Skip to content
Back to skills

Departments

ASecurity

Add, change or remove a department of a ceo-of-agents hq, or grant one a new tool or connector, through a decision card. Holds the grant review (a one-line reason for every elevated grant), the account checks and the files to write once the CEO decides, including moving a removed department's folder to the archive. Use when the CEO asks for a new department, a change to one, or access for one, and when recurring work suggests a new department.

  • 2 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added November 13, 2026
ai-agentsgobashgitsecurity

Works with

  • mcp

Security analysis

A100/100

Scanned November 18, 2026

npx -y skills add iamgiolaga/ceo-of-agents --skill departments --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Departments?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Departments
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/iamgiolaga-departments/badge)](https://www.skillsdirectory.com/skills/iamgiolaga-departments)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: departments
description: Add, change or remove a department of a ceo-of-agents hq, or grant one a new tool or connector, through a decision card. Holds the grant review (a one-line reason for every elevated grant), the account checks and the files to write once the CEO decides, including moving a removed department's folder to the archive. Use when the CEO asks for a new department, a change to one, or access for one, and when recurring work suggests a new department.
user-invocable: false
allowed-tools: Read Grep Glob Edit Write Bash(git status:*) Bash(git ls-files:*) Bash(git mv:*) Bash(mkdir:*) Bash(ls:*) Bash(date:*)
---

# departments: add, change or remove a department, or grant access

Reply to the CEO in the chat language set in `org.yaml` (`language.chat`); write every file in the file language (`language.files`).

Closed org: when `org.status` in `org.yaml` is `closed`, say that the org is closed and read-only, and stop: this skill writes nothing. Only the Reopen procedure of the `close-org` skill writes in a closed org.

Only the chief of staff (CoS) runs this skill. A department never changes its own charter, agent file or tools.

Every file this skill writes in the hq is written by a `worker` (one tier down, explicit model) from your instructions, and you read the diff; commits, pushes and `gh` writes stay with you.

## Prepare the card

1. Frame the change in one line: a new department, a change to one (scope, exclusions, tools, model, name), a removal, or a new tool or connector for one.
2. A new department can be built in four ways, the same as at setup. Ask in plain words which one the CEO wants, unless the request already says it; the first is the plain path and needs no further question:
   - **the catalog entry as is**: `${CLAUDE_PLUGIN_ROOT}/catalog/<id>/charter.md` and `agent.md`;
   - **the catalog entry plus changes**: the same files, plus a "Changes for this org" section (`${CLAUDE_PLUGIN_ROOT}/templates/department/changes.md`) at the end of both, and the `tools:` line with the abilities added or dropped. Everything else stays from the catalog. The rules are those of question 5 in the `init-fleet` skill ("Changing a catalog department"): a one-line reason per ability added (only web search can be added or dropped; `Bash` and `Agent(reader, worker)` are fixed for non-advisory departments), and an advisory entry never gains commands or helpers. `adapted: true` in `org.yaml`. The same applies to changing an existing catalog department later;
   - **fully custom**: `${CLAUDE_PLUGIN_ROOT}/templates/department/` with its scope, triggers, exclusions, single core job. It gets the standard non-advisory tools (`Read, Grep, Glob, Bash, Agent(reader, worker)`, no `Edit` or `Write`: its file writes go through a `worker`), so there is no question about commands;
   - **the CEO's own agent or skill**: registered as it is, never rewritten, copied or moved: a charter from `${CLAUDE_PLUGIN_ROOT}/templates/department/own-charter.md`, an item with `from_catalog: own`, `kind`, `location` and no `model`, and a routing row. It can sit next to the others, replace one (the replaced department is removed in the same decision), or be the only one. If the CEO gives its file path, read it once, read-only, and show its tools; it brings its own tools, and this org grants it nothing.

