/seo obsidian-kb

Build an Obsidian knowledge base from any product website's sitemap. Uses Python for fast bulk-fetching and parallel Claude agents for quality content analysis and MD generation.

Usage

/seo obsidian-kb --sitemap <URL> --product "<Name>" [--vault <path>] [--base-path <path>]

Parameters:

• --sitemap (required): Sitemap URL (e.g., https://www.zoho.com/campaigns/sitemap.xml)
• --product (required): Product name for folder and tagging (e.g., "Zoho Campaigns")
• --vault (optional): Obsidian vault path. Default: ~/Documents/Obsidian Vault
• --base-path (optional): URL base path prefix (e.g., /campaigns/). Auto-detected if omitted.

What You Must Do When Invoked

Follow these steps in order. Do not skip steps.

Step 1 — Ensure dependencies are installed

python3 -c "import aiohttp, bs4, lxml" 2>/dev/null || python3 -m pip install aiohttp beautifulsoup4 lxml -q --break-system-packages

Step 2 — Run the Python fetcher

python3 ~/.claude/skills/seo-obsidian-kb/fetch_pages.py \
  --sitemap "SITEMAP_URL" \
  --output /tmp/seo-obsidian-kb-pages.json \
  --base-path "BASE_PATH" \
  --concurrency 15

Replace SITEMAP_URL and BASE_PATH with the user's values (or omit --base-path for auto-detection).

Read the output JSON silently. Present a clean summary:

Fetched: X / Y pages
Categories: {category: count, ...}
Domain: example.com

If total_pages is 0: stop with "No pages found. Check the sitemap URL." If total_pages > 300: warn the user and ask if they want to proceed or limit scope.

Step 3 — Create vault structure

Derive the product folder name from --product by lowercasing and hyphenating (e.g., "Zoho Campaigns" → "zoho-campaigns").

Create directories:

{vault}/
  _HQ/                    # Only if it doesn't exist
    _SEO Command Center.md
    _Templates/
  {product-folder}/
    {category}/            # One folder per category from the JSON

If _HQ/_SEO Command Center.md already exists, update it to add the new product row — do NOT overwrite.

Step 4 — Generate MD notes with parallel agents

This is the critical quality step. Split pages into batches and dispatch parallel agents.

Batch size: ~20 pages per agent. Model: Use sonnet for the worker agents (fast + quality balance).

For each agent, provide this exact prompt structure:

You are generating Obsidian knowledge base notes for the product "{product_name}".
Product folder: {product-folder}
Vault path: {vault_path}

For EACH page in the batch below, create a markdown file at:
  {vault_path}/{product-folder}/{category}/{slug}.md

Use this EXACT frontmatter + body template:

---TEMPLATE START---
---
product: {product-folder}
category: {category}
slug: {slug}
alias: "{human-readable title derived from h1 or title}"
url: "{url}"

page_type: {classify as: feature|pricing|comparison|alternative|case-study|landing|blog|integration|docs|homepage}
primary_intent: {classify as: informational|commercial|navigational|transactional}
funnel_stage: {classify as: tofu|mofu|bofu}
target_audience: "{infer from page content and type}"

title: "{title}"
h1: "{h1}"
word_count: {word_count}
has_schema: {true|false}
schema_types: [{list}]

tags:
  - {product-folder}/{category}
  - page-type/{page_type}
  - intent/{primary_intent}
  - status/live
---

# {title or h1}

**URL:** [{url}]({url})
**Product:** [[{product-folder}/_{product_name}|{product_name}]]

## About This Page
{Write 2-3 sentences describing what this page is about, its role in the site, and who it serves. Use the heading outline and meta description to inform this.}

## Page Sections
{Format the headings array as a structured outline:}
### {H2 heading}
- {H3 sub-heading}
- {H3 sub-heading}

## Key Takeaways for LLM Context
{Write 5-8 bullet points summarizing the most important facts Claude should know about this page. Focus on: what the page offers, key features mentioned, CTAs, unique value propositions.}

## Internal Links (Outbound)
{For each resolved_outbound_link, write:}
- [[{target_slug}|{target_title}]] - anchor: "{anchor_text}"

## Internal Links (Inbound)
{For each inbound_link, write:}
- [[{from_slug}|From: {from_category}/{from_slug}]] - anchor: "{anchor_text}"
---TEMPLATE END---

IMPORTANT RULES:
1. page_type classification: pricing pages = "pricing", feature descriptions = "feature", X-vs-Y = "comparison", customer stories = "case-study", product landing = "landing", help/docs = "docs", blog articles = "blog", integration pages = "integration", homepage/index = "homepage"
2. primary_intent: pricing/plans = "commercial", features/how-it-works = "informational", login/signup = "transactional", brand/homepage = "navigational"
3. funnel_stage: awareness/educational = "tofu", evaluation/comparison = "mofu", purchase/signup/pricing = "bofu"
4. Wikilinks use the slug only (Obsidian resolves by filename). For cross-category links, use category/slug format.
5. Write the actual file to disk using the Write tool.
6. If a page has 0 headings, write "No structured headings detected on this page." in Page Sections.
7. Deduplicate inbound/outbound links (same slug appearing multiple times with different anchors — keep the first occurrence).

