Claude Code Studio v7.10.0

A turn that walked away from a background job now finishes the work instead of
printing "Done" over it; a terminal pane that loses its socket reconnects when you
touch it instead of staying dead until a page reload; and an error_max_turns that
did not come from this chat's budget is named as such rather than sending the user to
the wrong dial. Plus a split tmux window that is composited rather than cropped, a
paste button that stopped pasting twice, corrected toolbar tooltips, an Electron
popup that no longer escapes the window-open policy.

A turn that strands a background task is not a finished turn

• The gap: subtype:'success' is the one rung the auto-continue ladder does not
  cover. A run that started something with run_in_background and then ended its
  turn saying it would wait and continue reports a clean end_turn. Nothing resumes
  a headless claude -p — the process exits and takes the background shell with it —
  so the chat printed ✅ Done over work that had not happened, and the promised
  continuation could never arrive.
• Detection is structural, never textual. A Bash call carrying
  run_in_background is the same JSON on every install; user-facing prose is written
  in the UI language, so a regex over English phrases like "I'll check back" is dead
  on a French or Ukrainian one. The flag is read off a parsed object, so a foreground
  grep '"run_in_background": true' is not mistaken for a launch.
• The debt is incremental and counted across the whole turn. An agent that starts a
  background job and collects it inside the same turn — exactly what the new
  system-prompt rule asks for — is not charged a rescue run. The harvest side matches a
  Read/View of the tasks/<id>.output file the CLI actually writes, not just a
  BashOutput call: measured on 2.1.231, BashOutput is never called. A harvest pays
  down only a debt that already exists, once per shell id — so reading a leftover log
  from an earlier turn cannot pay for a launch that comes afterwards, and polling one
  job twice cannot cancel a second job that really was abandoned.
• One harvest run, and an honest ending if it is not enough. When even the rescue
  run walks away, the turn says how many tasks were left running, as a real --- status
  line so the SPA does not stamp its own "Done" badge over the warning.
• The system prompt states the constraint up front — collect a background job's
  result inside the same turn, because there is no later moment in which to get it.
  The rescue is the second line of defence, not the first.
• Kanban/scheduled tasks, multi-agent members and bots get that instruction too,
  each through the channel that actually reaches it. --system-prompt is dropped
  whenever there is a session to resume, and is absent entirely for a task with no bot —
  so for a multi-agent worker, a bot with a live session, and every Kanban task, the
  instruction rides the user turn instead. An unattended task that walks away from a
  background job is the worst version of this bug — nobody is reading that chat to
  notice. Riding the user turn also covers a Kanban task on the subscription engine
  for free: it passes no system prompt at all, but its prompt is typed into the tmux
  pane like any other.
• Known limit: a process backgrounded with shell syntax (cmd &, nohup) inside a
  foreground Bash call is not detected — separating that from a && b and 2>&1
  needs a shell parser. The system-prompt rule covers it in words.

An error_max_turns that is not ours says so (#67 follow-up)

• "Raise Max turns" is wrong advice when that is not the cap being hit. Measured
  against CLI 2.1.231, a run capped at N reports num_turns === N + 1, so a genuine
  exhaustion lands AT the cap. A stop at 3 turns against a 50-turn dial is a limit
  imposed somewhere else — a different CLI version, a settings.json, a hook on the
  machine the agent runs on — and the message now says that instead, on the FIRST
  retry notice rather than only after all three are spent.
• The remote auto-continue notice names the budget and what was spent. The local
  CLI loop has done this since it was written; the SSH one said only "resuming on
  remote", which is why diagnosing the report needed a round trip.
• The SSH auto-continue path logs. It previously logged nothing at all, on either
  the max-turns or the non-success branch.

A terminal pane that loses its socket now reconnects when you touch it

• Typing into a dead pane was a silent drain. term.onData guarded on
  readyState === 1 and did nothing else, so every keystroke vanished with no
  reconnect and no sign anything was wrong. xterm renders locally, so the pane looked
  alive — that is the "terminal freezes until I reload the browser" report.
• Paste and the send line named the problem and did nothing about it. Both toasted
  "Terminal is not connected" and returned, on the two controls that exist precisely
  because a pane has gone unresponsive. All three paths now revive the socket and
  carry the frame through, so the first thing you type is not the thing that is lost.
• A terminal socket failure used to leave no trace. ws.onclose took no argument
  (the close code was discarded) and there was no onerror at all. Both exist now, so
  the next occurrence says whether it was an abnormal close, an oversized frame or a
  server error instead of looking identical to every other cause.

Note: this fixes what the UI does about a dropped terminal socket. What drops it in
the first place is not yet identified — the close code added here is what will name it.

A split tmux window is composited, not cropped

• Claude Code's agent-teams splits whatever tmux window it runs in, one pane per
  teammate. Measured on a live session: a 155x39 window with the user's own claude
  squeezed into 46x38 on the left and four teammates stacked 108x9 down the right —
  and the viewer showed the 46-column strip and nothing else. The window is now
  composited whole.

The ›_ paste button no longer pastes again on every Enter

• Click ›_ beside a saved command or skill, press Enter to submit, and the same text
  was pasted a second time — and a third. The button kept focus, so Enter kept
  re-triggering it; it only stopped once focus moved elsewhere.

Terminal tooltips, and three of them were wrong

• The mode, agent and model toolbar buttons and the max-turns input gained tooltips,
  following the pattern the Room button already used.
• turns.tip claimed Claude "stops automatically" at the limit. It does not: it
  auto-continues up to three more times before surfacing an incomplete-run notice, so
  real usage can reach roughly four times the number shown. Corrected in all five
  locales, along with two other inaccurate tips.

A popup window no longer escapes the window-open policy

• setWindowOpenHandler was applied only to the top-level Electron window, so a
  same-origin link opened inside a popup fell back to Electron's allow-anything
  default instead of the app's own same-origin/external policy. The policy is now
  re-attached to every window it creates.
• Kanban: a stale response from a previously opened task could be appended to
  whatever modal happened to be open by the time it arrived.

Bot bubbles no longer flicker on reload

• A full browser reload mid bot-turn lost streaming.agent (in-memory only), so the
  next live chunk looked like a genuinely new speaker and opened a second bubble.

Room mode and the mid-turn bot hand-off are documented

• Both shipped as user-facing features with no mention anywhere in the README. Added
  in all three translations (EN/RU/UA).