PROJECT ONE - A MINIMAL HARNESS

RUNNING AND OBSERVING YOUR FIRST AGENT Open the terminal in your project folder and run the two commands below. The first creates the data and the second starts the agent. [START OF BASH CODE] # Run the script that writes the fictional job posting and resumes to disk python make_sample_data.py # Run the main think/act/observe loop, which reviews every resume in turn python main_loop.py [END OF BASH CODE] You should see five lines like the following, one per resume. Your token numbers may differ slightly, and that is fine. [START OF PYTHON CODE] # Example printed outcome for the first resume: finished in 4 steps, using about 570 tokens {'resume': 'resume_01.txt', 'status': 'done', 'steps': 4, 'tokens': 570} # Example printed outcome for the second resume: finished in 4 steps, using about 552 tokens {'resume': 'resume_02.txt', 'status': 'done', 'steps': 4, 'tokens': 552} # The remaining resumes print similar outcome lines, one per resume ... [END OF PYTHON CODE] Now open the reports folder. You will find one file per resume, and the report for resume_01.txt reads like this. Jordan Sampleperson matches 4 of 4 required skills (Python, SQL, Airflow, Docker). Missing: none. Match level: strong. Next open logs/run_log.jsonl. Each line is one event, and together they tell the whole story: a model reply asking to read the file, a tool result, a reply asking to search the posting, and so on. Four steps per resume means three tool calls and one final reply. You have just watched a complete agent run, and you can reconstruct exactly what happened from the log alone, which is the test of a good log.