Simplify chat navigation and activity controls

docs/development.md

@@ -128,12 +128,18 @@ message out of the queue for editing; clicking a pending message does the same.
 Escape stops the current turn. Pending messages then run in their original
 order. A failed send preserves the queue for an explicit retry.
 
-`/goal <objective>` starts work toward an objective. `/loop [interval] <prompt>`
-repeats a prompt, immediately once and then at the given interval. Intervals use
-`s`, `m` or `h`, from one second to 24 hours; the default is five minutes.
+`/goal <objective>` starts work toward an objective. `/loop 2 <prompt>` runs two
+consecutive iterations and stops. Counts range from 1 to 1000. A count written
+as "2 iterations" or "2 itérations" in the prompt is also recognized.
+`/loop 5m <prompt>` explicitly schedules repetition, with the delay counted
+after each finished turn. Intervals use `s`, `m` or `h`, from one second to
+24 hours. A loop without a count or interval is refused; there is no default timer.
 Both commands belong to Boite and work with every driver. Goals and loops can
-coexist with the agent's task list above the composer. Hover or click the task
-row to expand it; its button also works from the keyboard and on a phone.
+coexist with the agent's task list above the composer. The compact overlay shows
+the current task and progress. Only a click expands it; updates and disclosure
+do not resize the timeline. Completed tasks and goals fade out on the next user
+prompt, and newly reported work brings the task list back. Loop details show
+the latest 50 iterations with their outcome and up to 4000 characters of result.
 
 The core owns this work, so switching threads or closing a client does not
 cancel it. A goal continues through scheduled turns until the agent emits
@@ -143,6 +149,11 @@ Escape also pauses a loop between runs. The activity bar has pause, resume,
 remove and manual goal completion controls. A restarted core preserves the
 activity but requires an explicit resume.
 
+Goal instructions are assembled only when invoking a driver. The journal stores
+the visible `/goal` or `/loop` message with its kind and iteration in the text
+part. The UI also cleans up goal prompts saved by older cores and hides standalone
+completion/blocker markers, including partial markers during streaming.
+
 Tasks come from ACP plans, Codex plan notifications or successful task tools
 such as Claude's TodoWrite and TaskCreate/TaskUpdate. An agent that reports no
 tasks gets no invented task list. Pi uses the same successful-tool observation.
@@ -237,6 +248,17 @@ from exact events to polling; without the second the focus guard never starts.
 
 ## Captures
 
+`tests/e2e/header.test.ts` checks the shared header, sidebar folding and saved
+state, project groups, machine menu ordering, and prompt navigation through a
+paged, virtualized conversation. It captures desktop, phone and light-theme
+layouts on the fake client. The shell suite checks that the thread controls sit
+inside the same title bar and that dragging excludes editable controls.
+
+The prompt outline uses at most 13 entries, keeping the first and last prompts
+and seven around the reading position. Distant prompts are grouped behind a
+keyboard-accessible list, so every loaded prompt remains reachable. Desktop
+markers are 12 px apart; the compact activity panel sits 4 px above the composer.
+
 The fake client is excluded from production bundles. Tests that need it must
 use the Vite development server. `tests/e2e/settings.test.ts` starts and closes
 one within the test process; the other end-to-end paths use a real temporary
@@ -312,10 +334,11 @@ panel, paragraph buffering, reasoning replacement, goal display and command
 highlighting through the fake client. It writes desktop, phone and light-theme
 captures under `tests/e2e/.artifacts/`.
 
-Scheduled goal and loop prompts keep their execution instructions in `text` and
-carry a separate optional `displayText` on the text part. The journal retains
-both; drivers read the execution prompt. The UI shows the command and objective,
-including when recalling a sent prompt, and hides standalone goal control markers.
+Scheduled goal and loop prompts journal the command and objective in `text`, with
+activity kind and iteration metadata. The core builds the execution instructions
+when starting the driver. Older messages can carry `displayText`, which the UI
+still honors when displaying or recalling a prompt. Terminal goal control markers
+stay hidden; examples inside answer text or code fences remain visible.
 
 Chat status uses two small receipts: core acceptance and the first nonempty
 assistant activity. Agent protocols do not provide a literal read receipt.

docs/model-switching.md

@@ -85,6 +85,13 @@ continuation of an image-bearing history explicitly.
 
