Farhan ArshadAI systems
The agency runs on skills I wrote.
Thirty-five custom Claude skills, agent workflows that watch client accounts every morning, and eight connections that put Claude inside the tools the work already lives in. Every screenshot on this page is a real file from the workspace, with client names removed.
What a skill is
A folder, not a prompt
A prompt lives in one person's head and dies when they leave. A skill is a folder: a SKILL.md that holds the workflow, the rules, the approval gates and the output format, plus reference files the model loads only when a phase needs them. Anyone on the team opens the skill and gets the same process, the same voice and the same quality checks, without knowing how to prompt.
Selected systems
Five systems, what they replaced, what happened
| System | Problem | Built | Result |
|---|---|---|---|
| Ecommerce content engineecommerce-content-engine | Store blog content took a researcher, a writer, an editor and a developer, and still read like AI. | A five-file pipeline: live research, ideation with two human approval gates, practitioner-voice drafting, HTML build with JSON-LD schema and author entities, publish to Shopify when connected, scheduled refresh. | One person runs the whole cycle for owned and client stores. The same writing engine sits behind a 56-article content engine for a client. |
| Daily AI health checkn8n + Zapier + Claude agents | Ad accounts drift quietly. By the time a client notices cost per lead has doubled, the relationship is already damaged. | Agents pull Meta Ads and Search Console every morning, check guard rails (frequency, cost per result, spend caps, learning resets) and draft the client update with the fix already proposed. A human reads and sends. | UAE logistics client: 486 conversations in seven days at AED 0.63 each. When cost climbed to AED 1.36 the report flagged it first, and the pilot became a retainer. Write-up. |
| GEO audit skillgeo-audit, four engines | Brands know their Google rankings and nothing about how ChatGPT, Claude, Perplexity or Google AI Mode describe them. | A skill that builds a buyer-prompt set from the site, runs it across the four engines, scores mention, recommendation, share of voice and citation source, and writes a fix playbook. | UK activewear brand: 100% mention, 32% share of voice, only 17% of citations from its own domain, cause found in an afternoon. Write-up. |
| Monthly SEO reportingclient-seo-reporting | Half a day per client per month turning Search Console exports into a report a client would read. | Export in; month-over-month keyword movement, click-through opportunities and pages with impressions but no clicks out, written in the agency's voice as a client-ready document. | Reporting time dropped to the time it takes to read the draft. For a Toronto pest-control client the zero-click detection became a 20-article plan. |
| UGC video factoryhiggsfield-content-factory | Short-form ad creative was the slowest, most expensive part of every launch. | A five-stage pipeline on the Higgsfield MCP: trend research, a content plan across five UGC formats, batch video generation, an image pack, scheduling to Meta Ads, and a cost comparison against traditional production. | Runs for Pet Buckets, the agency's own Shopify brand; button-driven so a non-technical operator can run a launch. |
Want one of these running for your business?
A morning report on your ad account, an audit of what ChatGPT says about you, or a skill written for your team. Tell me what you are trying to do and I reply myself within one working day.
Built in public
Tools anyone can use
Some of the work is public. The customer lifetime value calculator is a single HTML file, built with one prompt, that shows a business owner what one customer is worth before asking for anything. Try it, or read how it was built.
The library
35 skills, grouped by the job they do
Every name below is a real folder synced to the agency's Claude workspace. Four skills built for personal projects are counted but not listed.
| Group | Skill | What it does |
|---|---|---|
| SEO and authority | seo-content-writer | live SERP research, verified entities, banned-word QA |
seo-content-strategy | monthly plan from Search Console data and clusters | |
client-seo-reporting | export to client-ready monthly report | |
local-seo-schema | JSON-LD, GBP plans, NAP audits, entity stacking | |
local-authority-stack | full citation and signal campaign with tracker | |
global-authority-stack | researched off-page plan for non-local sites | |
gmb-crush-content-refresh | client-specific refresh workflow | |
| Shopify and ecommerce | ecommerce-content-engine | research to publish, five files |
shopify-brand-ops | store audits, PDP copy, CS templates, ad copy | |
clone-link-to-my-shopify | any landing page to editable Liquid sections | |
digital-product-hunter | low-saturation digital product research | |
ebook-builder | designed multi-page PDF products | |
ebook-pdp-image-builder | nine-image branded product gallery | |
| Agency operations | agency-proposal-writer | proposals, SOWs, tenders, retainers |
cv-architect | research-backed CVs with a deterministic ATS checker | |
ui-ux-designer | design direction, tokens, working prototypes | |
website-design-director | creative direction before any code | |
blog-featured-image-generator | branded 16:9 editorial images | |
| Video and paid creative | higgsfield-content-factory | five-stage UGC pipeline on MCP |
ebook-fb-ad-script | paid social scripts for digital products | |
ebook-fb-ad-video | script to generated video in Marketing Studio | |
yt-idea-radar | structured idea hunt against channel medians | |
yt-thumbnail-system | thumbnail house system and test log | |
yt-video-packaging | titles, descriptions, chapters, A/B plan | |
| Business planning | pd-1-person-business | offer, lead gen and monetisation plan with live research |
pd-ai-business-idea-generator | idea generation against real demand | |
pd-listbuilder | list-building system | |
pd-personalbrand-launchpad | personal brand launch plan | |
pd-youtube-niche-finder | niche selection from channel data | |
pd-youtube-script-writer | long-form script structure | |
| Also running | searchfit-seo | plugin, 11 skills: audits, clustering, schema, internal links |
design | plugin: critique, handoff, accessibility review | |
marketing | plugin: campaign plans, email sequences, brand review |
Connections
Claude connected to where the work lives
Shopify Admin over MCP: products, collections, orders, analytics, digital products
Meta Ads MCP: campaign builds, creative upload, insights, catalog, scheduling
Self-hosted Search Console MCP: analytics, indexing checks, sitemaps, period comparisons
Higgsfield MCP for image, video and voice; Figma MCP for design context and generation
Hostinger MCP for hosting, DNS and VPS; n8n self-hosted for the agent workflows moving off Zapier
n8n and Zapier for triggers and hand-offs; Claude Code for builds; Cowork for the team-facing skills
Operating rules
How I build
Research first, then the model writes
Every fact in an output traces to a source the skill fetched. The model never gets to invent a number for a client.
Automation watches, people decide
Agents draft, flag and propose. A human keeps the send button. That is why clients trust the morning report.
Ship small, measure adoption
One team, one task, one skill. If nobody opens it a second time, the skill was wrong, not the team.
Vet what the agent can reach
Scoped permissions, no credentials in prompts, third-party skills read before they run.
Contact
Say hello. I answer myself.
For a role, a consultation on AI search visibility, or an automation build. Tell me what you are trying to do and I will reply within one working day.
Or skip the form: WhatsApp me for a quick answer, or write to hello@farhanarshad.com if you would rather explain properly.