Evaluate

This is how you actually test whether your agent workflow works in practice. It walks you through five evaluation dimensions (task completion, output quality, error behavior, user experience, consistency) and forces you to run concrete scenarios including edge cases and adversarial inputs. You get a structured table to document what happened versus what should have happened, then a graded report with specific improvement actions. The real value is that it won't let you skip the uncomfortable tests. Most people only check the happy path, but this pushes you to test malformed input, tool failures, and tricky cases. If you're shipping an agent workflow to users, run this first and prepare to feel slightly embarrassed by what breaks.

skills.sh installs: 363
GitHub stars: 602
last updated: Aug 21, 2026

View on GitHub

Install

$ npx -y skills add sharpdeveye/maestro --skill evaluate --agent claude-code

Installs into .claude/skills of the current project.

Files

MANDATORY PREPARATION

Invoke /agent-workflow — it contains workflow principles, anti-patterns, and the Context Gathering Protocol. Follow the protocol before proceeding — if no workflow context exists yet, you MUST run /teach-maestro first. Consult the feedback-loops reference in the agent-workflow skill for evaluation patterns, golden test sets, and regression detection.

Evaluate the workflow's actual interaction quality by testing it against scenarios that represent real usage.