Last Thursday I was stuck waiting for a client call that never happened. Forty-five minutes blocked on my calendar, nobody showed, and I had a browser with 11 tabs open and nothing useful to do. So I started poking around GitHub.
I found a repo called ai-marketing-claude — a full marketing analysis toolkit built on Claude Code. Fifteen slash commands. Six parallel agents. PDF output. I ran it on Calendly just to see what would happen, and when the report landed I stopped what I was doing and read the whole thing.
It was good. Like, actually good.
How to install Claude Code for beginners
You type /market-audit https://anysite.com and the agents take over. No config, no scraping setup, no API keys to juggle beyond what Claude Code already uses. The toolkit fetches the homepage, maps the site architecture, identifies the business type — and then launches six subagents in parallel to tear the site apart from different angles simultaneously.
That parallel execution is the part that matters. A single sequential audit takes forever. Six agents running at the same time — one on content, one on SEO, one on brand, one on competitive positioning, one on conversion rate elements, one on growth strategy — finish in 2–5 minutes total. I ran it on a medspa site I'd driven past the week before. The site had broken booking buttons and a homepage hero that said nothing meaningful. The audit scored it 29/100 on content and messaging. Brutal — but accurate.
Every audit produces a score across six categories:
Content & Messaging, Conversion Optimization, SEO & Discoverability, Competitive Positioning, Brand & Trust, Growth & Strategy.
Each gets a numeric score out of 100, a weight (Conversion Optimization carries 30% of the overall), and a status label — Good, Average, or whatever the equivalent of "fix this immediately" is. The full breakdown goes into both a markdown file (MARKETING-AUDIT.md) and a PDF with a cover page, findings section, action plan, and SWOT analysis for identified competitors.
I ran it on my own site at one point (49/100 overall, don't ask) and on three competitor sites I'd been meaning to look at for months. Total time: about 90 minutes including reading the reports. Total cost: $0.61 in API credits. (For comparison, the last manual competitive analysis I built in Notion took me two full days.)
Claude Code skills and automation workflows
The main audit is just the start. The repo ships with 14 other slash commands, each with its own SKILL.md playbook that tells Claude exactly how to approach the task:
/market-copy [url] generates optimized copy with landing page examples. /market-emails [url] writes a full email sequence. /market-ads [url] does creative and copy for all platforms. /market-intel [competitor] runs a competitive intelligence report. /market-proposal [url] generates a client proposal you can actually send.
That last one is interesting. You run an audit on a prospect's site, then immediately run the proposal command — and you have a document ready to send cold, before you've had a single conversation with them. The PDF looks polished. I was skeptical, so I tested it on three fake prospects I made up. Two of the three outputs I would have paid a junior copywriter $80–100 to produce.
/market-landing [url] specifically focuses on CRO — conversion rate optimization elements — which is usually the most actionable thing to tell a small business. Most local business sites have catastrophically bad landing pages and no idea.
Here's the honest part: web crawling is messy. Pages 404. JavaScript-heavy sites don't render the same way. Redirects happen mid-fetch.
The toolkit handles errors by logging them and continuing. If a page fails, the agent marks it and moves to the next one. The audit doesn't stop because one URL bounced — it keeps going and notes the failure in the report. That's the right behavior. Plenty of tools just crash.
That said, heavily JavaScript-rendered sites are still a weak point. If a site loads its entire content client-side (SPAs, heavy React apps), the crawler is going to miss things. It's not a dealbreaker for 80% of business sites — most local businesses and B2B tools have server-rendered pages — but it's worth knowing before you run it on something like a complex SaaS dashboard.
One curl command. It drops all 15 skills into your Claude Code setup. If you're running the VS Code extension (search "Claude Code" in Extensions, install the official Anthropic one), everything shows up immediately. Terminal users get the same setup.
The GitHub README has both options — Command Install for the curl route and Local Install if you want to clone and modify. The repo is fully open source, MIT license, so you can fork it, break it, add commands, do whatever.
Setting up Claude Code with VS Code extension
Agency builders who want to sell audits as a service. Freelancers who do marketing consulting and spend too much time on the deliverable and not enough on the actual thinking. Founders who want a fast gut-check on a competitor before a sales call.
I ran it on a local aesthetics clinic — Nob Hill Aesthetics, a medspa I passed on my walk to the office — on a Thursday morning just to test it on a real business outside tech. The report came back with specific callouts about their booking flow, their social proof gaps, and a competitor comparison with two other local medspas I didn't know existed. The kind of research I'd normally spend two hours on.
The PDF output is the thing that makes this useful for agency work specifically. You generate the audit, generate the proposal, attach the PDF, and send it cold. No cover letter needed. The report explains itself.
The average agency charges $300–$1,500 for a marketing audit this comprehensive. The average API cost to produce it is $0.08.
That gap is where the business model lives.
