GitHub Copilot Evals
InstallSearch /github/evals

Create

Turn a real task and a clear expected outcome into an eval.

Choose your app

Find your setup instructions
Create an eval form bound to the sample release-note skill while workspace trust is off.
Describe the task and expected behavior for this source. Evals Canvas · sample data

Choose your app to see how to start. Start with the first-eval guide for the general steps.

Choose what to evaluate

Select your real skill, delegated agent, or supported tool source. Check its path when names repeat. Enter another asset path… handles source outside the discovered list; changing projects needs approval.

Define a useful check

Use requirements and repository scoring guidance, not the AI’s answer or an example threshold. Try a correct answer, a valid alternative, a plausible mistake, and an empty answer. Record unchecked requirements.

For skills, consider a request that should not trigger the skill. Using the skill does not mean its answer is correct.

Copilot asks about missing requirements. Canvas’s Send to Copilot continues in chat and retains a read-only snapshot. After saving, optionally View saved eval in Canvas where supported; nothing runs.

Review and save

Choose a configured eval folder or another safe project-relative destination. With no configured folder, Create uses evals. Destination and recovery rules preserve existing files and configuration.

Read the full proposal before saving new files. Normal host permissions or the Canvas Save action authorize the write; Evals adds no further confirmation. The Evals save service checks file format and source before saving; Run checks execution support separately. With only your app’s file tools, files are saved but not checked by the runner. Draft-only requests stay unsaved; saving never approves execution.

Your first eval · Checks and limits