What changed
The instructions you write for Claude can now be read by more than 70 AI tools, ChatGPT included
A skill is a folder of written instructions, sometimes with small scripts, that teaches an AI assistant to do one job your way: how you price a quote, how you check an invoice. On October 22nd Anthropic handed the format, called Agent Skills, to the Agentic AI Foundation, which Anthropic, Block and OpenAI set up under the Linux Foundation. Anthropic says more than 70 tools already read it, including ChatGPT and GitHub Copilot, and counts about 1 million published skills and 280 million installs.
Until October 22nd, writing down how your team does a job as a Claude skill was a bet on Claude. Now it works more like a procedure manual that stays yours whichever AI your team uses next year, so the effort worth making is getting your processes written down properly.
Each tool still reads the folder its own way, so a skill that runs small scripts may work in one and stall in another.
Try: load one skill your team already uses in Claude into ChatGPT, give both the same task, and see where the answers differ.
Companies that already buy from Google Cloud can now buy Claude the same way
Since October 22nd, Anthropic runs the full Claude Platform, the version your developers and business software connect to, inside Google Cloud. Access goes through your Google Cloud accounts and the charges go on your Google Cloud bill.
If your company already has a Google Cloud contract, your IT or finance team can now buy Claude without setting up a new supplier, and the spend may count toward what you have already committed to Google.
From TigerTiger
What we shipped: three new skills and six updates
All nine use the format from our first item and reach your Claude through your TigerTiger account. One, save-learnings, is in our free Memory plan, and the other eight come with Pro.
New:
better: rewrites a request the way an expert would brief it, including how to check the result.
research-plan: gathers context, then writes a plan with steps, a finish line and the open questions.
skill-quality-criteria: scores one of your skills and says whether Claude does that job just as well without it.
Updated:
writing-critic: now also reviews how-to guides and process notes.
save-learnings: adds to your existing note on a topic before starting a new one.
wrap: checks claims such as "sent" before writing them into your end-of-day notes.
spring-cleaning: reads your usage report before its quarterly check of your Claude setup, then flags instructions that conflict or go unused.
use-case-discovery: offers 12 plain-language ideas matched to your department.
url-research: reads the image attached to a post before deciding a claim has no source.
Source: TigerTiger platform, October 14th to 22nd