Why AgentKits Organizes Agents by Funnel Stage, Not by Tech Stack

The Default Way to Organize an Agent Repo

Open a large, well-known Claude Code agent repository and you’ll almost always find the same organizing principle: agents grouped by technical domain. wshobson/agents, one of the bigger multi-harness plugin marketplaces in the ecosystem, ships 203 agents, 175 skills, 109 commands, and 94 plugins across six coding harnesses (Claude Code, Codex CLI, Cursor, OpenCode, GitHub Copilot, Gemini CLI), tiered by which model should run them — Opus for architecture and security review, Sonnet for docs and testing, Haiku for fast, cheap operations. rohitg00/awesome-claude-code-toolkit takes a similar shape from a different angle: 135 agents across 10 categories, 176+ plugins, organized by directory type — agents/, skills/, plugins/, hooks/, rules/ — the way an engineering monorepo organizes itself.

Both are useful, and both are structured the way engineers naturally structure things: by what kind of technical work an agent does.

AgentKits — our open-source, MIT-licensed toolkit of AI agent kits for Claude Code, Cursor, and GitHub Copilot — takes a different axis for its flagship Marketing Kit, and it’s worth explaining why.

Organized by Funnel Stage, Not by Capability

The Marketing Kit’s 18 agents aren’t grouped by “what tool do they use” — they’re grouped by where a prospect sits in the marketing funnel:

• Top of Funnel (TOFU) — Attraction Specialist, SEO Specialist, Content Creator
• Middle of Funnel (MOFU) — Lead Qualifier, Email Wizard, Conversion Optimizer
• Bottom of Funnel (BOFU) — Sales Enabler, Copywriter, Campaign Manager
• Retention & Expansion — Continuity Specialist, Upsell Maximizer, Brand Voice Guardian

The commands follow the same logic: they’re namespaced by workflow, not by feature — /campaign:plan, /campaign:brief, /content:blog, /seo:keywords, /cro:page. A marketer reaching for a command reasons in terms of “I’m running a campaign” or “I need SEO copy,” not “which agent module handles content generation.”

Installation is opt-in and modular in the same spirit: a required Core module (18 agents, base commands, essential skills) plus optional modules for SEO, CRO, Content, Email, Analytics, and a Training module — 23 interactive modules and a pattern library for teams who want to onboard rather than just install.

On disk, none of this breaks from how Claude Code itself expects extensions to be laid out — agents, commands, and skills drop into .claude/agents, .claude/commands, and .claude/skills, the same native directory convention as everything else in the ecosystem. You can install via npx @aitytech/agentkits-marketing install (an interactive installer that detects your IDE and sets up persistent SQLite memory), clone the repo directly, or install as a Claude Code plugin. The kit supports Claude Code, Cursor, and GitHub Copilot as headline targets, with broader compatibility across Windsurf, Gemini CLI, Cline, Roo Code, and several others.

The Kit That’s Actually Live

Worth being precise here, since AgentKits is still filling out its portfolio: the Marketing Kit — 18 agents, dozens of namespaced commands, and skill packs covering SEO, CRO, email sequencing, and pricing strategy — is the one kit that’s fully shipped today. An Engineer Kit (17 agents, 40 skills) is in progress but not yet at the same maturity. Nine more kits — Sales, Legal, Finance, HR, Education, Creator, E-commerce, Healthcare, Real Estate — are planned, not built. Across the whole portfolio-in-progress, the aggregate headline is 12+ kits, 100+ agents, 200+ commands, and 50+ skills — but only one vertical is actually finished and usable end-to-end right now.

That’s a deliberate choice, not a gap we’re hiding: go deep on one function first, prove the funnel-stage organization actually holds up in real marketing workflows, then repeat the pattern for the next vertical.

A Memory Layer That Matches the Same Philosophy

The Marketing Kit doesn’t work in isolation from context. A companion project, AgentKits Memory, is a persistent memory system for AI coding assistants that runs entirely as local SQLite storage over MCP — no server, no account, no cloud sync. It exposes exactly five MCP tools (memory_save, memory_search, memory_recall, memory_list, memory_status), supports optional HNSW-based semantic search, and ships hooks that auto-capture session context so an agent remembers a brand’s voice or a campaign’s history across sessions instead of starting cold every time. It’s a small, sharp tool rather than a general-purpose vector database — consistent with how the rest of the kit is built.

Depth-First in a Breadth-First Market

The wider “awesome-claude-code” ecosystem is starting to split into two recognizable strategies. One is breadth-first: ship as many agents, skills, and plugins as possible on day one, and let users assemble their own subset — that’s the wshobson/agents and awesome-claude-code-toolkit approach, and it shows in raw numbers (200+ agents, 176+ plugins). The other is narrower and vertical-specific: Agensi’s 2026 landscape analysis describes a parallel trend toward curated SKILL.md and MCP registries with quality and security review baked in, and tools like NanoSkill.ai focus purely on marketing-category skills — SEO, ads, email, lead gen, content — rather than trying to cover every technical domain at once.

AgentKits’ Marketing Kit sits closer to that second camp: fewer agents than the largest general-purpose repos, but organized around how a marketing team actually thinks about its own funnel, with a memory layer built specifically to remember brand voice and campaign history rather than arbitrary code context. Whether business-function-first organization outlasts raw-count breadth as an adoption strategy for non-developer teams is an open question — but it’s the bet this kit is making, one vertical at a time.

---

AgentKits is open source and free forever under the MIT license. Explore the Marketing Kit, install it in your own project, or follow what’s next at agentkits.net, or reach us at [email protected].