# headcount: A 16-Department Agent Organisation for Claude Code and ChatGPT

> headcount is an MIT-licensed plugin repository that organises Claude Code into a company-shaped hierarchy of 172 skills across 16 departments, each independently installable. Skills load automatically when a request falls within their domain, and each department ships an agent charter for subagent delegation.

**cbrock84/headcount** — An agent organization for Claude Code, structured as a company — 15+ departments, 125+ skills, each independently installable.

- Repository: https://github.com/cbrock84/headcount
- Website: https://cbrock84.github.io/headcount/
- Stars: 1,773 · Forks: 267
- Language: Markdown
- License: MIT
- Published: 2026-09-10 · Updated: 2026-09-10 · Language: en
- Canonical page: https://hysenlabs.com/projects/cbrock84-headcount

## The Problem headcount Addresses in AI-Assisted Development

A general-purpose AI agent handles a wide range of requests adequately but handles specialised domain questions shallowly. A startup founder asking about unit economics, a developer asking about threat modelling, or a designer asking about landing page conversion rates all get generic responses unless the agent has been given targeted instructions for that domain.

headcount addresses this by structuring specialised skills as a plugin library organised like a company. The README describes it as "an agent organization structured as a company: a chief executive over 16 departments, 172 skills in total." Each department contains skills for a specific functional area. A project installs only the departments it needs, keeping the loaded skill set focused.

## How Skills Load and the Department-Skill Model

Skills in headcount load automatically when a request matches their territory. The README gives concrete examples: a question about why a landing page is not converting loads `demand-generation:landing-page-cro-expert`, a request to review a design before building it loads `security:threat-modeling`, and a question about whether a hire is affordable loads `finance:unit-economics`.

Skills are addressed in the format `department:skill`, so names never collide across departments. You can also invoke a skill directly by name when you want a specific lens: `/finance:financial-modeling` calls that skill explicitly rather than waiting for auto-detection.

The README distinguishes between regular skills and reviewer-class departments. Reviewer-class departments stop work and flag an issue rather than simply contributing an opinion. This distinction matters for departments like security, where the goal is to block a dangerous decision rather than add a comment.

## Installing headcount and Invoking Skills

headcount installs through the Claude Code plugin marketplace. The README shows a two-step process: add the repository as a marketplace source, then install the specific department you want:

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

Install as many departments as the project needs. Each department is independently installable, so a project that only needs security and finance capabilities can skip the remaining 14 departments. ChatGPT and Codex users can also use headcount by dropping the department directory into `.agents/skills/` in their own project; the README notes the skills are identical across Claude and ChatGPT, since both sets of manifests are generated from the same source tree.

For new users, the README points to `docs/GETTING-STARTED.md`, which covers which departments to install first and the three ways to invoke a skill. The `docs/USE-CASES.md` file walks through 11 cross-department situations in full, including scenarios like a SOC 2 audit demand from an enterprise prospect and a SaaS link outage.

## Departments Included in headcount

The Office of the CEO contains 7 skills covering direction-setting, fundraising, SaaS idea validation, and AI research analysis. The Technology department covers 19 skills including code review, cloud infrastructure, observability, prompt optimisation, release and deployment, and skill authoring.

The README lists the following as additional departments beyond these two: the 16-department total includes functional areas across engineering, finance, security, and demand generation, among others, though the full department list is visible in the `docs/` directory and the AGENTS.md file in the repository rather than being fully enumerated in the README summary.

Each department also has a subdirectory under `.claude/agents/` containing an agent charter. The charter defines the exclusive write surface the department controls when running as a subagent.

## Agent Charters and Subagent Delegation

Each department in headcount ships an agent charter in `.claude/agents/`. This means a department can be delegated to as a standalone subagent with its own scope of action, not just loaded as a skill that adds instructions to a single session.

The README describes this as giving each department an "exclusive write surface": when a department runs as a subagent, it only operates within its designated area rather than having general access. The `agent-hierarchy` skill in the Office of the CEO section is specifically designed for splitting work across subagents: the README describes it as designing "orchestrator-and-subagent hierarchies for a repository, splitting agents by exclusive write surface, pairing every producer with an independent auditor."

This architecture is a practical way to run parallel specialised agents on a complex task without subagents overwriting each other's work.

## Limitations and What headcount Does Not Do

headcount is a Markdown-based skill library. Skills are instruction sets, not executable code. They guide agent behaviour during a session but do not persist state, run background jobs, or enforce outcomes independently of the agent. A skill that advises on threat modelling cannot prevent a developer from writing insecure code after the session ends.

The auto-loading mechanism depends on request phrasing. Skills load when the agent infers a request matches their domain, which means unusual phrasing or ambiguous requests may not trigger the right skill. Direct invocation by name is more reliable when you know which skill you want.

headcount does not include domain-specific tools, APIs, or integrations. It adds instructions and domain expertise as prompts; it does not add new tools to the agent's toolset. A finance skill can advise on unit economics but cannot run spreadsheet calculations independently.

The last push to the main branch was on 2026-09-17 and the project has no GitHub releases.

## Conclusion

headcount is worth installing for Claude Code or ChatGPT users who run into the same domain-specific gaps repeatedly and want specialised instruction without writing their own prompts. Installing a single department is low-risk; the plugin format lets you add only what a project needs. Before relying on a reviewer-class department to stop risky work, read its charter in `.claude/agents/` to understand exactly what triggers it, since auto-loading depends on request phrasing.

## FAQ

### How do I install headcount for Claude Code?

Run `/plugin marketplace add cbrock84/headcount` to add the repository as a plugin source, then `/plugin install <department>@headcount` to install the department you want. Each department installs independently, so you can add only the ones a project needs.

### What departments are available in headcount?

headcount includes 16 departments with 172 skills total. The README describes the Office of the CEO (7 skills covering strategy, fundraising, and SaaS validation) and Technology (19 skills covering code review, cloud infrastructure, prompt optimisation, and deployment). The full list is in AGENTS.md in the repository.

### Does headcount work with ChatGPT as well as Claude Code?

Yes. The README states the same repository works with ChatGPT and Codex. Add it as a plugin marketplace or place the department directory into `.agents/skills/` in your own project. The skill manifests differ between Claude and ChatGPT but are generated from the same source tree.

## Sources

- [cbrock84/headcount on GitHub](https://github.com/cbrock84/headcount)
- [Issues](https://github.com/cbrock84/headcount/issues)
- [License: MIT](https://github.com/cbrock84/headcount/blob/main/LICENSE)
- [Project website](https://cbrock84.github.io/headcount/)
- [README](https://github.com/cbrock84/headcount/blob/main/README.md)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/cbrock84-headcount