
When you split work across several agents, you first have to sort out who sets direction and who takes on research and execution. Matt Pocock's mattpocock/skills repository has a chief-of-staff Skill that guides this kind of role split. It is an experimental set of instructions for coordinating subagents and schedules while pursuing a long-term goal and for separating strategic and tactical work.
A Skill that handles strategy and subagents that handle execution
The repository's Chief of Staff Skill describes keeping one long-term goal going while coordinating subagents and schedules. The Skill looks at goals and priorities from a strategic viewpoint, and the subagents take on concrete tactical work like research and implementation, so the roles are divided.
It also includes guidance to use context pointers, which point to where you need to look, such as research notes or earlier commits, instead of copying long work context every time. This is a working method for where to pick up information.
An in-progress beta that is still being tested
This Skill is in the repository's skills/in-progress category. The repository guide marks this category as beta and explains that its contents may change or disappear before they move into a stable category. So it is more accurate to treat the current behavior and wording as public guidance at an experimental stage.
Installing the Chief of Staff Skill
Following the repository's installation guide, you can specify the repository and the name of the Skill to install in the Skills CLI.
npx skills@latest add mattpocock/skills --skill chief-of-staffAfter installing, you can organize your long-term goal and sub-tasks according to the Skill's instructions and set it up to refer to the context you need, such as research notes and earlier commits.
If you want to explore agent skill repositories, you can read Marketing Skills and Building an AI Agent Team alongside this post. For an example of delegating roles, see also Grok Bot's Primary Bot. A different approach to long-running agents is covered in the OpenAI dots post.
