Module assessment

1.

Your meta-agent planner produces a decomposition plan, but early execution results reveal that the original plan was built on a flawed assumption about the data structure. What pattern should you apply?

Discard all intermediate results, fix the meta-agent's prompt, and start the entire plan from scratch.

Continue executing the original plan regardless—replanning mid-execution introduces unpredictable state.

Pass the intermediate results and the discovered discrepancy back to the meta-agent, which reevaluates and updates the remaining plan while preserving completed steps.

2.

A receiving agent returns an acknowledgment rejecting a handoff from an analysis agent, citing 'task outside domain.' How should the orchestrator handle this acknowledgment?

Treat the rejection as an agent failure and trigger the circuit-breaker for the receiving agent.

Log the rejection with the reason, query the registry for an alternative agent with the required capability, and reissue the handoff.

Add the rejected task back to the original sending agent's queue. Provide fallback instructions that specify how to handle the rejection without specialized knowledge.

3.

You're comparing two decomposition approaches for a financial research task. Approach A uses six specialized agents. Approach B uses two agents. Your evaluation shows similar quality scores but Approach B completes in 40% of the time at 35% of the cost. What does this evaluation indicate?

Approach A should be used in production because more specialized agents produce higher-quality research in edge cases not covered by the evaluation.

Approach A exhibits over-decomposition: the four extra agents are adding coordination overhead without proportional quality improvement, suggesting task boundaries were drawn at suboptimal granularity.

The evaluation dataset is insufficient—add more test cases before making a granularity decision.

Feedback