See Every Terminal

**See every terminal open on your machine** — a new **Machine** tab in Add Terminal and a **Machine** section in Mission Control list the terminal windows 1DevTool did not start — Ghostty, iTerm, WezTerm, kitty and friends — with the AI agent running inside, its working folder, and the project it probably belongs to. Scanning runs only while the list is visible, so the feature costs nothing when it is closed. A window reached through tmux is labeled as such instead of guessing at its owner. macOS and Linux today; Windows support comes in a later release.

**Continue here** — one click adopts an external agent's session into a real 1DevTool tab and resumes the conversation, so work started in another terminal doesn't have to stay there.

**A straight answer instead of a dead end** — running an attributed command (`resolve`, `team start`, task writes) from an outside terminal now explains that the window is not 1DevTool-owned and shows the commands that do work from there, replacing the cryptic "No compatible 1DevTool instance owns the calling terminal".

**`onedevtool terminal` learns `status` and `close`** — poll whether an agent tab is idle or busy in one shot, and close a tab (and its process) from the command line; a seat claimed by a live team refuses to close and points at the proper stop command.

**Every submit returns a receipt** — a submission id plus the agent's readiness at the moment the prompt landed; a prompt sent to a busy agent exits with its own code so a script knows it is queued behind the current turn, not delivered.

**Sends held for your approval say so** — a delegated prompt that 1DevTool is holding for confirmation now reports "accepted but awaiting the user's approval — not delivered yet" instead of looking like a success and leaving the sender waiting for a reply that cannot come.

**Dead shell tabs clean up after themselves** — a shell that exits cleanly closes its tab, a crashed shell restarts in place, and a shell that dies immediately twice in a row closes instead of looping. AI agent tabs are kept with their transcript, offering **Restart** and **Close** — never again a frozen pane that ignores your typing.

**A team no longer freezes behind one confirmation click** — when an agent's screen cannot prove a run finished but the matching result shows up in its transcript, the run completes; parked confirmations keep re-checking in the background and resolve themselves when the result lands late.

**Waited link sends honour late receipts** — a delivery confirmed a few seconds after the normal watch window now reports as delivered instead of failed.

**Delegated agents find version-managed tools** — headless runs see your real login-shell PATH plus asdf, mise and proto shims, so a machine whose only Node lives under a version manager no longer fails with "env: node: No such file".

**Model pickers fill correctly on Windows** — OpenCode and PI installed through npm are probed properly, AGY's new list format is understood, an unconfigured PI shows as needing sign-in instead of a probe failure, and Cursor's trailing tip line no longer appears as a selectable model named "Tip".