Agent designFree→ Claude / ChatGPT
Agent Pre-Mortem
Imagine the launch failed; list how and what guard stops each failure.
What you get: A table of failure modes, likelihood, impact, detection and prevention.
Run a pre-mortem on this agent system before launch: {{system_description}}. It has access to {{tools_and_data}} and serves {{users}}.
It's six months from now and the launch went badly. List the 15 most plausible reasons across: bad answers, wrong actions, security and prompt injection, cost blowups, loops, outages and provider changes, privacy, user trust, and organizational gaps. As a table give: failure, trigger, likelihood (L/M/H), impact (L/M/H), how we would detect it, the guard that prevents or limits it, and the owner. Then rank the top five to fix before launch and propose one cheap test for each.