APPENDIX H: RESUME PORTFOLIO BUILDING CAPSTONE PROJECTS

ARTIFICIAL INTELLIGENCE ASSISTANT PROMPTS These two prompts reflect work a harness engineer does when an agent starts forgetting or repeating itself. Replace each bracketed placeholder with your own material before use. [START OF PROMPT] I run a research agent whose conversation grows long, and I compact it by replacing old messages with a short summary while keeping findings in a notes file. Here is an example of my current compaction summary text and my notes format: [PASTE ONE SUMMARY AND ONE NOTES EXAMPLE, WITH ANY PRIVATE DETAILS REMOVED]. Review them as a skeptical editor. Tell me what kinds of information a summary like this is likely to lose, which facts must never live only in the summary, and what fields I should add to each note, such as source, date, and confidence. Then propose five small tests that would reveal a harmful loss of information after compaction. Output format: Likely Losses, Facts That Must Live in Notes, Suggested Note Fields, Five Tests. [END OF PROMPT] [START OF PROMPT] My agent sometimes repeats work after a restart. Here is a description of how it saves progress and how it decides what to do on startup: [DESCRIBE YOUR SAVE AND STARTUP LOGIC]. Walk through at least six different moments when a run could be interrupted, such as between reading a file and saving its note, and for each moment explain what the agent would do on restart, whether work would be repeated or lost, and how to make that step safe to repeat. Output format: a table with columns Interruption Moment, What Happens on Restart, Risk, and Fix. [END OF PROMPT] PROJECT 3: THE CAREFUL CATALOG CLERK (A SAFE MCP-STYLE TOOL AGENT)