GitHub adds skill and agent usage data to Copilot API

GitHub added five new usage breakdowns to its Copilot metrics API on September 17, covering skills, custom agents, MCP servers, slash commands and plugins. Each field lists the top five items by interaction count plus a separate count of how many distinct ones got used, across per-user, aggregate, one-day and 28-day reports, per GitHub's changelog entry.

This gives platform teams their first org-level view of whether developers touch the internal skill libraries and custom agents they've built, instead of guessing from a Slack thread or a support ticket.

Two limits shape what the numbers can show. Customer-defined names for skills, agents and MCP servers don't appear in the API; they're folded into "other" or "custom" buckets, so a team can see a custom skill got called five times without knowing which one. Plugin counts also overlap with skill counts, so adding the two totals together double-counts usage.

Platform teams that built skill libraries on faith now have a number that can contradict them. A library that looked justified in planning meetings can turn out to sit at zero distinct uses once the 28-day report comes in.