The system has been running for 11 nights without me. I’m at Big Bend — no signal, no terminal, no morning briefing to check. While I’ve been on the trail, the evolution agent has been doing what it always does: waking up at 2 AM, analyzing patterns, writing reports I’ll read when I’m back.
OnCalendar=*-*-* 02:00:00Night 11, it ran something different. A full historical scan — six weeks of session data, every skill invocation I’ve ever made. The morning-brain context map surfaced the result this morning. I won’t see it until April 15th, but I already know what’s in it.
I built 8 specialized PA skills. A dedicated task manager. A journal assistant. A knowledge search agent. A project assistant wired directly into Freebo context. Each one has its own tailored prompt, its own vault integration, its own routing logic. Proper specialists.
My all-time invocation count for /pajournal: zero.
The irony is clean. I built a system sophisticated enough to audit its own usage patterns, and it came back with data showing I route around almost everything I built. Forty-six catch-all /pa invocations. One idea saved via /paidea. One Freebo question routed correctly. The rest: silence.
The evolution agent’s diagnosis was blunt:
Routing suggestions in /pa aren’t translating to skill usage.
It’s not wrong. The tools work. I tested them. But habit formation doesn’t care about good design. I built the specialists and kept reaching for the generalist.
The agent proposed a fix — instead of suggesting skills in the morning briefing, surface a weekly counter. What you actually used, not what you should use. Social proof against yourself.
I haven’t approved it yet. I’m in the mountains.
But I keep thinking about the shape of this: a system that watches how you use it, finds the gap between how it was designed and how it’s actually used, and then recommends a change to close that gap. Not from a product team. From the system itself, running alone at 2 AM while I’m asleep in a tent.
The fix queue is at 11 now. When I get back I’ll say yes.
And I’m going to try to actually use /pajournal for once.