Flow Test

Use this skill to design tests for tasks that cannot be validated reliably with traditional unit-test assertions alone.

This skill is for flow testing: the agent performs a realistic task, records key evidence from the process, and then judges success with an explicit semantic rubric.

Invoke this skill when:

• the task depends on live or changing web content
• the output can vary but still be correct
• the workflow spans multiple model or tool steps
• intermediate evidence matters more than one exact final string
• you need to verify user intent was satisfied, not exact wording

Do not use this skill when:

• the result is deterministic and easy to assert directly
• a schema check, exact match, snapshot, or pure function test is enough
• the requirement can be covered fully by normal unit or integration tests

Objective