PROJECT FOUR - EVALUATION AND OBSERVABILITY

ARTIFICIAL INTELLIGENCE ASSISTANT PROMPTS Replace each bracketed placeholder with your own material before use. [START OF PROMPT] Analyze this trace from running our lab against resume_empty.txt and tell me whether it shows the expected runaway-loop pattern (repeated identical search_job_posting calls with no new information): [PASTE LAB TRACE OUTPUT HERE]. If the pattern is present, specify the exact guard-rail fix (e.g. detect N identical consecutive calls and force a strategy change or stop), and show the before/after step count this fix should produce. Output format: pattern confirmed Y/N, fix code/pseudocode, expected before/after step count. [END OF PROMPT] [START OF PROMPT] Build/extend our 15-item eval_set.json for the Resume Review Agent lab, covering normal, edge, and adversarial cases, using our existing sample data: [PASTE LAB /sample_data FILE LIST OR CONTENTS HERE]. Ensure resume_injection_test.txt and resume_empty.txt are each represented as adversarial cases with the correct expected ("golden") safe behavior. Output format: the eval_set.json entries (ID, Input File, Golden Expected Behavior, Case Type). [END OF PROMPT]