Diagnose before redesigning — identify whether trigger…
When an agent fails, diagnose which component broke—trigger (never activated), condition (activated but context wasn't right), or action (executed but too vague/complex)—before attempting any redesign.
When This Fires
- Every time a behavioral agent fails — before attempting any modification
- During failure log review (Log every agent misfire with date, name, event…) when analyzing patterns
- When an agent has been failing and you're about to redesign it — pause and diagnose first
- Complements Diagnose failing behavioral agents by component — trigger… (two-week diagnosis) with a per-failure diagnostic framework
Common Failure Mode
Redesigning the entire agent after a failure: "My morning routine doesn't work, let me build a completely new one." Most failures are single-component issues. Rebuilding from scratch discards the components that were working and introduces untested new components. Diagnose the specific failure, fix that component, and preserve the rest.
The Protocol
(1) When an agent fails, ask three diagnostic questions in order: (2) Trigger check: "Did I notice the trigger event when it occurred?" If no → trigger failure. The event happened but didn't activate the agent in your awareness. Fix: increase salience (visual cue, alarm, environmental placement). (3) Condition check: "I noticed the trigger — did the conditions allow firing?" If no → condition failure. The trigger activated but context blocked the agent. Fix: broaden conditions or redesign for the context that actually occurs. (4) Action check: "Trigger fired, conditions matched — did I execute the action?" If no → action failure. The behavior was too complex, vague, or effortful. Fix: simplify (When an agent fires below 80% after 30 days, simplify…, Start every new agent at under two minutes with zero…). (5) Apply only the fix for the diagnosed component (Change one agent component per iteration — multi-variable…).