OpenAI launched Skills in Codex this week to let developers customize the Codex coding agent with task-specific capabilities using pre-built packages or custom scripts generated through natural language prompts.
The Skills in Codex service combines instructions, resources, and optional scripts, allowing Codex to execute specific workflows with reliability. OpenAI detailed this functionality in its developer documentation. Developers gain flexibility by crafting their own skills either through natural language descriptions or by writing scripts manually. Pre-built skills appear through GitHub repositories, providing ready-to-use options for common tasks.
This launch arrives shortly after OpenAI released GPT-5.2-Codex last week. OpenAI described GPT-5.2-Codex as its most advanced agentic coding model yet, setting the stage for enhanced customization features like Skills.
OpenAI’s Skills in Codex draws directly from the Agent Skills standard, originally created by Anthropic and released as an open specification on December 18. The specification resides at agentskills.io, complete with a reference SDK. Multiple platforms have integrated this standard, including Microsoft, GitHub, Cursor, Goose, Amp, OpenCode, and additional coding agents.
Mahesh Murag, a product manager at Anthropic, explained the initiative to VentureBeat. He stated, “We’re launching Agent Skills as an independent open standard with a specification and reference SDK available at https://agentskills.io.” Murag continued, “Microsoft has already adopted Agent Skills within VS Code and GitHub; so have popular coding agents like Cursor, Goose, Amp, OpenCode, and more.” This open approach facilitates widespread compatibility across developer tools.
Anthropic first rolled out Agent Skills in October 2023 as a feature within its Claude AI assistant. Since then, Anthropic established partnerships with Atlassian, Figma, Canva, Stripe, Notion, and Zapier. These collaborations deliver pre-built skills tailored to workflows in those platforms, expanding practical applications for AI-assisted development.
The development reflects broader industry efforts among major AI providers to advance coding agents. Amazon Web Services introduced powers for its Kiro coding agent during the re:Invent conference earlier this month. Powers function through Model Context Protocol servers and steering files, serving as expertise modules that equip the Kiro agent with immediate knowledge of particular technologies and frameworks. AWS characterized powers as these expertise modules that give the Kiro agent instant knowledge of specific technologies and frameworks.
Skills in Codex employs progressive disclosure for context management. This method loads only essential information during startup, then accesses full details as required. OpenAI’s developer documentation specifies installation options: skills install globally in the ~/.codex/skills/ directory or remain scoped to individual repositories, accommodating varied project needs.
