cloud-os-builder

name: cloud-os-builder description: Set up a Cloud OS — your second brain hosted either in a synced cloud folder (Google Drive, OneDrive, iCloud, Dropbox) or as a Notion workspace. Asks where to host it, then takes the matching route: for a folder it builds the Local OS structure (root CLAUDE.md + Context + Projects/Intelligence/Daily/Resources/Skills, Obsidian-free) and gives provider-specific sync steps; for Notion it builds pages + databases via the connector. Runs the same personalized onboarding either way. Two modes — Solopreneurs/Professionals (default), Business/Teams. The Cloud OS tier of the AIOS family. Use when the user says "set up a cloud OS", "host my second brain on Google Drive/Dropbox/iCloud/OneDrive/Notion", or runs /cloud-os-builder.

Cloud OS — Host Selection + Setup + Onboarding

USE WHEN the user runs /cloud-os-builder or asks to set up a Cloud OS, or to host their second brain in a synced cloud folder (Google Drive, OneDrive, iCloud, Dropbox).

This is the Cloud OS tier of the AIOS family. It builds the same structure the Local OS uses, but inside a synced cloud folder so it follows the user across devices — and it omits the Obsidian .obsidian/ layer (that layer does nothing without Obsidian and is the part that syncs badly on consumer cloud). This skill is self-contained: every reference template it needs lives in its own references/ folder; it does not depend on any other skill.

This is a multi-step process:

• Phase H: Host Selection — ask which cloud and confirm the exact folder path (do this FIRST)
• Phase 0: Mode Selection — ask which OS variant to create
• Phase A: Bootstrap — create the directory structure and system files inside the cloud folder (skip Obsidian-specific steps)
• Phase B: Onboarding — interview the user and personalize everything
• Phase C: Cloud Sync Setup — exact provider step-by-step + verify sync

Phase H: Host Selection (do this first)

Two questions, via AskUserQuestion (the second only if needed). This keeps each question within the 4-option limit.

H.1 — Route

Ask: "Where do you want to host your OS?" Pass both label and description:

• Notion — "Build it as a Notion workspace (pages + databases with relations + a home dashboard). You operate from inside Notion; Claude reads/writes via the Notion connector. Realtime + multiplayer, but cloud-only (no local file copy)."
• Synced cloud folder — "A plain folder synced via Google Drive, iCloud, Dropbox, or OneDrive. Real files on disk; Claude operates it directly. Single-user, with a sync delay."

Branch:

• If they pick Notion → jump to ## Notion Route near the bottom. Skip Pre-flight, Phase A, and Phase C (those are the file route). The Notion Route reuses Phase 0 and the Phase B interview.
• If they pick Synced cloud folder → continue with H.2, then Pre-flight → Phase 0 → A → B → C.

H.2 — Provider (file route only)

Ask: "Which cloud?" Pass both label and description:

• Google Drive — "Most free storage (15 GB); simple desktop sync."
• iCloud — "Best if you're all-Apple; near-realtime across Mac/iPhone/iPad."
• Dropbox — "Most reliable sync; best mobile path (via Remotely Save)."
• OneDrive — "Best if you live in Microsoft 365 / Windows."

If they're unsure, read references/comparison.md and recommend by their devices (Apple-only → iCloud; mobile matters → Dropbox; Microsoft shop → OneDrive; most free storage → Google Drive).

Then confirm the exact folder path inside that provider's synced location, and create it:

• iCloud → ~/Library/Mobile Documents/com~apple~CloudDocs/<OS-name>
• Google Drive → ~/Google Drive/My Drive/<OS-name>
• Dropbox → ~/Dropbox/<OS-name>
• OneDrive → ~/OneDrive/<OS-name>

This cloud folder is the OS root. cd into it (or treat it as the working directory) so every ./… write path in Phase A/B below resolves inside this folder. Remember the chosen provider — Phase C uses it.

Pre-flight Check

Check if claude.md or CLAUDE.md exists only in the current working directory (do NOT search subdirectories or parent directories — check only the exact CWD path).

• If it exists: The vault is already set up. Ask the user:
  • "This vault is already set up. Would you like to:"
  • Re-run the interview — Keep existing structure, update memory files based on new answers
  • Full reset — Delete everything and start fresh (confirm twice before proceeding)
  • Cancel — Do nothing
• If it does NOT exist: Proceed with full setup (Phase 0 + Phase A + Phase B)

Phase 0: Mode Selection

Ask the user to pick a mode using AskUserQuestion with these exact label and description values:

• Question: What type of vault do you want?
• Option 1 label: Solopreneurs/Professionals — description: Blends work and personal. Best for solo founders, freelancers, consultants.
• Option 2 label: Business/Teams — description: Org structure with departments, processes, stakeholders. Best for teams and companies.

CRITICAL: You MUST pass both label AND description for each option in AskUserQuestion. The description field is what explains each mode to the user. Never leave description empty.

Mode mapping:

• Solopreneurs/Professionals → os-mode: professional
• Business/Teams → os-mode: business

Accept any clear signal: "solo", "professional", "freelancer", "business", "org", "team", etc.

If the user skips or says "I don't know", use Solopreneurs/Professionals (professional mode).

Store the selected mode. It will be written to CLAUDE.md frontmatter as os-mode: professional | business.

