Productivity: Switchable Skill Groups

Ideas Feature Requests
,
1

Feature request for product/service

Cursor IDE

Describe the request

Feature Request: Switchable Skill Groups

Background
While developing with LLMs, I’ve tried several Skill collections (e.g. SuperPower, MattPocock Skills). I currently rely on MattPocock Skills day to day. I’ve also started experimenting with Poteto’s Poteto Stack.

Experience
Different skill combinations fit different scenarios:

Scenario Better fit
Fine-grained control, strong personal preferences: MattPocock Skills
Let the agent run with less hand-holding: SuperPower / Poteto Stack

A useful analogy: self-drive vs. guided tour — one gives you control; the other is more hands-off.

Problem
Loading all Skills into Cursor at once causes context interference:

Overlapping Skills across collections get cross-referenced by the model
Each collection stops performing at its best
Behavior becomes less predictable

Request
I’d like Skill Groups — similar to VSCode configuration profiles — so I can switch skill sets by context, for example:

How much of the codebase I fully control
How fine-grained I want control over project details
Solo project vs. collaborating with others

Proposed layout (reference)
Organize groups as folders:

.cursor/
├── skills/ # always-on: knowledge, conventions, etc.
└── skill-groups/
├── group-a/ # e.g. personal projects
└── group-b/ # e.g. team / collaborative projects

When a group is active, only always-on + that group’s Skills are loaded — unrelated skills stay out of context.

Expected benefits
Less context noise — clearer, more consistent instructions per session
Mode switching by scenario — precision vs. autonomy
More predictable behavior — each group can work as designed

Operating System (if it applies)

Windows 10/11
MacOS
Linux

8

Hey @sinar_pandora, thanks for the detailed suggestion!

There isn’t a skill-groups switcher today, but you can get partway there with custom modes: any skill can be entered as a sticky mode (it stays active and the agent keeps following it until you exit), so you could make a small entry-point skill per collection and switch between them per session.

To keep the other collections from adding context noise, you can also set disable-model-invocation: true in a skill’s frontmatter so it only loads when explicitly invoked.

It’s not the full grouping system you described, but it covers the mode-switching part, and the underlying context-noise concern is something the team is aware of!