headcount: 16 departments and 172 specialists inside Claude Code, free

# headcount

Every solo builder makes the same trade a hundred times a week without noticing
it. A decision comes up that belongs to somebody with a title — is this schema
safe to expose, is this price above or below cost, does this clause mean what it
looks like it means — and it gets made by the only person in the room. Who is
you, and who has never done that job.

headcount is one answer to that. It is an agent organization for Claude Code,
structured as a company: a chief executive over 16 departments, 172 skills,
MIT licensed and free. Each department is an independently installable plugin,
so a project loads only the specialists it needs rather than all of them at once.
Skills address as department:skill — security:threat-modeling,
finance:unit-economics — and load themselves when a question lands in their
territory.

## Install it in two lines

The first line is once per machine. The second is once per department, per
project.

/plugin marketplace add cbrock84/headcount
/plugin install security@headcount

There is no account, no API key and no paid tier. Both lines are typed inside
Claude Code itself.

Once a department is installed, you do not invoke anything. Ask a question in
its territory and the right specialist engages:

You can also call one by name when you want a specific lens:
/finance:financial-modeling.

These counts are from a read-only clone taken on 2026-09-04 —
find plugins -name SKILL.md — not from the badges in the README. The badges
happen to be honest, which is worth saying, because plenty are not.

172 skills across 16 departments, 9.3 MB on disk.

headcount is 172 markdown files. In the entire repository, 43 files are not
markdown, and nearly all of those are its own build and lint scripts — the
product itself is prose.

That is not a criticism, but it does decide what you should expect. The skills
tell Claude how a particular specialist frames a problem: what a threat model
asks first, what unit economics checks before it answers. They do not run
anything, they do not call an API, and they do not add knowledge the model does
not already have. If Claude cannot do the job, a folder will not teach it.

What a good skill file does is stop the model from answering as a generalist —
which, for the kind of question you were about to answer alone, is most of the
value.

Installing all sixteen into every project is the one thing the project's own
README tells you not to do. So the kit above contains deptscan.py: it walks
your project three levels deep, reads path names only — it never opens a
file — and prints the /plugin install lines your own tree argues for, ordered
by evidence.

python3 deptscan.py            # scans the current directory
python3 deptscan.py ~/code/app # scans somewhere else
python3 deptscan.py --all      # all 16, with their skill counts

A repo with a Dockerfile gets it-operations. One with a checkout/
directory gets revenue. One with an .env.example gets security. It is a
heuristic over filenames and it says so; selfcheck.py builds three synthetic
project trees and asserts the recommender names what those trees should
obviously produce, so you can check it before you trust it.

Both files are standard-library Python. Nothing talks to the network, and
neither one installs anything — they print commands and you decide which to
type.

Source: cbrock84/headcount, MIT.

Comment HIRE on the reel and I'll send the kit over.