Question 11
Domain 2An agent has been going for 47 turns with no progress on a stuck test. Which mitigation is most aligned with agentic best practice?
Correct answer: B
Explanation
Session-budget watermarks prevent endless looping by forcing a graceful stop before full exhaustion: “~80% commit + exit cleanly, ~95% hard exit.” This preserves partial progress so the next run can resume instead of wasting more turns on a stuck test.
Why each option is right or wrong
A. Increase max_tokens.
More tokens increase output length, not the agent’s ability to escape a stuck loop.
B. Add session-budget watermarks: ~80% commit + exit cleanly, ~95% hard exit. Partial progress is preserved; the next run resumes.
Agentic best practice is to impose explicit session-budget thresholds rather than letting a run continue indefinitely once it is clearly stalled. In this scenario, the relevant control is a soft stop at about 80% of the allotted budget to checkpoint work and exit cleanly, followed by a hard stop at about 95% if the agent still has not recovered; that preserves partial progress and allows the next invocation to resume from the saved state instead of burning the remaining turns on a stuck test.
C. Switch to a more powerful model mid-run.
Changing models mid-run may alter behavior, but it does not enforce bounded execution or recovery.
D. Disable the test tool until the agent commits to a written plan.
Blocking the test tool removes feedback; agents usually need tool results to validate and iterate.