durable-statelisted

# durable-state **The problem this solves:** you kick off a long run — an overnight content batch, a big client audit, a job that spans a whole afternoon — and then the laptop sleeps, the app crashes, the session context fills up and resets, or you have to switch machines. The work is *in the chat window*, so when the window dies, the work dies with it. You come back to nothing, and you don't even know how far it got. This skill fixes that by moving the run's memory **out of the chat and onto disk**. A run gets its own small folder of plain files. If anything interrupts it, the agent (or you) reopens the folder, reads **three files**, and picks up exactly where it left off — on the same machine or a different one. Nothing is lost, and you can always see what happened, in order. Think of it like a save file for a long job. You don't hold your progress in your head across a whole workday; you write it down. This does the same thing for your agent. > This is the **memory** half of always-on. The `night-shift` skill is the **contract** that bounds an unattended run (time box, read-only-by-default, a failure ledger, fail-loud-not-silent). The `host-your-agent` skill is the **plumbing** that gets the run off your laptop and onto a schedule. `durable-state` is what makes any of that survive a crash. Read `night-shift` once for the rules of the road — this skill is the seatbelt. --- ## Say this to your agent > "Before you start this long job, set up a durable-state run folder