An agent organization for Claude Code, structured as a company — 15+ departments, 125+ skills, each independently installable.
Overview
Open the interactive org chart — search every skill, open a department, jump to the source. An agent organization for Claude Code, structured as a company: a chief executive over 16 departments, 172 skills in total. Every department is an independently installable plugin, so a project loads only the functions it needs rather than all of them at once. Install as many departments as the project needs. Skills are addressed as department:skill — security:threat-modeling, finance:unit-economics — so names never collide. Skills load themselves when a request matches. Ask a question in the department's territory and the right specialist engages: Invoke one directly by name when you want a specific lens: /finance:financial-modeling. New to this? docs/GETTING-STARTED.md covers which departments to install first, the three ways to invoke a skill, and what reviewer-class departments do differently.
README
headcount
Add a department, not a prompt.
Open the interactive org chart — search every skill, open a department, jump to the source.
An agent organization structured as a company: a chief executive over 16 departments, 172 skills in total.
Every department is an independently installable plugin, so a project loads only the functions it needs rather than all of them at once.
Install
/plugin marketplace add cbrock84/headcount
/plugin install security@headcount
ChatGPT and Codex — the same repository. Add it as a plugin marketplace, or drop the
department you want into .agents/skills/ in your own project.
The skills are identical in both; only the manifests differ, and both sets are generated from
this tree, so a fix reaches both at once. See AGENTS.md.
Install as many departments as the project needs. Skills are addressed as department:skill —
security:threat-modeling, finance:unit-economics — so names never collide.
Use
Skills load themselves when a request matches. Ask a question in the department’s territory and the right specialist engages:
| You ask | What loads |
|---|---|
| “why isn’t this landing page converting?” | demand-generation:landing-page-cro-expert |
| “review this design before we build it” | security:threat-modeling |
| “can we afford this hire?” | finance:unit-economics |
| “our growth has stalled” | executive:business-growth-consultant |
Invoke one directly by name when you want a specific lens: /finance:financial-modeling.
New to this? docs/GETTING-STARTED.md covers which departments to install first, the three ways to invoke a skill, and what reviewer-class departments do differently.
Eleven situations that cross departments — a SOC 2 demand from an enterprise prospect, a link down between two sites, a renewal that auto-renewed because nobody owned the date — are worked through end to end in docs/USE-CASES.md, including what comes back and where a reviewer-class department stops the work rather than adding an opinion.
Each department also ships an agent charter in .claude/agents/, so a department can be delegated
to as a subagent with its own exclusive write surface.
Departments
Reviewer-class departments (security, legal-risk) review what other departments build, and
their blocking findings are not overrulable by the department under review. That is why the CISO
and the CLO report to the chief executive rather than into the function they oversee.
Sources
A skill states what a competent practitioner knows. It cannot state what the regulator
published last month — it is written once and the obligation moves. So skills that answer
questions an outside authority settles carry a list of those authorities, in
references/sources.md inside the skill, which is where an agent reads it while answering.
184 sources across 150 skills so far — tax and accounting, law and
employment, security and controls frameworks, education standards.
The full index is in docs/SOURCES.md.
References, never copies, and every entry carries what you may actually do with it. That second part is the point: most of what a professional must cite is not open. ISO standards are sold, SANS papers are copyrighted, the FASB Codification needs an account — while US federal works are public domain by statute and EU legal texts are reusable with attribution. 125 of the 184 are quotable; the rest are read-and-cite, and the entry says so in the imperative next to the link.
Links are re-checked weekly by their own workflow rather than on every push, because a publisher being briefly down is not a reason to fail an unrelated pull request.
How it is organized
plugins//
.claude-plugin/plugin.json department manifest, Claude Code
.codex-plugin/plugin.json the same department, ChatGPT and Codex
skills//SKILL.md frontmatter name equals the directory name
skills//references/ supporting files, including the skill's sources
.claude-plugin/marketplace.json the marketplace Claude Code reads
.agents/plugins/marketplace.json the same departments, for ChatGPT and Codex
sources/*.toml the source catalog, mapped to the skills it serves
verticals// industry packs, emitted as standalone repositories
.claude/agents/.md one charter per department
AGENTS.md repository context for any agent working on this repo
docs/AGENT-SURFACES.md every path has exactly one owner, enforced in CI
docs/DECISION-LOG.md numbered decisions with options and recommendations
docs/GETTING-STARTED.md install, what to take first, and how to invoke a skill
docs/SOURCES.md every source in the catalog, and what may be done with it
docs/USE-CASES.md situations worked end to end across departments
docs/org-chart.html interactive org chart, searchable across every skill
docs/index.html GitHub Pages entry point, redirects to the chart
Agents split by exclusive write surface, not by topic — a topic split has no checkable
boundary, and two agents working on “SEO” and “UI” both end up in the same file. See
executive:agent-hierarchy for the method.
Contributing
./scripts/check-all.sh
Every check CI runs, in one script. The surface map is coherent; every skill’s frontmatter is
valid and unique; no third-party license text has appeared; the README, social card and org
chart are current; every department:skill reference resolves; spelling is US English; no
## Never block mixes two styles; the source catalog is valid and every skill’s source file
matches it; the ChatGPT manifests match the Claude ones; every vertical emits a repository that
passes its own checks; and every manifest parses. CI calls this same script, so local and CI
cannot drift.
A new department needs its roster row in docs/AGENT-SURFACES.md, a surface block, a charter in
.claude/agents/, and an entry in .claude-plugin/marketplace.json — all in the same change, or
the check fails.
Contributors
The ChatGPT and Codex support in this repository started as a contribution from
@adi-dibra, who worked out that the same SKILL.md files load
in both tools and that only the manifests differ.
Writing
Notes from building and running this, and from the day job — technology, security, AI, and the operating side of all three — go out at cbrock84.substack.com.
The piece on why this is shaped like an org chart at all, and what broke before it was: Giving AI agents an org chart.
License
MIT — see LICENSE. Every skill here was written for this repository.
Built by Chris Brock.
README generated by scripts/build-readme.py — edit that, not this file.
Recommended Tools
Try a different keyword or remove a filter.
Install
npx skillfish add cbrock84/headcount