/
Navigation
Chronicles
Browse all articles
Explore
Semantic exploration
Research
Entity momentum
Nexus
Correlations & relationships
Story Arc
Topic evolution
Drift Map
Semantic trajectory animation
Posts
Analysis & commentary
Pulse API
Tech news intelligence API
Browse
Entities
Companies, people, products, technologies
Domains
Browse by publication source
Handles
Browse by social media handle
Detection
Concept Search
Semantic similarity search
High Impact Stories
Top coverage by position
Sentiment Analysis
Positive/negative coverage
Anomaly Detection
Unusual coverage patterns
Analysis
Rivalry Report
Compare two entities head-to-head
Semantic Pivots
Narrative discontinuities
Crisis Response
Event recovery patterns
Connected
Search: /
Command: ⌘K
Embeddings: large
TEXXR

Chronicles

The story behind the story

days · browse · Enter similar · o open

OpenAI quietly adopted Anthropic's “skills” mechanism in ChatGPT and Codex; ChatGPT's skills include creating and modifying spreadsheets, docx files, and PDFs

One of the things that most excited me about Anthropic's new Skills mechanism back in October is how easy it looked for other platforms to implement.

Simon Willison's Weblog Simon Willison

Context & Ripple Effects

Anthropic had framed Skills as folders of instructions, scripts, and resources for task-specific performance in its October Skills launch. OpenAI had already been moving ChatGPT toward bundled tool access and document analysis, amid concerns that this could sideline third-party ChatGPT plugins.

Adopting a reusable skills layer extends that trajectory from individual tools toward packaged workflows. It also provides a technical precursor to the shared, Codex-powered workspace agents OpenAI later described as an evolution of GPTs.

First-order effects

  • ChatGPT and Codex gain reusable task packages for creating and editing spreadsheets, DOCX files, and PDFs, making those document workflows more directly available inside OpenAI’s products.
  • Anthropic’s Skills design is validated as a mechanism that can be implemented beyond Claude, even as OpenAI applies it within its own product stack.

Second-order effects

  • Competition between ChatGPT and Claude shifts further toward the quality and breadth of reusable workflow packages, not only the underlying model or single-tool access.
  • The move increases pressure on standalone ChatGPT plugins and document-automation add-ons where their core value is now available through integrated skills.

Third-order effects

  • If skills become a common assistant primitive, AI products may increasingly compete as configurable work surfaces: reusable instructions, resources, and tools layered on top of general-purpose models.
  • That would favor platforms able to connect skills to shared agents and work context, while leaving narrower tool vendors to differentiate through specialized workflows or integrations.

The trend: Generative-AI assistants are evolving from prompt interfaces into workflow-native work surfaces built from reusable task components and agents.

Discussion

  • @morqon Morgan on x
    what the year of enterprise looks like
  • @natebjones Nate on x
    chalk another one up for the anthropic team injecting an AI primitive into the broader ecosystem
  • @simonw Simon Willison on x
    ChatGPT has skills for creating and manipulating spreadsheets, docx files and PDFs tucked away in a new /home/oai/skills folder in their Code Interpreter environment - you can ask it for a zip file to see the code, here's what it gave me back: https://tools.simonwillison.net/ ...…
  • @simonw Simon Willison on x
    Whoa. ChatGPT has a /home/oai/skills folder now! [image]
  • @nummanthinks Numman Ali on x
    @simonw Yes they've already added experimental support for skills in the Codex CLI, which likely powers the ChatGPT sandbox to a degree https://github.com/...
  • @simonw Simon Willison on x
    Meanwhile Codex CLI added experimental support for skills a couple of weeks ago, documented here: https://github.com/... I used that to run a new Datasette plugin skill and create a cowsay Datasette plugin from a single prompt: https://simonwillison.net/...
  • @simonw Simon Willison on x
    I used ChatGPT's new PDF skill to have it create a report on this year's Kākāpō breeding season - it took 11 minutes because it kept rendering, noticing and then fixing issues like the ā-ā-ō letters not appearing correctly. Here's the final PDF it made: https://tools.simonwilliso…
  • @_simonsmith Simon Smith on x
    Verified this with GPT-5.2 Thinking! It even overcame my typo. GPT-5.2 uses skills, which Anthropic introduced. This suggests skills may be come a standard like MCPs, which is fantastic. [image]
  • @simonw Simon Willison on x
    OpenAI aren't talking about it yet, but it turns out they've adopted Anthropic's brilliant “skills” mechanism in a big way Skills are now live in both ChatGPT and their Codex CLI tool, I wrote up some detailed notes on how they work so far here: https://simonwillison.net/...
  • @simonw Simon Willison on x
    I got curious and decided to dig around in the Codex CLI source code to see if I could find the prompts that drive their skills implementation - here they are, I extracted them from the Rust code and formatted as Markdown in a gist for better readability https://gist.github.com/.…
  • @joedevon Joe Devon on x
    @simonw Since everything is prompts all the way down I'd often just point a rule/skill/plugin/etc built in a different tool and it tends to work really well. lol. That said, Claude Code is best at using any kind of tool
  • @mweinbach Max Weinbach on x
    These are the skills OpenAI included with GPT 5.2 in ChatGPT Hopefully they add more in the future! [image]