Runs

Sections can be collapsed, and Runs remembers which ones. A section with nothing in it keeps its header and says so (Nothing needs you.), so the four words are always in the same place. Every state is written as a word in the row, never only as an icon or a colour.

A run is in Needs you only while something real is blocking it, and it leaves the moment that thing is decided, by you, by another device, or by a timer.

The list is computed from what is actually blocking, not from a label a run last wrote, so a run that crashed with an old "waiting" label does not sit in Needs you forever. Approving from the Agent panel removes the row immediately, without a refresh.

Done is "finished while you were not looking". A turn you watched finish, in the window you were using, never appears there: it goes straight to Recent.

Opening a Done run (or keeping it open for a second) marks it seen. The mark is shared: if you open a run from the phone, it is seen on the desktop too. An opened Done row stays where it is, with a check and the word Seen, until you next come back to Runs (the next time it is shown, or the window regains focus). Then it moves to Recent. Nothing ever moves out from under your cursor.

Press M on a Done row to mark it seen without opening it.

• Code mode: the tray icon in the activity bar, or Ctrl+Shift+U (press it again to close the sidebar). The icon carries one badge: how many runs need you. It does not count working runs.
• Chat mode: the Runs tab of the sidebar, labelled Runs · 3 when three need you. The search box at the top of the sidebar searches runs while that tab is open.

Clicking a row opens that session in the Agent panel (Code mode) or the chat (Chat mode). A chat session opens in Chat mode and a code session in Code mode, switching mode if needed.

Runs does not make a sound or show a toast. If you want a desktop notification when a Fleet lane or a loop finishes, that is Settings → Automations → Background runs → desktop notifications, unchanged.

The New ▾ button in the Runs header (Code Mode panel, and the top of the Runs tab in Chat Mode) starts something without the composer:

• Background run… - a task in its own worktree (or this folder, per the project's background defaults). It creates a fresh session in the project.
• Parallel run… - the same task in 2 to 4 lanes, with a model per lane if you want, then compare.
• Loop… - a new loop on a schedule or when a comment keyword appears.
• Fleet dashboard and All loops · n - the Fleet and Loops panels.

The first three need a project folder and say so when there is none. Each sheet reports a problem inside itself; a run that started simply appears in the list.

A finished run shows how its result was checked: Passed, Repaired (it failed a check and was fixed), Unverified (no checks ran, or the checks could not grade it), or Failed. The word is the signal; the dot beside it only repeats it. Hover the chip for the score. These words are the same ones used everywhere Bodega reports a check, which is QEL, the Quality and Evidence Loop; the interface says "verification" and keeps the acronym for this tooltip.

The list is a tree with one tab stop: Tab enters it once and leaves it once.

Letters typed quickly jump to a run whose title starts with them. A, S and M are commands when they start a word and ordinary letters after that.

An item that does not apply to a run is not shown, so there is nothing greyed out to explain.

• Any run: Open, Copy session id.
• Waiting on an approval: Approve… (opens the session on the approval card, it never approves from the menu).
• Waiting on a question: Answer….
• A diff to review: Review diff, Apply changes…, Discard worktree… (all open the Fleet panel, where the diff and the confirmation are).
• Working or queued: Stop.
• Finished and unread: Mark seen.
• A Fleet lane or a loop run: Open in Fleet, Open in Loops.
• Finished or idle: Pin or Unpin (pinned runs sort first in their section), Archive.
• Any run: Delete…, which asks first and whose safe choice is Cancel.

Open in a split, Move to background, Run now and Take over here are not in the menu yet.

Typing in the search field filters all four sections in place by title, by the kind word (Chat, Fleet, Loop) and by the reason word (approve shell). A section with no match says No matches, and its count reads 2 of 7. Search never changes which run is open and never marks anything seen.

The list is served by GET /api/sessions/attention and kept current by status frames over the app's single WebSocket, so a status change reaches Runs without polling; once a minute Runs re-reads the list and records a [reconcile] attention_drift line in the developer console if push ever missed one. PATCH /api/sessions/:id/seen is what marks a run seen.

The bodega terminal client does not have commands for this list yet. The routes above are the contract a later CLI release reads (bodega sessions for the list, bodega attach <session> to follow one, bodega sessions seen <id> to clear unread).

Two settings belong to this view, and both are yours, not the project's: ui.runs.collapsed (which sections you collapsed) and ui.runs.pinned (the runs you pinned).