Phase A: Bootstrap (file route only)

Create the directory structure and write all system files for the selected mode. (For the Notion route, skip this entirely — see ## Notion Route near the bottom.)

Resolving reference file paths

Every references/<file>.md mentioned below lives in the references/ subdirectory next to this SKILL.md — not in the user's working directory. Two conventions matter:

• Read paths (references/foo.md) → resolve relative to this SKILL.md's directory.
• Write paths (./Foo/CLAUDE.md) → resolve relative to the user's current working directory (the vault root).

If the Read tool can't open a references/... path directly (some harnesses mount the skill at a path that differs between Read and Bash), run a quick discovery step once before Step A.2:

# Find the references directory; cache the result for the rest of Phase A.
find / -type d -path '*/cloud-os-builder/references' 2>/dev/null | head -1

Use that absolute path as the prefix for every reference read in Phase A and Phase B. Don't retry path resolution per-file — do it once and reuse.

Step A.1: Create Directory Structure

All modes share this base:

mkdir -p .claude
mkdir -p Context
mkdir -p Projects
mkdir -p Daily
mkdir -p Resources
mkdir -p Skills

Solopreneurs/Professionals mode adds:

mkdir -p Intelligence/meetings/team-standups
mkdir -p Intelligence/meetings/client-calls
mkdir -p Intelligence/meetings/one-on-ones
mkdir -p Intelligence/meetings/general
mkdir -p Intelligence/competitors
mkdir -p Intelligence/market
mkdir -p Intelligence/decisions
mkdir -p Intelligence/archive

Business mode adds:

mkdir -p Intelligence/meetings/team-standups
mkdir -p Intelligence/meetings/client-calls
mkdir -p Intelligence/meetings/one-on-ones
mkdir -p Intelligence/meetings/board-reviews
mkdir -p Intelligence/meetings/all-hands
mkdir -p Intelligence/meetings/cross-team
mkdir -p Intelligence/meetings/general
mkdir -p Intelligence/competitors
mkdir -p Intelligence/market
mkdir -p Intelligence/decisions
mkdir -p Intelligence/processes
mkdir -p Intelligence/archive
mkdir -p Departments
mkdir -p Team
mkdir -p Onboarding
mkdir -p Resources/templates

Team/ is created empty here. Profile-first subfolders (Team/{org}/Profiles/{person}/...) are scaffolded in Phase B once Q6 answers are in.

Step A.2: Write System Files from References

Read each reference file and write it to the corresponding local path. The reference files contain the complete content for each system file.

All modes — shared system files:

Mode-specific root CLAUDE.md template:

Per-folder routing indexes (every major folder gets its own CLAUDE.md — matches production vault convention):

For each row applicable to the selected mode: read the reference file, then write its content to the local path.

Step A.3: Initialize Starter Context Files

All modes — create placeholder skill folders:

mkdir -p Skills/linkedin-writer/references
mkdir -p Skills/newsletter-writer/references

Then write placeholder files from references:

• Read references/skills-placeholder-linkedin-notes.md → write to ./Skills/linkedin-writer/notes.md
• Read references/skills-placeholder-linkedin-example.md → write to ./Skills/linkedin-writer/references/example-post.md
• Read references/skills-placeholder-newsletter-strategy.md → write to ./Skills/newsletter-writer/strategy.md
• Read references/skills-placeholder-newsletter-example.md → write to ./Skills/newsletter-writer/references/example-edition.md

Solopreneurs/Professionals mode:

• Read references/context-me.md → write to ./Context/me.md

Business mode:

• Read references/context-operator.md → write to ./Context/operator.md
• Read references/context-organization.md → write to ./Context/organization.md
• Read references/context-team.md → write to ./Context/team.md
• Read references/context-strategy-business.md → write to ./Context/strategy.md

Business mode — Team/ is created empty in Phase A. Profile-first scaffolding (Team/{org}/Profiles/{person}/...) happens in Phase B Build Step 3 once Q6 answers identify the actual people.

Step A.4: Make Hooks Executable

chmod +x .claude/hooks/*.sh

Step A.5: Confirm Bootstrap

Tell the user:

• "Cloud OS structure created successfully in [mode] mode inside your [provider] folder."
• List the main folders created (varies by mode), including Skills/
• Mention Resources/ for storing prompts, frameworks, swipe files, and templates
• Note that this folder is plain markdown — Claude Code operates it directly, and they can optionally open it in Obsidian later (that just adds the Obsidian layer on top).
• "Now let's personalize it for you."

Then proceed to Phase B.

Phase B: Onboarding — Guided Brain Dump

This skill runs inside Cowork. Phase B uses Cowork's rich-HTML widget tool — not AskUserQuestion — to render a real form with stacked categories, free-text textareas, and proper styling (matches the look of os-optimizer's "Audit run details" form).

It's a guided brain dump across 12 categories of the user's life and business, batched into 3 rich-HTML forms (4 categories per form). Bullet points inside each category are inspiration prompts — riff on whatever lands.

The pitch to the user: sit down for an hour or two, pour a beer, order a pizza, and brain-dump. It's not only for the assistant to feel personal on day one — it's a useful exercise in itself.

The tool: mcp__visualize__show_widget (Cowork-only)

Each of the 3 forms is one call to mcp__visualize__show_widget. The tool accepts:

The user fills in the form and submits. The submitted values come back to the agent as the tool result. The agent then proceeds to the next form. No AskUserQuestion. No radio buttons. No "Other" box.

