The shared working agreements for AI-driven work across the mesh: how to track it, when to stay out of the owner’s way and when to put something in front of them, and — the one that has cost real rework — not building past the brief. Distilled from Pokered Save Editor 2’s collaboration notes and Random AI Prompt’s working-agreements. The canonical copy is in the repository at hub/standards/working-rhythm.md; it complements planning (the what) with the rhythm.

The rules

• Track the work with tasks — early, comprehensively, live. Open a task list at the start of anything with more than one step, break it down properly, and keep statuses live. The task list carries the in-flight trail; it doesn’t replace the durable notes.
• Work in the background; foreground the moment it’s worth a look. Agent-driven builds, tests, runs, and git all run hidden — nothing steals the owner’s screen while you work. The other half is just as binding: when it’s ready to look at, open it, already on the right screen, without being asked.
• Adjacency is not a brief — don’t build what wasn’t asked for. A feature gets its own brief first, then research, then a design, then code. A phase does not get to absorb a neighbouring feature because the data happens to sit next to it. When a briefed feature genuinely needs an un-briefed one, it reads it — it doesn’t build a UI for it.