Skip to content

An agent organization structured as a company — 15+ departments, 125+ skills, each independently installable, citing the standards and regulators that settle the question. Runs in Claude Code and ChatGPT.

Bring the documentation up to what the repository actually is

Claude03357300ebfb
diff --git a/AGENTS.md b/AGENTS.md
index aff0ecf..e866618 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -27,7 +27,9 @@ Every check CI runs, in one script — the workflow calls this same file, so the
- **One owner per path.** `docs/AGENT-SURFACES.md` maps every tracked path to exactly one agent, and no two owners may claim the same path. A new department adds its roster row, its surface block and its charter in the same change.
- **Generated files are never hand-edited.** The README, the org chart, the social card, each skill's `references/sources.md`, and these manifests are all emitted from the tree and verified with `--check`. An edit to one is lost on the next regenerate.
-- **US English**, by exact word form. `scripts/check-us-english.py --fix` rewrites.
+- **Sources are references, never copies.** `sources/*.toml` maps outside authorities to the skills they settle questions for, and `scripts/build-sources.py` emits each skill's list into its own `references/sources.md`. The `license` field decides whether an agent may quote a source or only cite it — most of what a professional must cite is not open, so when torn between two classes take the more restrictive one.
+- **Verticals emit one way.** `verticals/<slug>/` plus the core emits a standalone repository via `scripts/build-vertical.py`. Generated output is never hand-edited and is not committed; `--verify` emits to a temporary directory and runs the emitted repository's own checks instead.
+- **US English**, by exact word form. `scripts/check-us-english.py --fix` rewrites. It leaves URLs alone, which it did not always do.
- **No third-party licensed content.** All prose here is original; `scripts/check-provenance.py` is the backstop.
## Departments
diff --git a/README.md b/README.md
index cf7bb37..a745216 100644
--- a/README.md
+++ b/README.md
@@ -3,9 +3,10 @@
<p align="center"><b>Add a department, not a prompt.</b></p>
<p align="center">
- <a href="https://claude.com/claude-code"><img alt="Built for Claude Code" src="https://img.shields.io/badge/built%20for-Claude%20Code-D97757?style=flat-square"></a>
+ <a href="AGENTS.md"><img alt="Runs in Claude Code and ChatGPT" src="https://img.shields.io/badge/runs%20in-Claude%20Code%20%C2%B7%20ChatGPT-D97757?style=flat-square"></a>
<img alt="16 departments" src="https://img.shields.io/badge/departments-16-3F4B5B?style=flat-square">
<img alt="172 skills" src="https://img.shields.io/badge/skills-172-3F4B5B?style=flat-square">
+ <a href="docs/SOURCES.md"><img alt="181 cited sources" src="https://img.shields.io/badge/cited%20sources-181-3F4B5B?style=flat-square"></a>
<a href="LICENSE"><img alt="MIT licensed" src="https://img.shields.io/badge/license-MIT-3F4B5B?style=flat-square"></a>
<a href="CONTRIBUTING.md"><img alt="PRs welcome" src="https://img.shields.io/badge/PRs-welcome-2EA043?style=flat-square"></a>
</p>
@@ -408,6 +409,8 @@ plugins/<department>/
.codex-plugin/plugin.json the same department, ChatGPT and Codex
skills/<skill>/SKILL.md frontmatter name equals the directory name
skills/<skill>/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/<name>/ industry packs, emitted as standalone repositories
.claude/agents/<id>.md one charter per department
diff --git a/docs/GETTING-STARTED.md b/docs/GETTING-STARTED.md
index 05b97d6..bc09b96 100644
--- a/docs/GETTING-STARTED.md
+++ b/docs/GETTING-STARTED.md
@@ -1,14 +1,20 @@
# Getting started
-This is a library of expertise, packaged as sixteen independently installable departments for
-[Claude Code](https://claude.com/claude-code). You install the ones you need, and the relevant
-specialist loads itself when you ask something in its territory.
+This is a library of expertise, packaged as sixteen independently installable departments. You
+install the ones you need, and the relevant specialist loads itself when you ask something in its
+territory.
-Nothing here changes your project. Installing a department adds skills to Claude Code — it does not
-write to your repository, add dependencies, or run anything on its own.
+It runs in [Claude Code](https://claude.com/claude-code) and in ChatGPT and Codex. The skills are
+the same files in both — only the manifests differ, and both sets are generated from the same tree,
+so a fix reaches both at once.
+
+Nothing here changes your project. Installing a department adds skills to the assistant — it does
+not write to your repository, add dependencies, or run anything on its own.
## Install
+### Claude Code
+
Add the marketplace once:
```
@@ -24,6 +30,16 @@ Then install departments one at a time:
`/plugin` on its own opens the plugin menu, where installed plugins can be reviewed and managed.
+### ChatGPT and Codex
+
+The same repository. Add it as a plugin marketplace — the manifest ChatGPT reads is at
+`.agents/plugins/marketplace.json` — or copy the department you want into `.agents/skills/` in your
+own project.
+
+One thing worth knowing on that side: skill descriptions share a context budget of roughly eight
+thousand characters, and all sixteen departments together come to about fifty thousand. Installing
+a few departments is not just tidier there, it is the only thing that works.