How the user should respond — per category

Inside each category's textarea, the user can:

1. Paste a Whisper / dictation transcript — open phone or Mac dictation, ramble for 2–5 minutes, paste the transcript.
2. Paste documents — links to PDFs, Notion pages, Google Docs, brand guides, About pages, LinkedIn profiles, OKR docs, decks. Or drop file paths.
3. Point at connectors — paste a Notion workspace URL, a wiki link, a Drive folder.
4. Type long-form free text.

Two kinds of knowledge: what lives in your head (Whisper it) and what already lives online or in a tool (paste links / docs). Mix freely per category. Leave a textarea blank to skip that category.

Before Form 1 — Send one orienting message

Send this verbatim (or close to it), no tool call yet:

Three short forms, four categories each, twelve categories total. This isn't a quiz — it's a guided brain dump.

Each category has three ways to give me context: a brain-dump textarea, a links & file paths field, and a file upload. Use any or all. Brain-dump anything around the bullet inspirations — you don't have to hit each one. Leave a category blank to skip it.

Best inputs: a Whisper / dictation transcript, an About page URL, a brand guide PDF, an OKR doc, a LinkedIn profile, a Notion page. The more you give me, the less generic your vault will be on day one.

Sit down for an hour or two. Pour a beer. Order a pizza. This is worth it.

Submit each form when ready. Type "skip all" anytime to jump to defaults.

Widget HTML template (every category in every form uses this shape)

Each category gets three inputs: a brain-dump textarea, a links/paths textarea, and a file upload input. Any or all can be filled. All blank = skip.

Inside widget_code for each form, build a <form class="elicit"> containing one header and four elicit-group blocks. Per category:

<div class="elicit-group">
  <label class="elicit-question">{N}/12 — {Category name}</label>
  <div class="elicit-bullets" style="font-size:13px; color:var(--color-text-secondary); margin:8px 0">
    <ul style="margin:0; padding-left:18px">
      <li>{inspiration bullet 1}</li>
      <li>{inspiration bullet 2}</li>
      <li>{inspiration bullet 3}</li>
      <!-- etc -->
    </ul>
    <p style="margin-top:6px; font-style:italic">Brain-dump in the textarea below, OR paste links / file paths, OR upload docs. Any combination. Leave all blank to skip this category.</p>
  </div>

  <textarea class="elicit-textarea" name="cat{N}_braindump" rows="6"
    style="width:100%; border-radius:10px; padding:10px; border:1px solid var(--color-border-subtle); font-family:inherit; font-size:13px; margin-bottom:8px"
    placeholder="Brain dump — paste a Whisper transcript, or type long-form…"></textarea>

  <textarea class="elicit-textarea" name="cat{N}_links" rows="2"
    style="width:100%; border-radius:10px; padding:10px; border:1px solid var(--color-border-subtle); font-family:inherit; font-size:13px; margin-bottom:8px"
    placeholder="Links & file paths — one per line (Notion URL, LinkedIn profile, /path/to/file.pdf, etc.)"></textarea>

  <input class="elicit-file" type="file" name="cat{N}_files" multiple
    accept=".md,.txt,.pdf,.docx,.pptx,.xlsx,.csv,.json,.yaml,.yml,.png,.jpg,.jpeg"
    style="font-size:12px; color:var(--color-text-secondary)">
</div>

And one header at the top of <form class="elicit">:

<div class="elicit-header">
  <svg viewBox="0 0 20 20" fill="currentColor" width="20" height="20"><!-- pencil/clipboard icon --></svg>
  <span>{Form title}</span>
</div>
<div class="elicit-body">
  <!-- 4 elicit-group blocks -->
</div>

Reuse the SVG icon pattern from the os-optimizer Audit run details widget (clipboard-with-marks icon). Form titles: "You & business", "Customer & brand", "How you operate" (solo) — or "You & company", "Offer, customer & brand", "How the company operates" (business).

Reading form submissions

When the widget returns, the result is a record mapping each input's name to its value:

• cat{N}_braindump → string (the typed text / transcript)
• cat{N}_links → string (newline-separated URLs and file paths)
• cat{N}_files → array of file references (Cowork uploads these into the workspace folder; the result gives you the paths or signed URLs)

A category is "skipped" only when all three inputs are empty/blank.

Ingestion between forms

After each form returns, for each category (N = 1..4 in this form):

1. cat{N}_braindump — if non-empty, tag and store raw in the working corpus under the category. Don't paraphrase.
2. cat{N}_links — split on newlines. For each line:
   • HTTP(S) URL → fetch with WebFetch / WebSearch.
   • Local file path → Read it.
   • Folder path → Glob, then Read each file.
3. cat{N}_files — for each uploaded file:
   • .md, .txt, .json, .yaml, .csv → Read directly
   • .pdf → Read with pages param if large
   • .docx/.pptx/.xlsx → use pandoc / textutil via Bash if available; otherwise note and continue
   • Images → Read (multimodal)

Merge everything into the corpus tagged by category. Then immediately fire the next form. No commentary or summarization between forms.

Both modes use Oskar's category breakdown. Bullet inspiration prompts are Oskar's prompt blocks verbatim, plus Ben's framing of "brain-dump anything around any of these bullets."

