Watch It Fail, Then Fix the Plan
Watch It Fail, Then Fix the Plan
Confident wrongness is the default agent failure. The model will keep calling a tool, or skip a tool, and still write a polished answer.
This lesson is a postmortem page: a trace of a bad run, the failure, and the smaller plan that would have caught it.
Learning outcomes
- Read a short tool trace
- Name the failure (wrong tool, skipped check, or no stop)
- Replace a vague prompt with a two-step plan
Instructions
Add an <h1> about failure, traces, or a postmortem
Hint: The page is a review, not a celebration.
Where does this go?
<!-- Step 1: h1 about failure or postmortem -->Write it on line 12 — the empty line right under that comment.
Add <section class="trace"> with at least two steps of a bad run
Add <section class="failure"> naming what went wrong
Add <section class="fix"> with a smaller plan of at least two steps
Your Code
Describe what you want to build for this step and I'll write the code — then click Apply to drop it into your editor. This step: Add an <h1> about failure, traces, or a postmortem