Copilot Usage Metrics Upgrade: Custom Skills and Agents Now Trackable by Name

GitHub Copilot's enterprise management capabilities have received a detailed but critical update: usage metrics can now directly display the names of custom skills, custom agents, MCP servers, slash commands, and plugins created within the organization, rather than lumping them all into "Other" or "Custom" categories. For administrators who need to evaluate the ROI of internal AI tools, this marks a leap from "knowing someone is using it" to "knowing who is using which tool."

From Vague Grouping to Precise Naming

Previously, Copilot's usage metrics reports grouped all organization-customized extensions into a single category, leaving administrators able to see only the total for a broad category without distinguishing which specific skill or agent was being invoked. After the update, each list entry can now hold up to 25 named items (up from 5), with remaining activity consolidated into an "Other" entry at the end. The way interaction counts are calculated remains unchanged, so historical data comparability is unaffected.

The core value of this change lies in its actionability. When administrators can see that one internal MCP server is being invoked frequently while another custom agent is barely used, they can adjust promotion strategies accordingly or reassess development priorities.

Slash commands are a quick interaction method in Copilot where users trigger preset operations by typing / followed by a command name in the chat box—for example, /explain to explain code, /fix to fix errors, or an organization-customized /deploy to execute deployment workflows. Custom slash commands allow teams to encapsulate common workflows into simple instructions, lowering the barrier to use. Plugins are a broader extension mechanism that can add new functional modules to Copilot or integrate third-party services. Both types of extensions, like custom skills and agents, are organization-customizable AI capabilities. Usage metrics can now display their names separately, meaning administrators can assess the adoption of different interaction modes (natural language, quick commands, plugin invocations) and thereby optimize training materials and promotion strategies.

Capacity Increase and Display Logic

Raising the item limit per list from 5 to 25 is not just a simple numerical change. For large organizations, there may be dozens of custom skills and agents in existence simultaneously, and a 5-item limit would compress a large amount of activity into "Other," losing analytical value. The 25-item capacity allows most organizations' active extensions to be presented individually.

Remaining activity beyond 25 items is still consolidated into a trailing "Other" entry, keeping reports clean. This "long-tail consolidation" design strikes a balance between readability and completeness.

Version Requirements and Historical Data Notes

To enable named display, organizations need to ensure Copilot CLI version 1.0.97 or higher, or Copilot SDK version 1.0.21 or higher. Activity generated by clients below these versions will still appear as "Other," and data from earlier dates remains as-is without retroactive name backfilling.

This means administrators comparing old and new data need to pay attention to the timing of version switches. The logic for calculating interaction counts has not changed, so overall trends remain trustworthy, but the completeness of granular dimensions will vary depending on client versions.

Practical Impact on Organizational AI Governance

This update addresses a common pain point in enterprise AI adoption: bustling during the promotion phase, blind during the operations phase. When organizations invest resources in developing internal skills and agents, the lack of fine-grained usage data makes it impossible to determine which investments have produced real value. Name-level metrics make "tracking adoption after launch" an executable process rather than an empty slogan.

For platform administrators, this also simplifies the path for security and compliance reviews—being able to clearly see which custom MCP servers are being invoked helps promptly identify unauthorized integrations or anomalous usage patterns.

Overall, this is not a flashy feature, but it fills a missing link in the enterprise Copilot management chain: the closed loop from deployment to measurement.

MCP (Model Context Protocol) is an open protocol for standardizing how AI models connect to external tools and data sources. In the Copilot context, MCP servers act as a bridge between the model and internal organizational systems (such as databases, APIs, and knowledge bases), enabling Copilot to securely access and operate these resources. Custom MCP servers are typically deployed by the organization itself and may involve sensitive data or internal processes, so monitoring their invocation is not just an efficiency concern but a key aspect of security governance. Through the naming information in usage metrics, administrators can identify which MCP servers are used frequently, which may be idle, or which pose risks of unauthorized access, allowing them to promptly adjust permissions or decommission servers no longer maintained.

One-line Summary

GitHub Copilot usage metrics now display specific names of custom skills, agents, MCP servers, and more, improving enterprise AI governance granularity.

Paragraph Summary

GitHub Copilot's enterprise management capabilities have received a detailed update: usage metrics can now directly display the specific names of custom skills, custom agents, MCP servers, slash commands, and plugins created within the organization, rather than grouping them all under "Other" or "Custom" categories. Each list entry's capacity has increased from 5 to 25, with remaining activity consolidated into a trailing "Other" entry. Enabling named display requires Copilot CLI 1.0.97+ or Copilot SDK 1.0.21+; earlier data is not backfilled, but the interaction count calculation method remains unchanged, so historical data comparability is unaffected. This update helps administrators precisely track the adoption of internal AI tools, supports promotion effectiveness evaluation and security reviews, and fills the closed loop from deployment to measurement in the enterprise Copilot management chain.