PAGE BATCH:
{paste the JSON array of page objects for this batch}

Dispatch all agents in a single message for maximum parallelism.

Wait for all agents to complete. Count total files written.

Step 5 — Generate category indexes

For EACH category folder, create _{category}.md:

---
product: {product-folder}
category: {category}
type: category-index
alias: "{Category Name}"
tags:
  - {product-folder}/{category}
  - type/index
---

# {Category Name}

> {page_count} pages in the {category} category for {product_name}

## Pages

| Page | Type | Intent | Words |
|------|------|--------|-------|
{for each page in this category: | [[{slug}\|{alias}]] | {page_type} | {primary_intent} | {word_count} |}

Step 6 — Generate product MOC

Create {product-folder}/_{product_name}.md:

---
product: {product-folder}
type: product-moc
alias: "{product_name}"
domain: "{domain}"
tags:
  - {product-folder}/moc
  - type/moc
---

# {product_name} Knowledge Base

> {total_pages} pages across {category_count} categories

## Categories

| Category | Pages | Index |
|----------|-------|-------|
{for each category: | {category} | {count} | [[_{category}\|View]] |}

## All Pages by Type

```dataview
TABLE page_type AS "Type", length(rows) AS "Count"
FROM "{product-folder}"
WHERE !contains(tags, "type/index") AND !contains(tags, "type/moc")
GROUP BY page_type
SORT length(rows) DESC

### Step 7 — Update master MOC

If `_HQ/_SEO Command Center.md` exists, add the new product to the Products table.
If it doesn't exist, create it:

```markdown
---
type: master-moc
alias: "SEO Command Center"
tags:
  - type/moc
---

# SEO Command Center

## Products

| Product | Folder | Pages | Status |
|---------|--------|-------|--------|
| [[{product-folder}/_{product_name}\|{product_name}]] | `{product-folder}/` | {total_pages} | Active |

## How to Use

1. **Navigate by product** — click a product above to see its MOC
2. **Graph view** — each product is color-coded as a distinct cluster
3. **Search** — use Obsidian search or Dataview queries on frontmatter
4. **Claude integration** — Claude reads this vault for context, uses Ahrefs/GSC MCPs for live metrics

Step 8 — Quality verification

After all files are written, verify:

1. Count files: find {vault}/{product-folder} -name "*.md" | wc -l — should match total_pages + category_indexes + 1 (product MOC)
2. Spot-check 3 random pages: Read them and verify frontmatter is complete (all fields populated, no empty strings where data existed)
3. Check for broken wikilinks: Grep for [[ patterns and verify target files exist
4. Report summary:

=== SEO Obsidian KB Complete ===
Product:    {product_name}
Vault:      {vault_path}
Pages:      {count} notes created
Categories: {count} ({list})
MOC:        {product-folder}/_{product_name}.md
Graph tip:  Open Obsidian > Graph View > filter by tag:#{product-folder}

Frontmatter Schema Reference

Field	Type	Source	Description
product	string	CLI arg	Product folder name
category	string	Python (URL path)	Category derived from URL
slug	string	Python (URL path)	Page slug from URL
alias	string	Agent (from h1/title)	Human display name
url	string	Python (sitemap)	Full page URL
page_type	string	Agent (classified)	feature/pricing/comparison/etc
primary_intent	string	Agent (classified)	informational/commercial/etc
funnel_stage	string	Agent (classified)	tofu/mofu/bofu
target_audience	string	Agent (inferred)	Who this page serves
title	string	Python (HTML)	Title tag content
h1	string	Python (HTML)	H1 tag content
word_count	int	Python (HTML)	Body text word count
has_schema	bool	Python (JSON-LD)	Schema.org markup detected
schema_types	list	Python (JSON-LD)	Schema types found
tags	list	Agent (generated)	Obsidian tags for graph/filter

Notes

• No SEO metrics stored — Claude fetches live data from Ahrefs/GSC MCPs when needed
• Vault is additive — running for a new product adds to existing vault, never overwrites other products
• Run /graphify after for community detection and surprise connections across pages
• Templates at _HQ/_Templates/ can be used for manual note creation in Obsidian