bootstrap

name	bootstrap
description	Fastest way to get started — give a company website URL and agents crawl it to populate context.md. Use when operator wants to set up fast without answering 20 questions.
argument-hint	[company website URL, e.g. 'acme.com' or 'https://www.acme.com']

Bootstrap the GTM Context OS from a company's website. Agents crawl key pages, extract business context, and populate context.md — all tagged as marketing claims, not verified truth.

When to Use

• First thing after cloning: "here's our website, get me started"
• Faster alternative to /quickstart when the operator doesn't want a Q&A session
• Anytime: "fill in context.md from our site"

Core Principle

A marketing website is a curated performance, not ground truth. Everything extracted gets [CLAIMED: website] attribution. The operator can promote claims to [VERIFIED] later with sales data, PULL analyses, or direct confirmation.

Don't be cynical about it — website content is useful starting context. Just be honest that it's the company's own positioning, not independent evidence.

Process

Step 1: Normalize the URL

Accept any format: acme.com, www.acme.com, https://acme.com. Normalize to https:// base URL.

Step 2: Crawl Key Pages

Fetch these pages in parallel (use agents or concurrent web fetches). Not all will exist — that's fine, work with what you get:

Page	What to extract	Typical URL patterns
Homepage	Value prop, tagline, key features, social proof	/
About	Company story, team size, founding date, HQ, mission	/about, /about-us, /company
Product / Features	Core product, feature list, platform details	/product, /features, /platform, /solutions
Pricing	Pricing model, tiers, plan names	/pricing, /plans
Customers / Case studies	Customer logos, industries served, company sizes	/customers, /case-studies, /stories
Careers	Team size signal, HQ/offices, culture, tech stack	/careers, /jobs, /open-positions
Integrations	Tech ecosystem, required tools, partnership signals	/integrations, /apps, /marketplace
Blog (recent)	Topics they care about, thought leadership angles	/blog (just the index, not individual posts)

Crawl strategy:

• Start with homepage — it usually links to the other key pages
• Follow navigation links to find the right URLs for each category
• If a page doesn't exist at expected URLs, skip it — don't guess
• Don't crawl more than 10-12 pages total. Breadth over depth.
• For case studies / customers pages, skim for company names and industries — don't read full case studies

Step 3: Extract and Attribute

For each page, extract structured facts into these buckets:

Company basics:

• Name, tagline, what they do (1-2 sentences)
• Stage signals (funding mentions, team size, customer count)
• Founded date, HQ location
• Any explicit numbers (ARR, customer count, team size)

Product:

• Core product description
• Key features (top 5-8, not exhaustive)
• Platform / tech details
• Pricing model and tiers (if public)

ICP signals:

• Customer logos → infer company sizes, industries
• Case study titles → infer buyer roles, triggers
• "Built for..." or "Designed for..." copy → explicit ICP claims
• Pricing tiers → infer target company size (self-serve = SMB, "contact sales" = enterprise)

Positioning:

• Value proposition (usually on homepage hero)
• Comparison pages or "vs." content → competitors
• "Why us" or differentiation copy
• Categories they claim (e.g., "the AI-native CRM")

Competitor signals:

• Named alternatives on comparison pages
• Integration partners (sometimes reveal the ecosystem)
• Category language → who else is in that category

Tag everything: [CLAIMED: website]

Step 4: Populate context.md

Fill in the context.md template with extracted data. For each section:

• Company — Fill from about/homepage. Straightforward.
• Product — Fill from product/features/pricing pages.
• ICP — Fill what you can infer. Firmographic criteria from customer logos and pricing. Buyer titles from case studies. Leave buying patterns mostly empty — websites don't reveal how people actually buy, only what the company wishes they'd do.
• Positioning — Fill from homepage/about. Tag as [CLAIMED: website].
• Competitors — Fill if comparison pages exist. Otherwise leave a note: "No competitor data found on website — fill in from sales experience."
• Disqualification rules — Usually can't determine from a website. Leave template with a note.

Step 5: Gap Analysis

After populating, report what's solid and what the website couldn't tell you:

## What I Learned From the Website

### Filled In
- Company basics: [what you found]
- Product: [what you found]
- Positioning: [what you found]

### Gaps — Needs Your Input
- **Buying patterns** — The website says who uses the product but not 
  how they find and evaluate it. How do prospects typically discover you? 
  What's the usual sales cycle?
- **ICP hard constraints** — I can see who your customers are, but not 
  who ISN'T a fit. What disqualifies a prospect?
- **Competitors from the buyer's perspective** — [Found X on comparison pages / 
  No comparison content found]. Who do prospects actually mention on calls?
- **Demand triggers** — Marketing copy describes benefits, not the 
  situations that make someone buy NOW. What's usually happening when 
  someone reaches out?

Frame gaps as things the website structurally can't answer, not as failures. Suggest the operator fill these in directly or run /quickstart to do it conversationally.

Step 6: Confirm and Next Steps

Show the populated context.md to the operator. Then:

"Your context is bootstrapped from the website. Everything is tagged [CLAIMED: website] — it's useful starting context but reflects your marketing, not verified buyer behavior.

To strengthen it:

• Paste a sales call transcript — I'll run a PULL analysis that tests these claims against real buyer language
• Use /quickstart to fill in the gaps I flagged (buying patterns, triggers, disqualifiers)
• Use /intake if you have existing documents (pitch decks, playbooks, competitor analyses)"

Update status.md with a bootstrap entry.

What This Skill Does NOT Do

• Does not verify any claims — everything is [CLAIMED: website] until confirmed
• Does not build segments, messaging, or campaigns — those need demand evidence
• Does not replace /quickstart — quickstart covers the human knowledge a website can't surface
• Does not deeply analyze blog content or case studies — that's /intake territory