Setup guide · no coding required

Set up VS Code like a writer's binder.

You don't need to be a programmer. VS Code is a free, fast text editor — and a folder of Markdown files is all a novel needs. This guide walks you from a blank editor to a project layout where every Prose Minion tool knows exactly where your characters, chapters, and research live.

1 · open a folder 2 · build your binder 3 · one chapter, one file
my-novel — Explorer
my-novel
▸ 📁 Manuscript polished chapters
▸ 📁 Drafts working chapters
▸ 📁 Outlines chapter outlines
▸ 📁 Characters sheets & arcs
▸ 📁 Locations world-building
▸ 📁 Themes motifs & symbols
▸ 📁 Things props & artifacts
▸ 📁 Brief query & synopsis
▸ 📁 Research notes & bibles
the layout this guide builds — your whole story bible in one window
Before you start

Grab VS Code — it's free.

Prose Minion runs inside Visual Studio Code, Microsoft's free editor for Mac, Windows, and Linux. Download it first — everything in this guide happens inside it.

Download VS Code ↗
Pairs well with

Add a chat agent, too.

Extensions like Claude Code or OpenAI's Codex put a chat-based agent in the same sidebar. Prose Minion runs focused craft passes; an agent gives you hands — it can build this guide's folders for you, split one long draft into chapter files, or batch-rename chapters while you chat.

Claude Code ↗ Codex ↗
Sister extension
Frame Minion

Meet Frame Minion.

From the same maker: AI video generation in VS Code — generate frames, push them into video, or hand a whole song to the Music Video Orchestrator. Same bring-your-own-key, keys-stay-local design.

frameminion.video ↗
First, a primer

Three ideas, and VS Code makes sense.

01

A project is just a folder.

VS Code doesn't have "documents" — it opens a folder (it calls this a workspace). Everything inside that folder is your project. One novel, one folder.

02

Everything is plain text.

You write in Markdown — ordinary .md files. No proprietary format, no lock-in, works with every backup and versioning tool ever made. Prose Minion reads .md and .txt.

03

The Explorer is your binder.

The file tree in the left sidebar is the same thing as a three-ring binder with tabs — folders are the tabs. Prose Minion uses those tabs to know what's a character sheet and what's a chapter.

That last idea is the whole trick: your folder structure is metadata. Name the folders well and the Context Assistant, Metrics, Word Search, and Category Search all inherit a map of your book for free.

Step 1

Create your project.

1

Install VS Code

Free, from code.visualstudio.com — Mac, Windows, or Linux. Open it once so it's in your dock.

2

Make one folder for your book

In Finder or File Explorer, create a folder named after the project — my-novel. Put it somewhere that gets backed up (Documents, iCloud, Dropbox).

3

File → Open Folder…

In VS Code, open that folder. If it asks whether you trust the authors, click Yes, I trust — it's your own folder. The Explorer sidebar now shows your (empty) project.

4

Install Prose Minion & add your key

Extensions view → search Prose Minion → Install. Click the gear in the panel header and paste your OpenRouter key — it's stored in encrypted SecretStorage, never in a settings file.

Translation table
Your binder / Scrivener projectthe workspace folder
Binder tabs / sectionsfolders
A document / chapterone .md file
The binder sidebarthe Explorer
Two habits worth stealing

Turn on autosave (File → Auto Save) so you never think about saving again. And make a new file with right-click → New File in the Explorer — type the name including .md at the end.

Step 2

Build your binder — nine folders.

These names match Prose Minion's default Context Resource Paths, so everything works with zero configuration. Skip any you don't need — an empty folder costs nothing. The tags show which tools lean on each folder.

feeds context briefs metrics scope search scope
my-novel/ — the recommended layout matches the default settings
📁 Manuscript/ Polished chapters, ready for submission. One file per chapter. contextmetricssearch
📁 Drafts/ Working chapters and scene sketches — the messy versions. Also one file per chapter. contextmetricssearch
📁 Outlines/ Chapter outlines and beat sheets — where the Continuity tool checks what was supposed to happen. context
📁 Characters/ One sheet per character — voice, biography, arc notes. Subfolders are fine (Protagonists/, Antagonists/). context
📁 Locations/ Setting descriptions and world-building notes, one place per file. context
📁 Themes/ Thematic notebooks, motif tracking, symbolic elements. context
📁 Things/ Notable objects, artifacts, significant props — anything that must stay consistent. context
📁 Brief/ Query letter, synopsis, pitch materials, project overview — the 30,000-foot view of the book. context
📁 Research/ Research notes, tone & style guides, series bibles — plus files like story-bible.md and voice-and-tone.md anywhere in the project. context
Capitalization doesn't matter

Characters/ or characters/ — the default patterns match both. Just be consistent so your Explorer sorts cleanly.

