AI NewsOpen sourceAnnouncement

headcount packages 146 coding agent skills into 16 department plugins

A Claude Code project called headcount puts 146 skills into 16 department plugins, each independently installable, and gained 846 stars in the three days since its public creation.

AI News

Editorial2 min read

LinkedInX
The headcount GitHub repository social card

Image: GitHub

Why it mattersSkills are addressed as department:skill and each plugin includes an agent charter, so a project loads only the parts it needs rather than one large prompt library.

A repository called headcount, created on 28 August by GitHub user cbrock84, packages an agent organisation for Claude Code as a set of 16 department plugins holding 146 skills between them. The repository is at 846 stars three days after it went public, and the README's badges put the current department count at 16 and the skill count at 146.

Each department is a separately installable Claude Code plugin, with its own manifest and its own agent charter. A project installs only the departments it needs.

What it contains

The tagline on the project's main page is "Add a department, not a prompt." The structure follows that literally. A user installs the marketplace once, then adds departments one at a time:

  • Office of the CEO holds strategic skills like chief-executive, saas-idea-validator and business-growth-consultant.
  • Security includes threat-modeling and related reviewer-class skills.
  • Finance includes unit-economics and financial-modeling.
  • Engineering, marketing, product, demand generation, executive coaching and a dozen other departments follow the same structure.

Skills are addressed as department:skill, for example security:threat-modeling or finance:unit-economics, so names cannot collide across departments. The README says skills load themselves when a request matches, and a specific skill can also be invoked directly by name.

A design decision worth copying

The project makes one architectural choice that is worth noting even for teams that never install it. Agents in headcount are split by exclusive write surface, meaning a set of files that only one agent may change, not by topic. The README states the reason directly: a topic split has no checkable boundary, and two agents working on "SEO" and "UI" both end up in the same file. Each department gets its own agent charter under .claude/agents/ and can be delegated to as a sub-agent with a set of files no other department changes. A CI check enforces the ownership map.

The repository also includes an interactive org chart at cbrock84.github.io/headcount/org-chart.html that indexes every skill, along with a decision log and a use-cases document that describes seven cross-department situations end to end. One of those situations, per the README, is where a reviewer department stops the work rather than adding an opinion.

For teams already running Claude Code with a growing number of ad-hoc skills, the interesting artefact is the boundary system rather than the skill count. The two rules the project builds in, exclusive write surfaces and one department per plugin, address the kind of problem where two skills overwrite each other's outputs without anyone noticing. The org chart is also a useful check on how big a skill library can grow before its skills start to overlap.

The interest is real: 846 stars and 129 forks in the first three days after the repo went public. Star counts alone say nothing about correctness. They do say other Claude Code users are looking for the same structure. Anyone maintaining a Claude Code skill library at scale should read the ownership rules and the decision log before dividing their skills in a new way.

Source

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

Share
LinkedInX
Start a project