Skills in OpenAI's API are reusable bundles of scripts, instructions, and assets, organized in a specific folder structure. Each skill is defined by a required SKILL.md manifest, which includes metadata like name and description. When skills are attached to a shell environment, the system uploads and unzips these bundles, allowing models to access and execute the contained scripts. This setup is ideal for repeatable workflows that require branching logic, version control, and the ability to run code. Skills are especially useful for tasks that benefit from being shared across multiple agents or prompts. For instance, a "company-specific report generator" or a "data-cleaning pipeline" can be encapsulated in a skill. They shine in scenarios where procedures are stable and well-defined, as opposed to tasks that change frequently or require live data. Skills should be used instead of system prompts for complex procedures to keep those prompts slim, reserving them for global behaviors and constraints. Creating a skill involves structuring your folder correctly and making an API call to upload the skill. OpenAI provides two methods for uploading: either as individual files or as a zipped bundle. Once uploaded, skills can be referenced in execution environments, either hosted or local. The article includes a practical example, detailing a "CSV Insights Skill" that summarizes a CSV file, computes statistics, and generates a report with visualizations. This showcases how skills can integrate scripts and assets to automate complex tasks efficiently.