   Ask the CEO for what is missing; never invent it. Then draft the display name and agent name (the persona in lowercase, or the id; a brought-in one keeps its own name; unique, never `reader`, `worker` or another department's id), the model tier (not for a brought-in one) and, if wanted, "top model, workers included" (`workers: same`). For a change to an existing department: the exact lines that change in its charter, its agent file and `org.yaml`.
   For a removal: who takes over its work, if anyone, and one plain line that its folder will move to the archive, for example "Its notes and analyses move to `archive/departments/support/`: nothing is deleted, and they can come back." The folder is deleted instead only if the CEO asks for that explicitly (see "After the CEO decides").
3. Overlap check: compare what the new or changed department covers with every other department (charter scopes, changes sections, the core job and triggers of brought-in ones). For each piece of work two of them cover, add one plain line with a suggestion to the card, for example "Engineering and Security both cover security reviews: drop it from Engineering, or keep both and route security work to Security." Warnings only: they never block the card.
4. Review the draft as a whole (scope, tools, connectors, model) and give a one-line reason for each elevated grant: any connector, `Edit`, `Write`, `Bash` or `Agent`. `Bash` (limited to git, gh, tests or builds and spot checks) and `Agent(reader, worker)` are the standard tools of every non-advisory department, because it delegates bulk reading and every file write to the workers; `Edit` and `Write` are only for advisory departments, which have no `Agent` tool and write their own files. Drop a grant that has no reason, and say so.
5. Account checks for a connector:
   - never one listed in `accounts.forbidden_connectors` in `org.yaml`;
   - only a connector of this org's own accounts (see "Account separation" in `CLAUDE.md`);
   - an advisory department gets only read-only connector tools, named one by one, never a whole server (`mcp__<server>__*`).
6. Write the decision card (`company/decision-queue.md`, own D code) with the draft, the reasons and a recommendation, and add it to "Open decisions" and to "Waiting on the CEO" as `- <ORG_ID>-D<n> [urgency] {decide}: decide <question>, see decision queue (since YYYY-MM-DD)` (`${CLAUDE_PLUGIN_ROOT}/schemas/state.md`).

A reply from the CEO to a denied permission is not a grant by itself: do not widen an allowlist on the strength of it; the grant goes through the card.

## After the CEO decides

Record the decision with the `record-decision` skill, then apply exactly what was decided:

- New department: `departments/<id>/charter.md` and `.claude/agents/<agent>.md` rendered from the catalog entry or the custom template, or only the charter for the CEO's own agent or skill, and in every case `departments/<id>/README.md`, the index of its documents, from `${CLAUDE_PLUGIN_ROOT}/templates/department/index.md` (placeholders as in the `init-fleet` skill's placeholder reference), an item in `departments:` in `org.yaml` (`connectors: []` unless the decision grants one), and a row in the routing table of `CLAUDE.md`. Writes into `.claude/` may raise a permission prompt.
- Grant: add the connector to the department's `connectors:` in `org.yaml` and to the `tools:` list of its agent file. For the CEO's own agent or skill, record it in `org.yaml` only and tell the CEO that its own file decides what it can use.
- `workers: same` gained or lost: also update the worker model lines in its agent file.
- Removal: remove its agent file (`.claude/agents/<agent>.md`), its item in `org.yaml` and its routing row, and archive its folder: nothing in it is deleted. For the CEO's own agent or skill, its own file stays where it is.
  - Archive `departments/<id>/` as one folder with steps 1 to 4 of `${CLAUDE_PLUGIN_ROOT}/schemas/document.md`, section "Archiving": it becomes `archive/departments/<id>/`, each document in it gets `status: archived`, and links in current files point to the new path. The removal decision is the CEO's OK for this move: do not ask again. Documents it wrote into an initiative's folder stay there.
  - Delete the folder instead only when the CEO asks for deletion explicitly. That is a `delete-data` action: ask for the CEO's OK on that one folder, by path, as a separate request, then `git rm -r departments/<id>/`. Git still keeps the history.

Then commit with notice, as for any hq change.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

View all in ai-agents →

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.