Department skill packs.

§ 01

What is a skill, in plain English?

A skill is a reusable workflow. Anthropic describes Claude Skills as folders of instructions, scripts, and resources that Claude loads when they are relevant. OpenAI describes ChatGPT skills as reusable, shareable workflows that can include instructions, examples, and code. In ChatGPT, a plugin can include one or more skills, and can also include connected apps.

The important part for a department is consistency. When someone asks for the weekly client report, the skill should already know the outline, the tone, what to include, and which numbers need a source. The assistant should not invent a new format each time.

Claude Skills can be used on Free, Pro, Max, Ultra, Team, and Enterprise plans, with code execution turned on. ChatGPT skills are available to eligible Business, Enterprise, Healthcare, and Edu users, subject to workspace settings.

§ 02

Why use a skill instead of a long prompt or a Project?

A prompt in a single chat disappears the next time someone starts from scratch, or gets edited until nobody is sure which version is current. A Project, in ChatGPT or in Claude, is useful for background knowledge and ongoing work in one workspace. Anthropic’s comparison is direct: projects provide static background knowledge for chats in that project, while skills are specialized procedures that activate when needed and work more broadly. OpenAI similarly notes that a project can be a good fit for ongoing writing, research, or planning, but it does not replace a shareable skill or plugin.

Our honest take

Use a Project for “here is the context for this body of work.” Use a skill for “here is how we always do this job.” If more than one person has to produce the same report, quote, or recap, it belongs in a skill.

§ 03

What does a department pack usually include?

The pack is named after the department’s recurring work, not after the vendor. These are examples of how to structure a pack, not case studies of named clients.

Our honest take

Start with the one or two jobs that already happen every week and have a known good example. Add more only after those run reliably. Anthropic notes that a skill’s description is what helps Claude decide when to use it automatically, so vague names are a common reason a skill never kicks in.

§ 04

Do skill packs need connectors too?

Sometimes. A skill can work from files you attach, from Company Knowledge or a Project, or from text someone pastes. That is enough for some writing and review work.

If the job depends on live records, such as invoices, pipeline stages, or last week’s tickets, the skill needs a way to read that system. In Claude, skills and MCP connectors are meant to be used together: the connector gives access, and the skill teaches the procedure. In ChatGPT, the same idea shows up as a plugin that includes both a skill and an app, including a custom MCP server when no official app covers the system.

If a person still has to download a CSV every Monday for the report to work, the skill is only half done. Either attach a connector, or be honest that the pack is a drafting aid and the export step is still manual. See how to connect QuickBooks, HubSpot or your CRM. A reporting skill usually needs read access only.

§ 05

How do you share a skill pack inside a company?

Keep the pack private to your workspace or to the group that needs it, unless you have a separate reason to publish something publicly. In ChatGPT, you can share a skill with people or groups, or publish it to the workspace library. Enterprise and Edu admins control who can create, upload, share, publish, and install skills. After a custom GPT is migrated, the replacement plugin starts private and has to be shared separately.

In Claude, Team and Enterprise owners can provision a skill for everyone, or, on Enterprise, put skills into a plugin and assign that plugin to a group. Anthropic’s own example is a set of marketing skills that only the marketing group should see. Test the pack with one real user and one real job before you provision it for a whole department. Review a skill before sharing it, especially if it includes code or was uploaded from a file.

§ 06

What does “done” look like?

• A person on that team can produce the recurring deliverable in the team’s real format, without rewriting the instructions.
• A second person can do it with the shared skill, not a private copy of a prompt.
• Numbers in the output can be checked against the source system or the files provided.
• Anything that would be sent or changed still waits for approval.
• Someone owns updates. A stale skill is worse than no skill, because people will trust the old outline.

Our honest take

Save a few good examples and at least one awkward case, such as missing data or a period with no activity, and run those after every change.

§ 07

When should you build this yourselves, and when should you hire?

If the job is a writing template, the source material is already in front of the person, and one operator is willing to own the skill, your team can often create it in the product. Both ChatGPT and Claude let eligible users create a skill from instructions, and Claude notes that simple skills can be written in Markdown without code.

Get help when the pack has to be shared cleanly across a department, when it must read or write a company system, when several skills have to work together, or when you already have half-finished prompts, GPTs, or Projects and need them turned into something reliable. Appearly’s department skill packs are a fixed fee per skill, quoted after a 15-minute call. If a connector is required, that is a separate fixed-fee build.

§ FAQ

Questions we hear