 ## Storage and concurrent changes
 
+The composer resolves an old `default` or `auto` model alias to the configured
+provider preset for the next prompt when one is configured. A named model stays selected. Before sending
+on an old thread, the UI verifies the preset against the account's model catalog
+and saves it with the thread's selection revision. An unavailable preset or a
+concurrent selection change refuses the send; it never silently runs the alias.
+The picker also ignores saved presets containing these aliases.
+
 Schema 9 adds `threads.session_generation`, `threads.selection_version` and
 `turns.execution`. Existing native session IDs survive migration. Execution
 snapshots record the target at acceptance, and both scheduler and driver use it.

packages/contracts/src/index.ts

@@ -438,7 +438,7 @@ export interface ImageAttachment {
 }
 
 export type MessagePart =
-  | { type: 'text'; text: string; /** User-facing command for internally scheduled prompts. */ displayText?: string }
+  | { type: 'text'; text: string; displayText?: string; activity?: { kind: 'goal' | 'loop'; iteration: number } }
   /** An image the user sent with the prompt, journalled with the message. */
   | { type: 'image'; mimeType: ImageMimeType; data: string; alt: string | null }
   /** The model's reasoning as the provider streams it, folded in the UI. */
@@ -514,9 +514,19 @@ export interface AgentTask {
 }
 
 export interface ThreadActivity {
-  goal: { objective: string; status: 'active' | 'paused' | 'complete'; iterations: number; error: string | null } | null;
-  loop: { prompt: string; intervalMs: number; status: 'active' | 'paused'; iterations: number; nextRunAt: number | null; error: string | null } | null;
+  goal: { objective: string; status: 'active' | 'paused' | 'complete'; iterations: number; error: string | null; dismissed?: boolean } | null;
+  loop: { prompt: string; intervalMs: number; maxIterations?: number | null; status: 'active' | 'paused' | 'complete'; iterations: number; nextRunAt: number | null; error: string | null; history?: ActivityIteration[] } | null;
   tasks: AgentTask[];
+  tasksDismissed?: boolean;
+}
+
+export interface ActivityIteration {
+  iteration: number;
+  turnId: TurnId;
+  status: 'running' | 'done' | 'error' | 'stopped';
+  summary: string;
+  startedAt: Timestamp;
+  finishedAt: Timestamp | null;
 }
 
 export interface Thread extends ThreadSummary {
@@ -902,7 +912,7 @@ export interface PairingGrant {
 // ---------------------------------------------------------------------------
 
 export interface RpcMethods {
-  'threads.activity.set': { params: { threadId: ThreadId; goal?: { objective: string } | null; loop?: { prompt: string; intervalMs: number } | null }; result: ThreadActivity };
+  'threads.activity.set': { params: { threadId: ThreadId; goal?: { objective: string } | null; loop?: { prompt: string; intervalMs: number; maxIterations?: number | null } | null }; result: ThreadActivity };
   'threads.activity.control': { params: { threadId: ThreadId; kind: 'goal' | 'loop'; action: 'pause' | 'resume' | 'remove' | 'complete' }; result: ThreadActivity };
   'quotas.list': { params: { refresh?: boolean }; result: AccountQuota[] };
   'quotas.configure': { params: { accountId: AccountId; enabled: boolean }; result: AccountQuota[] };

packages/core/src/activity-prompt.ts

@@ -0,0 +1,10 @@
+/** Instructions for automated turns stay out of the journal's user message. */
+export function activityPrompt(kind: 'goal' | 'loop', text: string, iteration: number): string {
+  const objective = text.replace(/^\/(?:goal|loop)\s*/, '');
+  if (kind === 'loop') return `Iteration ${iteration}. Execute the following task once for this iteration. Boite owns the repetition and stopping count; do not start another loop or repeat the task yourself. Report the result of this iteration.\n${objective}`;
+  return `Work toward this goal: ${objective}\nContinue until the objective is achieved. When you have verified completion, end your answer with [BOITE_GOAL_COMPLETE] alone on its own line, outside code blocks. If blocked or waiting for user input, explain what is missing and end with [BOITE_GOAL_BLOCKED] alone on its own line, outside code blocks.\nTrack the work with your native planning tool (Codex: update_plan; Claude: TodoWrite or TaskCreate/TaskUpdate). Boite displays those task updates in this thread. Create the plan before working and update its statuses as you verify results. The Boite goal already exists; do not create a second goal or use legacy Boite todo tools.`;
+}
+
+export function activityResult(text: string): string {
+  return text.replace(/^\s*\[BOITE_GOAL_(?:COMPLETE|BLOCKED)\]\s*$/gm, '').trim();
+}