Already organized differently?

Don't reorganize your book — point the extension at your names instead. Every group's pattern is editable in Settings → Context Resource Paths.

Step 3 · the one rule that matters most

One chapter, one file — under 5,000 words.

Both Manuscript/ and Drafts/ hold chapters — polished versus working. Either way, keep each chapter in its own Markdown file and aim for under ~5,000 words per file.

the whole book in one file
📄 the-whole-book.md · 92,000 words
  • Metrics average the whole book — a saggy middle hides inside one blended number.
  • Search reports one giant result instead of per-chapter counts and clusters.
  • The Context Assistant must trim aggressively — big files blow past the token budget.
  • Chapter-vs-standard page counts can't be compared.
one file per chapter
📄 chapter-01.md · 3,842 words
📄 chapter-02.md · 4,210 words
📄 chapter-03.md · 3,566 words
  • Metrics per chapter — compare pacing, dialogue %, and readability across the book.
  • Search returns a per-file table: which chapters overuse a word, and where it clusters.
  • The Context Assistant attaches only the chapters that matter — cheaper, sharper briefs.
  • Publishing-standard comparisons work at the chapter level.
Zero-pad the numbers

chapter-01.md, not chapter-1.md — so chapter 10 doesn't sort before chapter 2 in the Explorer.

Long chapter? Split the file.

A 9,000-word chapter can live as chapter-07a.md and chapter-07b.md. The tools don't care about your chapter numbering — only file size.

Drafts graduate

Write in Drafts/; when a chapter is submission-ready, move the file to Manuscript/. Both stay searchable.

The payoff

Set up once — four tools get smarter.

Every tool that asks "which files?" answers itself from your folder structure.

Context briefs

Click the assistant button next to a context field and the Context Assistant scans your folder groups — Characters, Locations, Themes — reviews only the files it believes are necessary, and lists every file it read under the context box. If Mira is in the excerpt, her sheet gets attached; the other forty don't.

reviewed: Characters/mira.md · Locations/harbor.md · Themes/salt-and-memory.md

Metrics

Run prose statistics on the active file, all of Manuscript/, all of Drafts/, or a path you type. Separate chapter files mean per-chapter pacing, dialogue %, and readability — and honest page-count comparisons against your publishing standard.

scope: 📁 Manuscript/ · 24 files · 87,300 words

Word Search

Hunt a tic — "suddenly," "just," a pet gesture — across one chapter or the whole book. With chapter files, results come back as a per-file table with hit counts, average gaps, and clusters, so you know chapter 12 has the problem, not "somewhere."

"suddenly" — chapter-12.md: 9 hits · 2 clusters · avg gap 310 words

Category Search

Semantic search by idea — "colors," "prepositions," "weather words" — scoped the same way. Ask for every color word in Drafts/ and see which chapters lean on the same palette. Scoping to a folder keeps the semantic pass focused and cheap.

"colors" in Drafts/ — 13 matched words · 20 occurrences
The only syntax you'll ever need

Glob patterns, demystified.

The Context Resource Paths settings use glob patterns — a shorthand for "which files count." There are exactly two you need to know, and they differ by one star.

Example folder
Characters/ ├── protagonist.md top level ├── main-cast.md top level ├── Protagonists/ │ ├── hero.md in a subfolder │ └── mentor.md in a subfolder └── Antagonists/ └── villain.md in a subfolder
Characters/* one level only
Characters/protagonist.md
Characters/main-cast.md
Characters/Protagonists/hero.md
Characters/Antagonists/villain.md

One * matches files directly inside the folder — subfolders are invisible.

Characters/**/* recursive · use this
Characters/protagonist.md
Characters/main-cast.md
Characters/Protagonists/hero.md
Characters/Antagonists/villain.md

The ** matches any depth of subfolders — including none. Files are found however you organize them.

