Init

Run 5 prompts via AskUserQuestion, then generate all files.

Step 1: Project Name

Ask the user:

What is the project/wiki name?
This becomes the vault root directory name and appears in CLAUDE.md.
Example: "ml-research", "company-wiki", "reading-notes"

Store as $PROJECT_NAME.

Step 2: Page Types

Ask the user which page types to include. Present the default 10 with checkboxes:

Which page types should the wiki support?
Default set (all selected):

1. concept     — Core ideas, definitions, mental models
2. entity      — External tools, technologies, protocols
3. architecture — System designs, data flows, infrastructure
4. decision    — ADRs, trade-off analyses, chosen paths
5. strategy    — Plans, roadmaps, positioning
6. org         — Team structure, processes, culture
7. comparison  — X vs Y evaluations, benchmarks
8. open-question — Unresolved problems, hypotheses
9. source-summary — Summaries of individual sources
10. synthesis   — Cross-source analyses, meta-insights

Enter numbers to toggle off, or "all" to keep defaults.
You can also add custom types (e.g., "tutorial", "glossary").

Store as $PAGE_TYPES array.

Step 3: Raw Source Directories

Ask the user which raw source directories to create:

Which raw source directories do you need?
Default set:

1. knowledge — Articles, papers, reference material
2. specs     — Specifications, requirements, RFCs
3. meetings  — Transcripts, minutes, recordings
4. docs      — Official documentation, manuals
5. external  — Third-party content, scraped pages
6. inbox     — Temporary landing zone for batch captures (PRs, git logs)

Enter numbers to toggle off, "all" for defaults,
or add custom directories (e.g., "podcasts", "code-reviews").

Store as $RAW_DIRS array.

Step 4: Capture Tools

Ask the user which capture tools are installed:

Which capture tools are available on this machine?
These determine which raw/ ingestion instructions go into CLAUDE.md.

1. defuddle — Web article extraction (HTML to clean markdown)
2. yt-dlp   — YouTube/video transcript download
3. pandoc   — Document conversion (PDF, DOCX, EPUB to markdown)

Enter numbers for installed tools, or "none".

Store as $CAPTURE_TOOLS array.

Step 5: Confirmation

Present the full plan and ask for confirmation:

Ready to create:

  $PROJECT_NAME/
  ├── CLAUDE.md              (schema + instructions)
  ├── wiki.config.md         (vault settings, YAML frontmatter)
  ├── index.md               (vault entry point)
  ├── log.md                 (change log)
  ├── .gitignore
  ├── .obsidian/             (minimal Obsidian config)
  ├── wiki/                  (LLM-maintained pages)
  └── raw/                   (immutable sources)
      ├── $RAW_DIR_1/
      ├── $RAW_DIR_2/
      └── ...

Page types: [list selected types]
Capture tools: [list selected tools]

Proceed? (yes / edit / cancel)

If "edit", loop back to the relevant step. If "cancel", abort.

File Generation

After confirmation, create all files in order.

1. Directory Structure

$PROJECT_NAME/
├── wiki/
├── raw/
│   ├── $RAW_DIR_1/
│   ├── $RAW_DIR_2/
│   └── ...
└── .obsidian/

2. CLAUDE.md

Copy skills/init/templates/claude-md-template.md and substitute:

• $PROJECT_NAME — the project name from Step 1
• $PAGE_TYPE_TABLE — a markdown table row per selected page type (type | description | use-when)
• $CAPTURE_TOOLS_SECTION — instructions only for tools confirmed installed in Step 4:
  • defuddle: npx defuddle "$URL" > raw/external/$(date +%Y-%m-%d)-slug.md
  • yt-dlp: yt-dlp --write-auto-sub --sub-lang en --skip-download -o "raw/external/%(title)s" "$URL"
  • pandoc: pandoc -s input.pdf -t markdown -o raw/docs/output.md

If no tools selected, write: "No capture tools configured. Add raw sources manually as markdown files."

3. wiki.config.md

---