+
**Install what you will use, not everything.** Sixteen departments is a lot of surface, and a
smaller set produces sharper triggering. You can add more at any point.
@@ -51,13 +67,15 @@ to reach for first.
> *"Our margins slipped this quarter and nobody can tell me why."*
-**2. Name the skill** when you want a specific lens rather than the one that would trigger:
+**2. Name the skill** when you want a specific lens rather than the one that would trigger. In
+Claude Code:
```
/finance:cost-accounting
```
-Skills are addressed as `department:skill`, so names never collide across departments.
+Elsewhere, name it in the sentence — "use the cost accounting skill" — since the slash form is
+Claude Code's. Skills are addressed as `department:skill` throughout, so names never collide.
**3. Delegate a whole department.** Each department ships an agent charter in `.claude/agents/`,
so it can be handed a body of work as a subagent with its own exclusive write surface. See
diff --git a/docs/USE-CASES.md b/docs/USE-CASES.md
index 9934420..c0f699d 100644
--- a/docs/USE-CASES.md
+++ b/docs/USE-CASES.md
@@ -7,8 +7,12 @@ These are the situations this catalog is shaped around. Every skill named here e
in CI fails if a reference stops resolving, so this page cannot rot as skills are renamed or
consolidated.
-New here? [Getting started](GETTING-STARTED.md) covers installing, which departments to take first,
-and the three ways to invoke a skill.
+Where a situation turns on something an outside authority settles — an obligation, a threshold, a
+standard — the skill carries that authority in its own `references/sources.md` and cites it. See
+[the source catalog](SOURCES.md) for what is covered and what may be quoted from it.
+
+New here? [Getting started](GETTING-STARTED.md) covers installing in Claude Code or in ChatGPT,
+which departments to take first, and the three ways to invoke a skill.
## Single asks
diff --git a/docs/assets/social-card.html b/docs/assets/social-card.html
index ab37026..0130dd7 100644
--- a/docs/assets/social-card.html
+++ b/docs/assets/social-card.html
@@ -71,7 +71,8 @@
<div class="meta">
<b>16</b> departments <span class="sep">/</span>
<b>172</b> skills <span class="sep">/</span>
- MIT <span class="sep">/</span> for Claude Code
+ <b>181</b> cited sources <span class="sep">/</span>
+ MIT <span class="sep">/</span> Claude Code &amp; ChatGPT
</div>
<!-- The chart is the organization's actual shape: one chief executive, departments beneath, and
diff --git a/docs/assets/social-preview.png b/docs/assets/social-preview.png
index 94af429..1aac11e 100644
Binary files a/docs/assets/social-preview.png and b/docs/assets/social-preview.png differ
diff --git a/scripts/build-port.py b/scripts/build-port.py
index d9a564b..8263ffe 100755
--- a/scripts/build-port.py
+++ b/scripts/build-port.py
@@ -120,7 +120,17 @@ def agents_md(department_count, skill_count, titles_by_slug, order):
"- **Generated files are never hand-edited.** The README, the org chart, the social card, "
"each skill's `references/sources.md`, and these manifests are all emitted from the tree "
"and verified with `--check`. An edit to one is lost on the next regenerate.",
- "- **US English**, by exact word form. `scripts/check-us-english.py --fix` rewrites.",
+ "- **Sources are references, never copies.** `sources/*.toml` maps outside authorities to "
+ "the skills they settle questions for, and `scripts/build-sources.py` emits each skill's "
+ "list into its own `references/sources.md`. The `license` field decides whether an agent "
+ "may quote a source or only cite it — most of what a professional must cite is not open, "
+ "so when torn between two classes take the more restrictive one.",
+ "- **Verticals emit one way.** `verticals/<slug>/` plus the core emits a standalone "
+ "repository via `scripts/build-vertical.py`. Generated output is never hand-edited and is "
+ "not committed; `--verify` emits to a temporary directory and runs the emitted "
+ "repository's own checks instead.",
+ "- **US English**, by exact word form. `scripts/check-us-english.py --fix` rewrites. It "
+ "leaves URLs alone, which it did not always do.",
"- **No third-party licensed content.** All prose here is original; "
"`scripts/check-provenance.py` is the backstop.",
"",
diff --git a/scripts/build-readme.py b/scripts/build-readme.py
index 8aa55fd..a1adcb3 100755
--- a/scripts/build-readme.py
+++ b/scripts/build-readme.py
@@ -49,6 +49,22 @@ def load_departments():
ORDER = load_departments()
+def _catalog_counts():
+ """Catalog totals, read from the catalog so the README cannot claim a number it does not hold."""
