How github/evals works
GitHub Copilot Evals is a plugin and VS Code extension for authoring, running, reviewing, and comparing evaluations of Copilot skills, custom agents, and MCP tools. The documented user workflow starts in the /eval commands or Evals Canvas, proceeds through eval creation and review, then an approved local run, saved results, and comparison. The README also documents setup, model-invocable skills, automation blueprints, and an MCP server. The supplied tree contains no implementation files, so undocumented component paths and internal relationships are omitted; edges shown are supported by README statements.
Components 14
Eval workflow
- Eval setup
- Eval authoring
- Eval review
- Run comparison
User interfaces
- `/eval` commands
- Evals Canvas
Plugin capabilities
- Six eval skills
- Automation blueprints
- MCP server
Execution and results
- Local eval runner
- Eval run
- Saved reports
Other components
- Project user
- Local project
Connections 17
- Project user → `/eval` commands: invokes
README.md - Project user → Evals Canvas: opens
README.md - `/eval` commands → Eval authoring: creates eval
README.md - Evals Canvas → Eval authoring: drafts eval
README.md - Evals Canvas → Eval review: reviews eval
README.md - `/eval` commands → Eval review: reviews results
README.md - `/eval` commands → Eval setup: configures
README.md - Project user → Eval setup: installs runner
README.md - Project user → Eval review: approves run
README.md - Local eval runner → Eval run: executes
README.md - Eval run → Saved reports: saves report
README.md - Evals Canvas → Saved reports: browses reports
README.md - Saved reports → Run comparison: compares runs
README.md - `/eval` commands → Run comparison: compares runs
README.md - Project user → Six eval skills: requests skill
README.md - Project user → Automation blueprints: uses blueprints
README.md - MCP server → Local project: resolves workspace
README.md