Solopreneurs/Professionals mode — 3 forms × 4 categories

Form 1 — You & business — one mcp__visualize__show_widget call. Title: os_setup_form_1_you_business. Contains Q1–Q4 as stacked elicit-group blocks.

Q1. You. Form header: You Bullets:

• Name, role/title, location, industry
• When and how you do your best work (mornings? deep blocks? after a walk?)
• If someone you respected had to introduce you in a room of people you respected, how would you want them to describe you?
• 5 attributes that describe you (one or two words each)

Q2. Your origin and POV. Header: POV Bullets:

• Why you started or joined what you're doing now
• A belief or POV you hold strongly, even when it's unpopular
• The "big idea" your work is built on (the wedge, the thesis)
• Who or what you're fighting against — a category, a behavior, a competitor archetype, a status quo

Q3. What you sell. Header: Lines Bullets (one paragraph per revenue line, or skip if none yet):

• Name, what it does, who it's for, stage
• Current revenue baseline if applicable
• How it came to exist. What made you start it.

Q4. The promise. Header: Offer Bullets:

• The 1–3 problems you solve for customers
• For each problem: are customers already aware they have it, or do you have to teach them?
• Your value proposition in one sentence
• The promise or guarantee you make (explicit or implicit)
• Why customers actually pick you — in their words if you've heard them say it

Form 2 — Customer & brand — one mcp__visualize__show_widget call. Title: os_setup_form_2_customer_brand. Contains Q5–Q8 as stacked elicit-group blocks.

Q5. The customer. Header: Customer Bullets:

• Title, role, industry, responsibilities
• What their day looks like, what tools they live in
• The language and words they use to describe their problem
• The dream outcome they want
• The situation they're in before they come to you — what triggered the search
• How long they typically take to decide to buy
• The media, podcasts, newsletters, or creators they follow
• 3–5 real examples (names, LinkedIn profiles, or company names)

Q6. Your voice and look. Header: Voice Bullets:

• Tone descriptors that fit (direct, warm, dry, technical, playful, serious, supportive…)
• 5 attributes that describe how you sound
• Signature phrases you actually use
• Words or phrases you'd never use
• Topics you love talking about
• Topics you refuse to discuss publicly
• Brand colors, fonts, taglines if you have them
• The feeling people should carry away after reading your stuff
• Or: paste a writing sample / link and I'll extract from it

Q7. Your positioning. Header: Position Bullets:

• The enemy you're fighting (the category, behavior, or competitor archetype)
• How you solve the problem differently from the obvious alternatives
• 3–4 distinct messages you want associated with your name or brand

Q8. This year's priorities. Header: Priorities Bullets:

• 1–3 outcomes with a number attached (revenue, audience size, ship date)
• The why behind each
• What you're explicitly saying no to in order to focus here

Form 3 — How you operate — one mcp__visualize__show_widget call. Title: os_setup_form_3_how_you_operate. Contains Q9–Q12 as stacked elicit-group blocks.

Q9. Active projects. Header: Projects Bullets (for each project):

• Name, one-line purpose, status, deadline if any
• Which business it belongs to (if multiple)
• Who else is involved

Q10. The people you work with. Header: People Bullets:

• Team, contractors, key external contacts
• For each: name, role, how you work together
• Skip if fully solo

Q11. Your stack. Header: Stack Bullets:

• Stack across communication, meetings, CRM, content, finance, dev, automation
• Source of truth for each main workflow — where deals live, where decisions live, where writing actually happens, where the calendar lives

Q12. Drains and workflows to automate. Header: Drains Bullets:

• Top 1–2 painful, repetitive workflows. Use this template: When X happens → I do Y → it takes Z time → output is W → what I want is V
• What's draining your attention right now — unclosed loops, decisions sitting unmade, things that should be done but aren't

Business/Teams mode — 3 forms × 4 categories

Same question shape as solo mode. Three AskUserQuestion calls, four questions each.

Form 1 — You & company — one mcp__visualize__show_widget call. Title: os_setup_form_1_you_company. Contains Q1–Q4 as stacked elicit-group blocks.

Q1. You, as operator. Header: Operator Bullets:

• Name, title, department, who you report to
• Decision authority (what you can sign off on alone)
• Location, working style
• What's draining your attention right now — unclosed loops, decisions sitting unmade

Q2. The company. Header: Company Bullets:

• Legal entity name, industry, stage
• Founded year, headcount (FT + contractors)
• Headquarters and where the team is based
• One-sentence mission
• Why the company started (origin)
• The belief or POV the company stands for

Q3. The market. Header: Market Bullets:

• The broad target industry
• The specific niche you operate in
• Trends and hot topics in the industry right now
• What's not going well in the industry — the inefficiency or broken thing you're betting against
• What changed in the last 5–10 years
• The main players (incumbents, competitors, adjacent categories)

Q4. What you sell. Header: Lines Bullets (for each revenue line):

• Name, what it does, who buys it
• Current revenue baseline, status (active, new, sunsetting)

Form 2 — Offer, customer & brand — one mcp__visualize__show_widget call. Title: os_setup_form_2_offer_customer_brand. Contains Q5–Q8 as stacked elicit-group blocks.

