Four bugs in one paragraph: what I learned building an autonomous loop that chases people who don't answer

CrisisMesh is a seven-agent system designed to respond to emergencies in a school setting. When a human reports an incident, the system receives a Slack command or WhatsApp text and handles the rest. The most fascinating aspect was what occurred after the system initiated its claim-making process. The key feature, which often goes unnoticed, is that once an incident is declared, it triggers a scheduler that repeatedly checks for unaccounted individuals, pings them, and relays the information to the appropriate floor warden.

Despite rigorous testing, four bugs were discovered within this one paragraph, all of which successfully passed their respective tests. In one instance, the system sent an escalation message to Mrs. Rodriguez, who was already addressed, resulting in an endless loop where the same warden was repeatedly paged about the same person every 25 seconds.

The issue was resolved by introducing a terminal state, but it resurfaced due to a mismatch between the functions called during testing and those called during the actual loop. Another bug caused the system to incorrectly count people as reachable, even if they couldn't be reached, leading to false claims of 34 of 34 reachable individuals.

The system also reported unreachable individuals as reachable, which is a fact that requires immediate action from a commander. The issue was fixed by verifying IDs against the workspace and reporting an accurate count of 4 of 34.