Rule of thumb: always write the recursive form. Characters/**/* finds every character file no matter how you reorganize later. Separate multiple patterns with commas: characters/**/*,Characters/**/*. Only .md and .txt files are ever reviewed.

Reference

Every setting, explained.

Open Prose Minion, then click the gear icon in the panel header. The defaults are good — you can ship a whole book without touching most of these. Settings save automatically.

🔑OpenRouter API Key +

Stored with OS-level encryption via VS Code SecretStorage — Keychain on macOS, Credential Manager on Windows, libsecret on Linux. It never appears in settings files and is not synced to the cloud. AI features require an OpenRouter pay-as-you-go account; get a key at openrouter.ai.

Save API Key Clear API Key
🧠Models — one per job +
Assistant Modelanthropic/claude-sonnet-5Prose and dialogue assistants — analysis and creative suggestions.
Dictionary Modelanthropic/claude-haiku-4.5Dictionary and utility tools — synonyms, word expansions. Fast and cheap.
Context Assistant Modelopenai/gpt-5.4Project-aware context generation and resource discovery.
Category Search Modelanthropic/claude-sonnet-5Semantic word matching for category search.

the panel also includes curated model dropdowns and a Model Reference Guide link

⚙️General +
Include Craft GuidestrueAdds writing guides and examples to prompts. More tokens, better suggestions.
Temperature0.7Creative diversity — higher is more varied, lower is more focused.
Max Tokens10000Maximum response length. Truncation notices appear when the limit is hit.
Apply Context Window TrimmingtrueTrims large inputs to a 128K-token window — another reason small chapter files win.
Show Token Usage WidgettrueRunning token totals in the header. Resets manually (Reset Token Usage) or on reload.
🧭Publishing Standards +
PresetnoneComparison standard for prose metrics — none, generic manuscript guidelines, or genre-specific expectations.
Trim SizedefaultWith a genre selected, pick a typical book trim size for accurate page-count comparisons.
📊Word Frequency +
Top N Words100How many most-frequent words to display.
Include Hapax ListtrueWords that appear exactly once — unique vocabulary, one-off expressions, or typos.
Hapax Display Max300Maximum hapax words shown inline. The total count is always displayed.
Include Stopwords TabletrueCommon function words — "the," "and," "of" — for rhythm and balance checks.
Content Words OnlytrueExcludes stopwords to focus on nouns, verbs, adjectives, adverbs.
POS Tagger EnabledtrueOffline part-of-speech analysis. Falls back gracefully if unavailable.
Include BigramstrueTop two-word phrases — recurring collocations and clichés.
Include TrigramstrueTop three-word phrases — repeated phrasing and voice patterns.
Enable LemmasfalseExperimental — groups inflected word forms under their base form.
Word Length Histogram Max10Longest word length shown before longer words group into an N+ bucket.
Minimum Word Length1Filter word frequency by minimum character count, 1+ through 9+.
🔍Word Search +
Context Words3Words shown before and after each match.
Cluster Window50Word distance within which matches group as a cluster.
Min Cluster Size2Minimum matches inside the window before a cluster is reported.
Case SensitivefalseWhen enabled, "Rose" and "rose" are different words.
Enable Assistant ExpansionfalseUses the dictionary model to suggest synonyms and inflections. Marked "coming soon" in the UI.
📁Context Resource Paths — the folder map +

Comma-separated glob patterns telling the Context Assistant where each kind of material lives. Matched against your workspace folders; only .md and .txt files are reviewed — and only when you click the assistant button. You can always type context by hand instead.

Characterscharacters/**/*, Characters/**/*
Locations & Settingslocations/**/*, Locations/**/*, Locations-Settings/**/*
Themesthemes/**/*, Themes/**/*
Things / Propsthings/**/*, Things/**/*
Drafts & Outlinesdrafts/**/*, Drafts/**/*, outlines/**/*, Outlines/**/*
Manuscript Chaptersmanuscript/**/*, Manuscript/**/*
Project Briefbrief/**/*, Brief/**/*
General Referencesresearch/**/*, Research/**/*, tone-and-style/**/*, literary-devices/**/*, **/story-bible.md, **/synopsis.md, **/voice-and-tone.md, **/genre-conventions.md

settings save automatically — no save button, no config file to break

Folder made · key saved · binder built

Now go write chapter one.

Add to VS Code See example output