title — default to the JSON title or the doc's first heading. Final frontmatter form is "Dev Diary #NN: <Title>" (NN = number without padding, e.g. #58).
• description — one line.
• author — the real developer handle (e.g. Luigi). NOT the in-character narrator ("Luigi IV von Limingly").
• version — default v2.0. This is its own frontmatter field (not a tag) and is schema-required: it must match ^v\d+\.\d+$ (see docs/src/content.config.ts), or bun run check fails.
• date — default today, format YYYY-MM-DD.

Derive the slug by kebab-casing the title without the "Dev Diary #NN:" prefix (e.g. Greenland: Holiday Paradise -> greenland-holiday-paradise). Build permalink: /dev-diaries/<N>-<slug>/ using the unpadded diary number (/dev-diaries/58-greenland-holiday-paradise/, matching the #NN in the title). The .mdx filename keeps the zero-padded <NNN>-<slug>.mdx form.

Step 5 — Assemble the .mdx

Write docs/src/content/devDiaries/<NNN>-<slug>.mdx, UTF-8 without BOM (it is not a loc file), in this exact shape:

---
title: "Dev Diary #58: Greenland"
description: <one line>
permalink: /dev-diaries/58-greenland-holiday-paradise/
author: Luigi
date: 2026-06-12
version: "v2.0"
tags:
  - dev diary
---

_By Luigi – 12 June 2026_

<body>

Body rules:

• No # H1. The title lives only in frontmatter; the body opens with the italic byline _By <Author> – <DD Month YYYY>_.
• Headings: keep the narrative intro and the sign-off header-less. Insert ## (or ### for sub-topics) only at genuine topic shifts. If the docx had heading styles, honor them; if it was flat prose (common — the author often writes one long stream), add headers yourself using the diary's own wording (typical focus-tree diary: ## The National Situation, ## The National Focus Tree, ## Decisions and Policies).
• Every extracted image goes into the body — always place the full content. No image is dropped. Image placement prefers the text's reading-order cues over the raw block position. Word docs frequently dump every screenshot in one cluster at the top; the blocks array reflects that. The index in each image block is the stable filename number (picture-<index>.png). Place each image immediately after the sentence/paragraph that introduces it, following the prose cues:
  • "The first picture shows …" -> picture-1 after that sentence.
  • "The next pictures in order are the economic tree and political trees" -> place those consecutively right after that sentence.
  • When two visuals are described in separate passages of one paragraph (e.g. the military tree, then "Moving to the GIS tree…"), split the paragraph at that boundary so each image sits with its own passage.
  • Match the repo's reference form exactly, with an absolute path (leading /): ![picture1](/assets/images/dev-diaries/<NNN>/picture-1.png) — alt text ![pictureN]. (Descriptive alt text is more accessible; offer it as an option but default to ![pictureN].)
  • If the prose gives no cue for an image (e.g. a Decisions screenshot the author never calls out), do not drop it — place it at the position it occupies in the document, i.e. after the body text that corresponds to the block immediately preceding that image in the extractor's blocks array. Then note it in Step 6 so the user can move it if they prefer.

Step 6 — Report and flag for confirmation

Tell the user, concisely:

• the new .mdx path and the image folder;
• image count — how many were extracted and placed (all of them), and which ones had no prose cue and were placed by document-position fallback (so they can move them);
• a reminder to eyeball the frontmatter (author = real handle, version tag, date) against an existing diary;
• suggested checks from docs/: bun run lint:md, bun run lint:remark, and prettier --write src/content/devDiaries/<NNN>-<slug>.mdx, plus bun run check to validate the content-collection frontmatter schema.