Agent system review is essential maintenance, not optional…

When This Fires

• At every scheduled review cadence (Review new agents weekly, established ones monthly…): weekly for new agents, monthly for established ones
• When you notice that your days feel less structured than they used to — agents may have silently died
• After any major context change (job, home, relationship, health) — immediate full review
• When adding new agents — check for conflicts with existing ones (Resolve inter-agent conflicts with documented priority…)

Common Failure Mode

Treating review as optional self-improvement that gets skipped when busy: "I'll do my agent review when things calm down." Things never calm down, and the review never happens. Meanwhile, agents are silently degrading. The review must be a non-negotiable scheduled event — an agent itself, with its own trigger, conditions, and success criteria.

The Protocol

(1) Schedule regular reviews: weekly for new agents (Review new agents weekly, established ones monthly…), monthly for established ones. The review itself is an agent with a time-based trigger. (2) For each agent, ask three questions: Did it fire? Check firing rate (Measure behavioral agent progress by displacement rate…, Define agent success as 80%+ firing rate, not subjective…). If below threshold → diagnose (Diagnose before redesigning — identify whether trigger…). Did it produce intended outcomes? Check whether the behavior is still serving its original purpose. If not → the schema may need updating (Name the schema before building the agent — a reliably…). Have conditions changed? Check whether the context the agent was designed for still exists. If not → redesign or retire. (3) Update agent documentation (Document every agent with five components) with review findings. (4) Retire agents that no longer serve a purpose — dead agents clutter your system and create false confidence about behavioral coverage.