Q5. The promise. Header: Offer Bullets:

• The 1–3 problems you solve for customers
• For each: are customers already aware they have it, or do you teach them?
• Value proposition in one sentence
• The promise or guarantee you make
• Key features and capabilities that deliver the value
• Why customers actually pick you over alternatives
• The kind of results you typically deliver — include a real example if you have one

Q6. The customer. Header: ICP Bullets:

• Who's in charge of buying — title, role, responsibilities
• What their day looks like, what tools they use
• The language and words they use to describe their problem
• Dream outcome they want
• Situation before buying — what triggered them to look
• How long the buying decision typically takes
• Market trends affecting them right now
• Media, podcasts, or creators they follow
• 3–5 real examples (LinkedIn profiles or company names)

Q7. The brand voice and look. Header: Voice Bullets:

• Tagline and value prop in plain language
• Voice in a paragraph or as descriptors (direct, warm, dry, technical…)
• 5 attributes that describe how the brand sounds
• Signature phrases used across your content
• Words or phrases you'd never use
• Topics you love covering
• Topics you avoid publicly
• Brand colors, fonts, logo notes
• The feeling readers and customers should carry away

Q8. The positioning. Header: Position Bullets:

• The enemy — the category, status quo, or competitor archetype you're fighting
• How you solve the problem differently from obvious competitors
• Brand personality in 5 adjectives
• The "big concept" the company is built on
• 3–4 distinct messages you want associated with the brand

Form 3 — How the company operates — one mcp__visualize__show_widget call. Title: os_setup_form_3_how_company_operates. Contains Q9–Q12 as stacked elicit-group blocks.

Q9. The team. Header: Team Bullets:

• The departments and the lead for each
• Team members getting their own profile folders. For each: name, role, reports-to, FT or contractor, location
• (Profile folders include their own daily notes, tasks, and sub-schedules.)

Q10. This year's OKRs. Header: OKRs Bullets:

• 1–3 objectives for the year (or quarter)
• For each KR: target number, owner, current status
• The why behind each objective
• What you're explicitly saying no to in order to focus here

Q11. Active projects. Header: Projects Bullets (for each project):

• Name, owner, status, deadline if any
• Client-facing or internal
• Which business unit or department it sits under
• Key collaborators

Q12. Stack, workflows, and stakeholders. Header: Stack Bullets:

• Stack across communication, meetings, CRM, PM, content, finance, dev
• Source of truth for each main workflow — where deals live, where decisions live, where writing happens, where the calendar lives
• Top 3 painful, repetitive workflows. Template: When X happens → we do Y → it takes Z → output is W → what we want is V
• External stakeholders: investors, partners, vendors, top clients. Name, type, nature of the relationship.

The user submits each form with one click. Per-category response patterns:

• Type / paste a brain dump, transcript, links, docs, or file paths into the textarea
• Leave the textarea blank to skip that category
• Reply "skip all" between forms — stop asking and move to Phase B+

Accept whatever they give. Don't ask follow-ups inside or between forms. Extract what you can.

If the user submits every form empty — proceed to build with defaults only.

Phase B+: Additional Context Drop

After Q12 (or "skip all") and before Phase B Build, ask one final AskUserQuestion to invite any leftover source material that didn't surface during the 12 categories. Most users still have brand decks, About pages, intake forms, LinkedIn URLs, Notion docs, PDFs, slide exports, voice/style guides, OKR docs, org charts, project briefs, etc. Always ask, even if Q1–Q12 looked rich.

Call AskUserQuestion (one question, header: Context):

• Question: "Anything else I should pull from before building? Upload files (PDFs, MDs, DOCXs), paste links (LinkedIn, websites, Notion pages, Google Docs), point me at a local folder, or paste raw text. The more I have, the more personalized your vault will be — instead of template scaffolds with placeholders."
• Options:
  • Yes — I'll paste links / upload files — "Walk me through it"
  • Yes — point me at a folder on disk — "I have local files"
  • No — use just the answers above — "Build with what we have"
  • Skip — "Skip this step"

If the user picks a "Yes" option (or pastes content directly):

1. Collect everything they share. Be greedy — accept anything they offer.
2. For each link: call WebFetch (or WebSearch if the URL is a search). Extract the relevant content.
3. For each uploaded file or local file path:
   • .md, .txt, .json, .yaml, .csv → read directly with Read
   • .pdf → read with Read (use pages parameter if >10 pages)
   • .docx, .pptx, .xlsx → use Bash with pandoc or textutil if available; otherwise tell the user to export as PDF or MD and re-share
   • Images / screenshots → read with Read (multimodal)
4. For a local folder path: use Glob to enumerate, then read each file.
5. Maintain a context corpus in working memory — every fact, name, number, quote you find. Tag each by likely target (me.md, brand.md, icp.md, strategy.md, projects/{name}, etc.).
6. After ingestion, briefly tell the user what you pulled (e.g., "Pulled 4 files: brand-guidelines.pdf, about-page.md, okrs-2026.md, team-roster.csv. 18 links fetched."). One sentence. Then proceed to Build.

If the user picks No or Skip: proceed straight to Build with only the Q1–Q12 answers from Phase B.

Phase B Build: Personalize the Vault

After Q12 + the additional-context drop (or skips), build everything you can from what the user gave you. Work silently — don't narrate each step.

CRITICAL: real personalization, not template scaffolds

The reference files in references/ are scaffolds — they show the section structure to use. They are not the output. Do not copy a template verbatim with placeholders intact.