+ import importlib.util
+ spec = importlib.util.spec_from_file_location("check_sources", "scripts/check-sources.py")
+ module = importlib.util.module_from_spec(spec)
+ spec.loader.exec_module(module)
+ entries = module.load()
+ skills = {ref for _, e in entries for ref in e.get("skills", [])}
+ quotable = {"public-domain-usgov", "public-domain", "cc0", "cc-by", "open-data",
+ "attribution-required"}
+ return len(entries), len(skills), sum(1 for _, e in entries if e["license"] in quotable)
+
+
+SOURCE_COUNT, SOURCE_SKILLS, SOURCE_QUOTABLE = _catalog_counts()
+
+
def summarize(path, limit=165):
text = open(path, encoding="utf-8").read()
front = re.match(r"^---\s*\n(.*?)\n---", text, re.S).group(1)
@@ -74,10 +90,14 @@ out = [
'<p align="center"><b>Add a department, not a prompt.</b></p>',
"",
'<p align="center">',
- f' <a href="https://claude.com/claude-code"><img alt="Built for Claude Code"'
- f' src="{B}/built%20for-Claude%20Code-D97757?style=flat-square"></a>',
+ # "Built for Claude Code" stopped being true when the ChatGPT manifests landed. Both tools read
+ # the same skills from this tree, and the badge is the first thing anyone reads.
+ f' <a href="AGENTS.md"><img alt="Runs in Claude Code and ChatGPT"'
+ f' src="{B}/runs%20in-Claude%20Code%20%C2%B7%20ChatGPT-D97757?style=flat-square"></a>',
f' <img alt="{len(ORDER)} departments" src="{B}/departments-{len(ORDER)}-3F4B5B?style=flat-square">',
f' <img alt="{total} skills" src="{B}/skills-{total}-3F4B5B?style=flat-square">',
+ f' <a href="docs/SOURCES.md"><img alt="{SOURCE_COUNT} cited sources"'
+ f' src="{B}/cited%20sources-{SOURCE_COUNT}-3F4B5B?style=flat-square"></a>',
f' <a href="LICENSE"><img alt="MIT licensed" src="{B}/license-MIT-3F4B5B?style=flat-square"></a>',
f' <a href="CONTRIBUTING.md"><img alt="PRs welcome" src="{B}/PRs-welcome-2EA043?style=flat-square"></a>',
"</p>",
@@ -161,21 +181,6 @@ for dept, title, exec_role in ORDER:
out.append(f"| `{os.path.basename(os.path.dirname(p))}` | {summarize(p)}. |")
out += ["", "</details>", ""]
-def catalog():
- """The source catalog, read the same way check-sources.py reads it."""
- import importlib.util
- spec = importlib.util.spec_from_file_location("check_sources", "scripts/check-sources.py")
- module = importlib.util.module_from_spec(spec)
- spec.loader.exec_module(module)
- entries = module.load()
- skills = {ref for _, e in entries for ref in e.get("skills", [])}
- quotable = {"public-domain-usgov", "public-domain", "cc0", "cc-by", "open-data",
- "attribution-required"}
- return len(entries), len(skills), sum(1 for _, e in entries if e["license"] in quotable)
-
-
-SOURCE_COUNT, SOURCE_SKILLS, SOURCE_QUOTABLE = catalog()
-
out += [
"**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",
@@ -210,6 +215,8 @@ out += [
" .codex-plugin/plugin.json the same department, ChatGPT and Codex",
" skills/<skill>/SKILL.md frontmatter name equals the directory name",
" skills/<skill>/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/<name>/ industry packs, emitted as standalone repositories",
".claude/agents/<id>.md one charter per department",
diff --git a/scripts/build-social-card.py b/scripts/build-social-card.py
index 7d205d8..2d1f722 100755
--- a/scripts/build-social-card.py
+++ b/scripts/build-social-card.py
@@ -25,6 +25,15 @@ WIDTH, HEIGHT = 1280, 640 # GitHub's recommended social preview size; rendered
# Same glob as build-readme.py. Both walk the tree directly rather than sharing state, so the
# card and the README cannot report different numbers for the same commit.
+def source_total():
+ """Distinct sources in the catalog, read from the catalog rather than typed."""
+ import importlib.util
+ spec = importlib.util.spec_from_file_location("check_sources", "scripts/check-sources.py")
+ module = importlib.util.module_from_spec(spec)
+ spec.loader.exec_module(module)
+ return len({entry["id"] for _, entry in module.load()})
+
+
def counts():
depts = sorted(
os.path.basename(os.path.dirname(os.path.dirname(m)))
@@ -107,7 +116,8 @@ TEMPLATE = """<meta charset="utf-8">
<div class="meta">
<b>{DEPTS}</b> departments <span class="sep">/</span>
<b>{SKILLS}</b> skills <span class="sep">/</span>
- MIT <span class="sep">/</span> for Claude Code
+ <b>{SOURCES}</b> cited sources <span class="sep">/</span>
+ MIT <span class="sep">/</span> Claude Code &amp; ChatGPT
</div>
<!-- The chart is the organization's actual shape: one chief executive, departments beneath, and
@@ -159,7 +169,7 @@ TEMPLATE = """<meta charset="utf-8">
def render_html():
depts, skills = counts()
- return TEMPLATE.format(W=WIDTH, H=HEIGHT, DEPTS=depts, SKILLS=skills)
+ return TEMPLATE.format(W=WIDTH, H=HEIGHT, DEPTS=depts, SKILLS=skills, SOURCES=source_total())
def find_chromium():