announcementschangelogproduct

This Week's Updates on Skills and Agents

This week we added connector pages for the catalog, improved the ask-assistant with memory and step-by-step planning, and updated the Skills and Agents Platform.

This week, we added connector pages to showcase the MCPs’ skills and agent use, improved the ask-assistant with memory and step-by-step planning, and made improvements to the Skills and Agents Platform.

Skills and Agents Marketplace

We added a page that shows connectors in the catalog. Connectors are the tools that enable LLMs to connect to a system of record. They can be MCPs or direct API connections. The /connectors page currently lists 25, from Gmail to QuickBooks to Slack, and each connector has a detail page that clearly explains what it can and can’t do, based on the vendor’s documentation. Skill and agent pages link to the connector pages they use, and connector pages link back to the skills that depend on them.

/agents and /plugins pages now let you view items in list or grid view, sort, and filter by source.

The ask-assistant on every page has two new features. It now remembers your recent questions, so if you ask something like “what about for agents instead?” it will understand the context. Also, if you ask a goal-based question like “I’m redesigning my website,” you’ll get a step-by-step, linked plan instead of just a generic answer.

Skills and Agents Platform

We made several updates to the skill detail page in the app this week. There’s now a breadcrumb and version chip in the header, a version history list, a section that shows every agent workflow using the skill (not just your own runs), a documentation notes section, and a working Run button.

Uploading a skill now uses a two-step guided wizard: first you upload, then you review. This matches the process already used for Add to the Brain and Maps. Building a new Map works the same way, with one flow that lets you choose between chat, step-by-step, or paste as entry methods, instead of having three separate pages.

Owners and Operations Engineers can now view Claude Code spend directly in the product. A new “Off-platform spend” card shows what Claude Code sessions reported for an organization, including the cost, right next to the existing skill-spend card. Previously, this information was only available through a database query.

The Workspace composer is now simpler. Match rows now start with the skill’s name, scope labels are shown as plain words instead of raw enums, and the composer now has two modes instead of three.

If you are new, start by learning what a skill is or browse the catalog. This update is from the Skills and Agents team. For a behind-the-scenes look at how the week’s work happened, check out Edwin’s weekly log.