For every file you write:

1. Read the reference template to learn the section structure (headings, frontmatter shape, section order).
2. Replace every placeholder (anything in [brackets] or marked as TBD) with real data extracted from the 12 Phase B answers + the Phase B+ corpus.
3. If a section has zero supporting data after exhausting both Q answers and the corpus: omit the entire section rather than writing [name] or TBD. The output should never contain bracketed placeholders.
4. If only some bullets in a section have data: keep the section, drop the empty bullets.
5. Use the user's actual words, names, numbers, URLs, and quotes wherever the corpus contains them. Don't paraphrase facts — preserve specificity (exact company names, exact dollar figures, exact dates, exact phrases the user uses).
6. Cross-reference: a single fact may belong in multiple files (e.g., "we sell to RevOps leaders at Series B SaaS" belongs in both icp.md and brand.md positioning). Place it in each file where it's relevant.
7. Frontmatter updated: = today's date.

A finished context file should read as a real human-written document about the user. If it reads like a fillable form, you did it wrong — go back and fill it.

Build Step 1: Create Context Files

Behavior depends on selected mode.

For every file below, source data from BOTH the Q answers AND the Phase B+ corpus (uploaded files, fetched links, folder reads). The corpus typically contains the depth — Q answers are anchors.

Solopreneurs/Professionals mode (Q1–Q12 = solo brain-dump categories):

• Context/me.md — Always created. Fill from Q1 (name, role, location, peer-intro line, attributes, working style) + Q2 (origin / POV / wedge / enemy) + Q12 (drains, unclosed loops) + corpus. Read references/context-me.md as scaffold.
• Context/business.md — Only if Q3 had content. Fill from Q3 (revenue lines: name, what it does, who it's for, stage, baseline, origin) + corpus (About page, business overview docs). Read references/context-business.md as scaffold.
• Context/services.md — Only if Q3 lists multiple revenue lines or corpus has product/service docs. Read references/context-services.md as scaffold.
• Context/pain-points.md — Only if Q4 named problems or Q2 surfaced one. Include awareness column (aware vs needs education) using Q4's awareness signal. Read references/context-pain-points.md as scaffold.
• Context/icp.md — Only if Q5 had content or corpus has ICP material. Fill role, day, language, dream outcome, trigger, decision time, media, examples. Read references/context-icp.md as scaffold.
• Context/brand.md — Only if Q6 (voice), Q7 (positioning), or Q4 (why-pick-you) had content, or corpus has brand material. From Q4 take value prop + why-pick-you. From Q6 take voice descriptors, signature phrases, words-to-avoid, feeling, colors/fonts. From Q7 take enemy, differentiation, key messages. Read references/context-brand.md as scaffold.
• Context/strategy.md — Only if Q8 had content. Fill priorities, why, and explicit nos. Read references/context-strategy.md as scaffold.
• Context/team.md — Only if Q10 had content (people / collaborators) or corpus has a team / contractor list. Read references/context-team.md as scaffold.
• Context/infrastructure.md — Only if Q11 (stack) or Q12 (workflows) had content, or corpus has a stack doc. Combine tool stack (Q11) + workflows-to-automate (Q12). Read references/context-infrastructure.md as scaffold.

Business mode (Q1–Q12 = business brain-dump categories):

• Context/operator.md — Always created. Fill from Q1 (name, title, reports-to, decision authority, working style, drains) + corpus. Read references/context-operator.md as scaffold.
• Context/organization.md — Always created. Fill from Q2 (legal name, industry, stage, founded year, headcount, HQ, mission, origin, POV) + corpus (About page, company deck). Read references/context-organization.md as scaffold.
• Context/market.md — Only if Q3 had content or corpus has market / industry material. Fill industry, niche, trends, what's broken, last 5–10y shifts, main players. Read references/context-market.md as scaffold.
• Context/services.md — Only if Q4 had content. Fill revenue lines from Q4 + corpus (sales deck, product pages). Read references/context-services.md as scaffold.
• Context/pain-points.md — Only if Q5 surfaced problems or corpus has them. Include awareness signal from Q5. Read references/context-pain-points.md as scaffold.
• Context/icp.md — Only if Q6 had content or corpus has ICP material. Fill role, day, language, dream outcome, trigger, decision time, market trends, media, examples. Read references/context-icp.md as scaffold.
• Context/brand.md — Only if Q7 (voice) or Q8 (positioning) had content or corpus has brand material. From Q7 take tagline, voice descriptors, signature phrases, words-to-avoid, topics, colors/fonts, feeling. From Q8 take enemy, differentiation, personality adjectives, big concept, key messages. Read references/context-brand.md as scaffold.
• Context/team.md — Always created. Fill from Q9 (departments + leads, team members) + corpus (org chart). Read references/context-team.md as scaffold.
• Context/strategy.md — Always created. Fill from Q10 (objectives, KRs, owners, why, explicit nos) + corpus (OKR doc). Read references/context-strategy-business.md as scaffold.
• Context/infrastructure.md — Only if Q12 listed tools or workflows, or corpus has a stack / SOPs doc. Combine tool stack + sources of truth + workflows-to-automate from Q12. Read references/context-infrastructure.md as scaffold.
• Context/stakeholders.md — Only if Q12 mentioned external stakeholders or corpus has investor / partner / client lists. Read references/context-stakeholders.md as scaffold.

