xAI Ships Grok Skills: Teach It Once, It Remembers Every Conversation
xAI shipped Skills on May 18, launching on Grok 4.3 across web, iOS, and Android simultaneously. The core pitch: describe a preference, workflow, or formatting rule once and Grok applies it to every subsequent conversation — no re-prompting required.
The feature places xAI squarely in competition with Anthropic’s Projects and OpenAI’s Memory, both of which offer persistent context layers. The difference is scope. Grok Skills extends past context retention into active file production.
What Ships Out of the Box
Every Grok account receives five built-in skills from xAI at launch:
- Word Documents — generates fully formatted .docx files with headings, tables, and consistent styling
- Presentations — builds complete PowerPoint-compatible slide decks with speaker notes
- Spreadsheets — produces structured Excel files with formulas, color-coded sections, and named ranges
- PDFs — creates and merges PDFs; can restructure uploaded documents
- Skill Creator — a meta-skill that saves anything Grok learned in a conversation as a reusable skill
All five are active without setup. Users who override a built-in skill with a custom version take precedence automatically.
Custom Skills in Seconds
The creation path is intentionally low-friction. Users describe a skill in conversation, upload a reference file, or write one from scratch. The example xAI highlighted: spent time perfecting an invoice format across multiple sessions — Skill Creator saves all of that as a persistent template that applies on demand going forward.
The Competitive Context
Anthropic’s Projects (launched with Claude in 2024) lets users attach files, set custom instructions, and maintain persistent context within a project boundary. OpenAI Memory, available to ChatGPT Plus users, captures preferences and facts across conversations. Neither ships with native file generation capabilities baked in as first-class skills.
xAI’s move combines both functions and extends them into office document output. Whether the execution matches the feature surface is a separate question — Grok 4.3’s baseline quality on complex reasoning tasks trails Claude Opus 4.7 and GPT-5.5 on independent benchmarks — but the product positioning is aggressive.
Skills are live now for all Grok users. No tier restriction was announced at launch.