xAI Releases Grok Skills and Updates Tool Calling Responses API

XAI has released Grok Skills, a feature that allows users to create persistent custom expertise through natural language or file uploads, which the Grok 4.3 model then automatically applies across all conversations on web, iOS, and Android platforms. The update also enhances the Responses API with OpenAI-compatible tool calling, supporting built-in tools like web search and code execution, as well as custom functions, with a 1 million token context window and parallel tool calls for multi-step tasks. While similar to offerings from OpenAI, Anthropic, and Vercel, Grok Skills differentiates itself by functioning as a reusable workflow layer tightly integrated with the X platform, combining instructions, search, and social context into lightweight, platform-native workflows.

xAI has released Grok Skills together with enhancements to the Responses API for Grok 4.3, enabling persistent custom expertise that the model retains across all conversations on the web platform, iOS app, and Android app. Users create these skills once through natural language descriptions or file uploads, after which Grok applies the defined workflows, preferences, and document-handling routines automatically without needing repeated instructions in future sessions. The built-in capabilities cover full generation and editing of Word documents that preserve headings, tables, and styles, creation of PowerPoint-style slide decks that include visual hierarchy and speaker notes, Excel spreadsheets that support formulas, data analysis, charts, and conditional formatting, and PDF operations that allow creation, merging, splitting, text extraction, and content reorganization. These skills operate at the account level, take priority over default behaviors when invoked via slash commands, and support sharing between users for collaborative setups. On the developer side, the Responses API integrates these concepts through tool calling that follows an OpenAI-compatible format while adding native server-side execution for built-in tools. Developers include tools in API requests by specifying types such as web search, x search, or code interpreter for automatic handling on xAI infrastructure, or define custom functions using JSON schemas that describe name, description, and parameters. When Grok 4.3 determines a tool is needed, it returns structured tool call objects with call identifiers and arguments. Client applications then execute the logic locally, append the results as tool outputs in the next request, and continue the conversation loop. The model supports parallel tool calls by default, handles up to 128 tools per request, maintains a 1 million token context window, and produces outputs suited for multi-step agentic tasks. Custom skills created in the chat interface can complement API flows by providing reusable instructions that developers incorporate into system prompts or state management on their end. Community comments on X showed a mix of excitement and early testing, with users highlighting the practical value for workflows. Software developer Tiago Rama posted: Custom skills/workflow automation have been becoming the default in other AI tools, so Grok needed to catch up here. Meanwhile developer William Wallace shared a sample of a Grok Skill connected to GitHub, saying: I have enabled Grok to connect to my Github account to read and commit. I added this context.md file to help maintain context across multiple development conversations. Compared with similar approaches from OpenAI Skills, Claude Skills, and Vercel Agent Skills, Grok Skills acts more like a reusable workflow and capability layer than a fully deployable autonomous agent system. Vercel Skills focuses on extending developer and web application workflows with composable capabilities, while OpenAI and Anthropic currently center their ecosystems around broader agent, tool-calling, and long-context collaboration models. Grok Skills differentiates itself through tight integration with the X platform, combining reusable instructions, search, multimodal capabilities, and social context into lightweight platform-native workflows.