Module 01 Summary

What this module established

Four pieces - Agent, Task, Crew, Flow - and most workflows need two. A task's expected_output is the contract that makes everything else checkable, and 'a good reply' is not one. A second agent earns its place through different instructions, tools or permissions; a long prompt is not one of those reasons.

Carry forward

• Score proposed agents on tools, permissions and judgement. Anything clearing fewer than two bars is a task in disguise.
• Write-access agents stay separate regardless of the score - permissions override the heuristic.
• Read a crew run as four numbers per task: agent, tokens, seconds, contract compliance. Reading the final prose tells you nothing about which agent caused what.

Before moving on

Move on when you have reduced an over-split proposal using the three questions, and every remaining task has an expected_output a reviewer could reject against.

Mark complete