Read the run as a sequence
Start at the end of the step history and find the last action that completed. Then inspect the following step or stop reason for an error, a permission boundary, or an incomplete result. This sequence helps distinguish an agent that made no progress from one that completed part of the task and then encountered a problem.
Compare the inputs and outputs around that point with the task you gave. A missing file, an ambiguous instruction, or an unavailable resource may explain the stop. Avoid repeating the entire request until you know whether the previous steps already changed data or produced a usable artifact.
Choose a safe next step
If the run stopped before a consequential action, clarify the missing information and retry only the unfinished work. If it produced a partial result, preserve that output and ask for a focused continuation. Review any proposed write, send, or publish action before allowing it to proceed.
When reporting a problem, include the run identifier, the last completed step, and the visible stop reason, while removing private content from logs. This gives support enough context to investigate without exposing more data than necessary. A clear run history turns an unclear stop into a specific issue you can address.