microsoftOfficial·6 Agent Skills Included

eval-guide

Plan, generate, and interpret Copilot Studio agent evaluations

Plans evaluation suites, generates importable test cases, and interprets pass/fail results for Copilot Studio agents. Eliminates guesswork around what to test, how to set pass-rate gates, and why specific eval cases fail. Walks users through a five-stage workflow with interactive review dashboards, from agent description to a SHIP, ITERATE, or BLOCK verdict.

123|20

↗

npx skills add microsoft/eval-guide --all -g -y

Available:

Routes user requests to the correct evaluation skill based on their stage, and enforces the shared methodology, CSV format, and dashboard review workflow across all six skills.

All Skills in This Repository (6)

Pure Emerald Level Indicators

📦 In Repo

microsoft

eval-generator

Generate eval test cases from an eval suite plan or agent description.

eval-triage-and-improvement

Triage Copilot Studio eval results into structured reports with owners and actions.

eval-result-interpreter

Analyzes Copilot Studio evaluation CSV results to generate a structured triage report.

eval-guide

Define agent purpose, success criteria, and evaluation scope into test artifacts.

eval-suite-planner

Generate a structured eval plan from an agent description for MS Learn Stage 1 Define.

eval-faq

Answer AI agent evaluation questions with cited Microsoft ecosystem guidance.

Official

Advanced

Frequently Asked Questions

FAQPage Schema

How to install eval-guide?▼

Run `npx skills add microsoft/eval-guide --all -g -y` in your terminal to install all six evaluation skills globally.

Do I need a running agent to use eval-guide?▼

No. The Discover, Plan, and Generate stages work from a plain-English agent description; only the Run stage requires a live Copilot Studio agent.

How does eval-guide interpret failing test results?▼

It classifies each failure as an eval-setup problem or an agent-quality problem, then returns a SHIP, ITERATE, or BLOCK verdict based on your hard gates and pass-rate targets.

Does eval-guide work with Claude Code and GitHub Copilot?▼

Yes. It ships as both a Claude Code plugin and GitHub Copilot prompt files, and all six skills behave identically in either environment.

Can non-developers use eval-guide?▼

Yes. You describe your agent in plain English and review results in interactive browser dashboards; no coding is required.