Backup agents and Task Improvement

**Backup agents — a seat that fills itself** — give any seat on a Charts org chart or pipeline an ordered list of up to three backups. When the primary agent's process actually dies — a crash, an out-of-quota exit, an API that gave up — the next backup opens in that seat and receives the conversation so far, so it picks up where the last one stopped. Each backup can carry its own *Terminal when starting* preset, and closing a tab or quitting yourself never counts as a failure.

**Switch to a backup by hand** — when an agent is visibly broken but its process is still alive, one click on the live seat hands it to the next backup and carries the transcript over. Active teams read *backup 1 · took over from codex*, so a result from the backup never reads as one from your first choice.

**Set up your agents on first run** — a *Set up your agents* card on the welcome screen finds the agent CLIs installed on your machine and lists each with the exact command it will launch, behind one permission choice: **Yolo** launches each agent with its full-permission flag, **Manual** launches the plain CLI so it asks before editing a file. Only the commands 1DevTool generates ever change — your saved presets and typed commands are left alone.

**Preview a team before it opens** — every orchestration card in Add Terminal now has a Preview that plays the chart step by step: a spark rides the wire on each beat, seats change state (briefed, working, reporting, done), and a caption says what just happened — so you see how four terminals will talk to each other before they appear. A preview only explains; Start is still one deliberate click.

**Edit a task, or have an AI rewrite it** — open a task's detail dialog to edit its title, body and acceptance criteria, with the body in Markdown across Write / Preview / Split. A **Rewrite with AI** bar takes an instruction like *tighten this* or *add acceptance criteria*; the reply lands in the editor as a draft you read and Save yourself, and a **Rewrite with** picker chooses which installed AI CLI does the rewriting.

**The composer stopped asking you to speak orchestration** — Agent Input's three setup switches (Hierarchy / Pipeline / Defaults) are replaced by a small drawing of the thing itself: one brief fans out, every agent works at once, one answer merges back. Everything those switches did already lives in Charts, the dashboard and the mode banner above the composer.

**The launch screen fits the pane it's in, not the window** — a terminal pane that's short because the editor is open above it, or narrow because another terminal sits beside it, now picks its own column count and scrolls the agent cards inside itself, so Agent teams, the plain shell and the layout picker stay on screen no matter how many presets you've saved.

**Each agent card shows the command it will run** — the launch screen's cards carry the actual startup command instead of a maker tag, which moved to the tooltip.

**Click a terminal in Resource Manager to open it** — a row in the per-process breakdown now switches project, unhides a collapsed pane and brings that terminal forward.

**Orchestration sits next to the agents it opens** — Add Terminal's tabs now read AI Agents · Shells · Orchestration · Machine Terminal.