My agent OS proposes work throughout the day that I can approve to run. It got much smarter after I gave it one thing: a story.
At first the proposals were ok, some useful, but often too generic. Never surprising or ahead of me.
The problem was that each session saw project snapshots but had no narrative. No sense of my goals, priorities, schedule, what I just finished, what’s nearing the finish line, what was starting to drift, etc.
So I built a story synthesizer.
Every hour it pulls from 10 sources: identity, goals, deadlines, sessions, projects & tasks, commits, notes, research, & more. Pipes it to Claude, which turns that into a living narrative of my current reality.
Every proposal run now reads that first and then project-specific data.
Now it sees something close to shipping and proposes the missing work before I ask: smoke tests, QA checks, signing verification, release prep, clean approval gates.
It sees another project with no real plan and goes off to do the planning legwork itself: researching gaps, mapping dependencies, laying out a realistic day-by-day build path.
It’s surfacing things I wasn’t tracking yet, gaps I didn’t notice, work that matters before I asked for it.
Part assistant, part coach, part consultant. It knows my story and pushes me towards finish lines now.