← All articles
Sep 5, 2026

headcount Turns Claude Code Into an Org Chart of 125 Skills

headcount adds 15+ departments and 125+ skills to Claude Code, so specialized agents handle marketing, finance, and security instead of one generalist.

Fifteen departments. A hundred and twenty-five skills. That's the shape of headcount, an open-source project that's picked up 1,149 GitHub stars in under a week by treating Claude Code less like a chat window and more like a company you're staffing. Its tagline says the whole idea in five words: "Add a department, not a prompt."

The project is maintained by developer Chris Brock, released under the MIT license, and actively updated, its most recent commit landed the same day this article was researched. That combination, real traction, a permissive license, and a maintainer still pushing code, is worth noting up front, because most week-old repos don't clear any of those three bars at once.

What headcount adds to Claude Code

Claude Code already lets you write custom skills and install plugins. headcount's contribution is structure: it organizes that capability into 15-plus departments, each one a self-contained plugin covering a real business function. Marketing carries 19 skills. Technology carries another 19, covering architecture, infrastructure, and code quality review. Security holds 8 reviewer-class skills built around risk assessment and threat analysis. Finance, People, Revenue, Legal, Product, and half a dozen more round out the rest, adding up to more than 125 skills across the whole set.

Each department installs independently through Claude Code's plugin marketplace:

/plugin install security@headcount

You don't have to install all 15 to get value. A one-person operation might only ever need Marketing and Finance; a small dev shop might install Technology and Security and skip the rest. That's a deliberate design choice, not a limitation, headcount is built to be picked apart rather than adopted whole.

How the skills actually activate

Once a department is installed, its skills work in one of two ways. You can call one directly by name, /finance:financial-modeling for a specific unit-economics pass, for instance, or you can just describe what you need in plain language and let the relevant skill auto-load based on what your request is actually about. In practice this means you don't need to memorize a command list the way you would with a smaller, flatter skill set. You ask for a pricing model and the Finance department's skill picks it up; you ask for a threat assessment and Security's does.

The more interesting engineering choice sits underneath that: headcount enforces exclusive write surfaces per department, meaning each one owns a defined set of files and doesn't step on another department's output. That's the actual answer to the org-chart metaphor holding up under real use. A flat pile of 125 skills with no boundaries would just be 125 ways to collide with each other on a shared codebase; giving each department its own lane is what makes running more than one of them at the same time on a single task something you can trust instead of something you have to babysit.

There's also a browsable interactive org chart on the project's GitHub Pages site, useful less as a gimmick and more as a way to actually see what you're installing before you commit to a department, since scrolling a README table doesn't give you the same sense of scope that a chart does.

Who this is actually for, and who should skip it

headcount is squarely built for someone whose Claude Code work has outgrown a single running conversation, an operator handling marketing copy, financial modeling, and codebase review all in the same week, without the budget to hire three specialists to do it. If that's not your situation, this isn't the tool to reach for first. A single developer working on one focused feature branch doesn't need a Legal department or a People department installed, and the overhead of deciding which of 15 departments to add is real overhead, even if each individual install is a one-line command.

It's also worth being honest about what a week-old project with a fictional-sounding backstory can and can't promise. There's no independent case study yet of a team running this at scale over months, and "125-plus skills" is a count of what's shipped, not a guarantee that any given skill is more polished than a prompt you'd write yourself for the same task. What headcount actually offers isn't magic quality, it's organization: a consistent place to add the next specialized capability instead of stuffing another instruction into an already-long system prompt.

Staffing an agent instead of prompting one

headcount is one entry in a pattern this site keeps tracking, Claude Code moving from a single generalist agent toward something closer to a staffed team, each member scoped to one job with its own boundaries. If your Claude Code setup has started to feel like one overworked assistant juggling every request, that's the actual signal to look at something structured like this rather than adding one more ad hoc instruction to a prompt that's already too long.

Sources: headcount on GitHub, including its README, live star count, and license; GitHub API repository metadata confirming creation date, license, and latest push.

Join the newsletter

AI workflows and systems, straight to your inbox.

No spam. Unsubscribe anytime.