ChatGPT testing new "Skills" feature to enhance complex task handling

ChatGPT is trialing a new "Skills" feature designed to significantly improve its ability to manage complex tasks, as reported by BleepingComputer. According to a leak by user Tibor on platform X, the internal codename for the project is "hazelnuts." This move aligns with competitor Claude's approach and marks a shift away from ChatGPT's current reliance on custom GPTs, which essentially package specific prompt instructions.

The new system is expected to introduce a more structured method for users to teach the AI workflows and domain knowledge—similar to a "folder-based instruction set"—rather than just tweaking prompts. This aims to make the AI more adaptable and efficient.

Four core pillars define the leaked feature: composability (stacking skills as needed), portability, efficiency, and powerful functionality. For example, when writing front-end code, ChatGPT could automatically apply design-related skills to better understand layout requirements.

Notably, these “skills” may extend beyond text generation to include executable code, enhancing reliability in tasks requiring precise calculations or logic verification. The feature will also be deeply integrated into the chat interface, allowing users to quickly activate specific skills using slash commands (e.g., `/design`).

To ease the transition, OpenAI plans to release a dedicated Skill Editor and offer tools to convert existing custom GPTs into skills with one click, ensuring a smooth migration for current users.

This upgrade promises to elevate ChatGPT’s performance in complex scenarios, delivering a smarter and more capable user experience.