Build Step 2: Create Project Folders

Solo: from Q9 (active projects). Business: from Q11 (active projects / initiatives). Plus any project briefs / Notion exports / project lists in the corpus. Intelligently structure each project based on what the user gave you.

Analyze the info and decide the right structure:

• Simple mention ("working on a podcast") → just a README.md
• Moderate detail (scope, deadlines, people) → README.md + relevant subdirs
• Rich info (briefs, specs, research, multiple workstreams) → full structure with subdirs and files

Create subdirectories only when the content justifies them:

README.md is always the index:

---
type: project
status: active
owner: [name]
business: [business unit if applicable]
created: YYYY-MM-DD
updated: YYYY-MM-DD
---
## Overview
[What this project is]

## Current Status
[Where things stand]

## Key Resources
[Links, tools, contacts]

## Next Steps
[What needs to happen]

Don't create empty subdirs. Don't cram everything into the README. Distribute content into the right files based on what it actually is.

Business mode only — from Q9 + corpus, also create Departments/{name}/README.md for each department with the lead's name, charter placeholder, and sops/ subfolder.

Build Step 3: Profile-First Team Scaffolding (Business mode only)

From Q9 + corpus (org chart, team roster), scaffold each person's profile workspace. Slug names are kebab-case.

{org-slug} is derived from Q2 (company name → kebab-case). If no company name given, default to team.

For each FT employee:

mkdir -p Team/{org-slug}/Profiles/{person-slug}/Daily
mkdir -p Team/{org-slug}/Profiles/{person-slug}/task-list
mkdir -p Team/{org-slug}/Profiles/{person-slug}/sub-schedules

Then write:

• Read references/team-profile-template.md → write to ./Team/{org-slug}/Profiles/{person-slug}/{Person Name}.md. Fill frontmatter and sections from Q9 (name, role, reports-to, FT, location) + corpus.
• Read references/team-tasks-template.md → write to ./Team/{org-slug}/Profiles/{person-slug}/task-list/Tasks.md.

For each contractor / advisor:

mkdir -p Team/External/contractors/{person-slug}

Then write the same team-profile-template.md (with employment: contractor or advisor) → ./Team/External/contractors/{person-slug}/{Person Name}.md.

If Q9 + corpus list no team members, don't scaffold anything under Team/{org-slug}/Profiles/. Leave Team/ with just the CLAUDE.md routing index.

Build Step 4: Mode-specific Additional Setup

Business mode only:

• If Q12 or corpus mentioned org-wide processes / SOPs, capture them in Intelligence/processes/{name}.md
• If user provided onboarding docs in the corpus, route them to Onboarding/{name}.md

Build Step 5: Create First Daily Note

Create Daily/YYYY-MM-DD.md (today's date):

---
type: daily-note
date: YYYY-MM-DD
---
# YYYY-MM-DD

## Session
- **Focus**: Initial vault setup and onboarding
- **Completed**: Full vault bootstrap + personalized onboarding
- **Next Steps**: [based on what was discussed]

Build Step 6: Confirm Completion

Tell the user:

• Quick summary of what was created (which context files, how many projects, any departments, any team profiles)
• "You can add more context anytime — just tell me and I'll update the right files."
• Suggest a next action based on what they told you

Then proceed to Phase C to set up cloud sync.

Phase C: Cloud Sync Setup (file route only)

The structure now exists inside the cloud folder. Walk the user through making it actually sync — one step at a time, using the guide for the provider they chose in Phase H.

1. Open references/<provider>.md (google-drive.md / icloud.md / dropbox.md / onedrive.md) and follow it step by step:
   • Confirm the desktop sync app is installed + signed in.
   • Confirm the OS folder is in the synced location (from Phase H).
   • Disable online-only placeholders for the folder ("Always keep on this device" / "Make available offline") so Claude and any editor see real local files, not cloud stubs.
   • Set up the mobile path if they want it (e.g. Dropbox → Remotely Save; iCloud → Obsidian iOS).
2. Verify:
   • Confirm a real local copy exists (not a placeholder).
   • Test sync: edit a file, confirm it appears on another device.
   • Point Claude Code at the synced folder so the root CLAUDE.md loads as the OS.
3. Set expectations: consumer cloud is single-user, multi-device — there's a sync delay, it's not realtime, and simultaneous edits create "conflicted copy" files. For true team + realtime, recommend the Local OS + Obsidian Sync/Relay.

Notion Route

Use this instead of Pre-flight, Phase A, and Phase C when the user picked Notion in H.1. The OS lives in a Notion workspace: documents become pages, collections become databases with relations and views. Claude operates it through the Notion connector (MCP). Notion holds the data; a CLAUDE.md page holds the master index + operating conventions (Notion can't auto-load it — the operator fetches it first each session).

Follow references/notion-template-spec.md exactly — it is the source of truth for the structure, the build order, and the connector gotchas. Read it before building. The result must match the polished out-of-the-box spec (Context is a database with a flat lead view, a CLAUDE.md master index, dashboard views in a fixed order, plus a generated "Notion OS Assistant" skill handed to the user so they can use the OS immediately). Do not ship a barebones version.

N.1 — Connect Notion

Make sure the Notion connector is available. If its tools aren't loaded, run the Notion authenticate flow and wait for the user to finish. Confirm you can search/list their workspace before building. Ask which page should be the parent for the new OS (or create it at the workspace root).

