Module 07 Activity

Scenario

Everything in this module is a failure your current monitoring would not show you.

What you produce

A failure-mode register for your workflow, with a detection method and a response for each.

Steps

1. Describe both loop detectors - a hop ceiling and a repeat check on the same action with the same inputs - and give an example each would catch that the other would miss.
2. Take one realistic run and list every action. Mark each as asked for or not, and for anything not asked for, say which capability you would remove rather than which instruction you would add.
3. Decide what happens when the agent proposes something that is not on its tool list, and what the refusal should tell it.
4. For every fact your workflow relies on, write how often that kind of fact changes and how old is too old. Show one fact going stale during an approval pause.
5. Describe a run that would produce a confident, well-formed, wrong answer with no error and no empty output.
6. Say what would have to exist for anyone to notice it.

Evidence to hand in

• Both loop detectors with an example each catches alone.
• The action audit and the capability you would remove.
• The unregistered-tool response.
• The staleness table with one fact expiring during a pause.
• The silent-failure scenario, and what would detect it.

Review checklist

• Both detectors are described, with distinct examples.
• Over-action is answered by removing a capability, not by adding an instruction.
• The refusal tells the agent what tools do exist.
• Staleness limits differ per fact, with a stated reason.
• The silent failure is specific to your workflow, and the detection method is named.

Mark complete