N.2 — Mode

Run Phase 0: Mode Selection above (Solopreneurs/Professionals vs Business/Teams). Mode decides which databases/pages to create — Business adds Departments, Team, Onboarding, and Processes.

N.3 — Build the structure (via MCP)

Build exactly per references/notion-template-spec.md, in its build order:

1. Home page (🏠 <Name> — Operating System).
2. CLAUDE.md page (the master index) + Skills/Onboarding pages.
3. Context as a database (NOT a page of sub-pages) + the 7/8 databases. Record each data_source_id.
4. Add non-relation properties, then relations in a second pass (both sides exist).
5. Per-database views (Projects board-by-status, Intelligence by-type, Daily calendar, Context by-category board, etc.).
6. Home dashboard linked views — create them via create-view in this exact order so they land in order (linked views only append): Context (flat, no grouping) → Projects (Active) → Intelligence (recent) → Daily → People → Departments. Context must be the lead view.
7. Fill the CLAUDE.md master index with the real database map (each DB's Open link + data_source_id) once IDs exist.

Critical connector gotchas (all in the spec): URL property writes use the userDefined:URL key; linked views can only be created with create-view (the markdown <database data-source-url=…> insert is rejected) and they append, so build them in order; never create files or .obsidian artifacts.

N.4 — Onboard

Run the Phase B onboarding interview above (the 3 forms / 12 categories / brain-dump, plus the Phase B+ context drop). Collect everything into the same corpus. Skip Phase B's file-writing "Build Steps" — you'll write to Notion in N.5 instead.

N.5 — Personalize (write to Notion)

Using the corpus, write real content into Notion via MCP, following the same "real personalization, not placeholders" rules from Phase B:

• Context → one row per topic in the Context database (Operator, Organization, Market, Services, Pain Points, ICP, Brand, Team, Strategy, Infrastructure, Stakeholders — or the solo set), each with a Category + a real page body. Only create rows with content.
• Projects → rows in the Projects database (Status, Owner, Deadline…), with the Overview / Status / Next Steps body in each project page.
• People / Team → rows in the People database, with Reports to + relations to Projects.
• Departments → rows with a Lead relation to People (business mode).
• Intelligence / decisions → rows in the Intelligence database, linked to Project + People.
• First Daily entry → one row in the Daily database for today. Omit anything with no data — never leave bracket placeholders. Keep the home title clean (🏠 <Name> — Operating System, no "test"/builder wording).

N.6 — Generate the Notion OS Assistant skill

So the user can start using their OS immediately, generate a small auto-triggered skill called Notion OS Assistant (name: notion-os-assistant) that routes Notion requests. Use references/notion-os-skill-template.md:

• Give the user the SKILL.md — present the full, filled-in file in a code block so they can copy it. Most users upload it into Cowork or drop it into their skills folder themselves, so handing them the file is the default. You may also offer to save it to ~/.claude/skills/notion-os-assistant/SKILL.md if they have a local filesystem.
• Fill every {{PLACEHOLDER}} with the real values from this build: OS name, home page URL, CLAUDE.md page URL, and each database's data_source_id.
• Solo mode: delete the routing rows for databases that weren't created (Departments, Processes).
• Keep the description broad so it auto-triggers on second-brain / OS / projects / notes / meetings / people requests. The skill operates entirely through the Notion connector.

N.7 — Verify + finish

• Confirm the home page, the CLAUDE.md index, the Context database, and all databases exist and are populated; confirm the dashboard views are in order (Context first).
• Hand over the Notion OS Assistant skill (notion-os-assistant) SKILL.md — for upload to Cowork or saving under ~/.claude/skills/; once active they can ask about their OS in natural language and it auto-triggers.
• Give the user the link to the home page.
• Tell them: they now operate from inside Notion; Claude reads/writes via the connector, and should fetch CLAUDE.md first each session before touching data.
• Set expectations: Notion is realtime + multiplayer (unlike the file route), but it is cloud-only — there is no local copy on disk.

Guidelines

• Phase 0 is one question — mode selection
• Phase A is fully automated — no user input needed
• Phase B is 12 categories (Oskar's structure), batched into 3 rich-HTML forms rendered via mcp__visualize__show_widget (Cowork-only). Each form has 4 stacked categories with title, bullet inspiration, and a single free-text textarea per category. It's a guided brain dump, not a Q&A box. The bullets are inspiration, not strict asks. Always recommend Whisper / dictation + pasting docs / links / file paths into the textarea
• No follow-ups, no drilling deeper between forms
• Phase B+ is one final AskUserQuestion (or visualize widget) inviting any leftover files / links / folders — always ask, even if Forms 1–3 looked rich
• Accept any format: typed brain dumps, Whisper transcripts, pasted docs, uploaded files, links (LinkedIn, websites, blog posts, Notion, Drive), local folder paths, or skips
• For every link the user pastes, fetch it (WebFetch / WebSearch); for every file or folder, read it (Read / Glob); merge into a single context corpus before building
• Templates are scaffolds, not outputs. Replace every [bracketed placeholder] with real user data. If a section has no data after exhausting Q answers + corpus, omit the section — never leave placeholders in the written file
• Preserve specificity: use the user's exact names, numbers, URLs, and phrasing
• Only create context files that have real content — don't create empty placeholder files
• Don't narrate every file you're creating — just build it and summarize at the end