@shepai/cli 1.179.1-pr536.c975f4d → 1.180.0

package/dist/src/presentation/web/components/features/chat/useChatRuntime.js

@@ -366,8 +366,17 @@ export function useChatRuntime(featureId, worktreePath, options) {
             // Cancel in-flight refetches so our optimistic update isn't overwritten
             await queryClient.cancelQueries({ queryKey: chatQueryKey(featureId) });
             const previous = queryClient.getQueryData(chatQueryKey(featureId));
-            // Optimistically add user message
+            // Optimistically add user message. CRITICAL: spread `old` first
+            // and only override `messages` (and clear `streamingText`). The
+            // earlier version rebuilt the object field-by-field and silently
+            // dropped `workflow` + `turnStatus`, which made `hasPlan` flip
+            // to false for one render, briefly expanding the workflow step
+            // tracker (it lost its `collapsedSummary` flag when allDone
+            // turned undefined) before the next refetch reinstated it. The
+            // user saw a flicker of the full step list every time they typed
+            // a follow-up message after the workflow finished.
             queryClient.setQueryData(chatQueryKey(featureId), (old) => ({
+                ...(old ?? { sessionStatus: 'booting', sessionInfo: null }),
                 messages: [
                     ...(old?.messages ?? []),
                     {
@@ -379,9 +388,7 @@ export function useChatRuntime(featureId, worktreePath, options) {
                         updatedAt: new Date(),
                     },
                 ],
-                sessionStatus: old?.sessionStatus ?? 'booting',
                 streamingText: null,
-                sessionInfo: old?.sessionInfo ?? null,
             }));
             return { previous };
         },
@@ -423,14 +430,28 @@ export function useChatRuntime(featureId, worktreePath, options) {
     const stepProgress = useMemo(() => {
         const workflow = chatState?.workflow ?? null;
         if (!workflow || workflow.steps.length === 0) {
-            return { hasPlan: false, steps: [], activeStepId: null, allDone: false };
+            return {
+                hasPlan: false,
+                steps: [],
+                activeStepId: null,
+                allDone: false,
+                liveStatus: null,
+            };
         }
-        // Group messages by their persisted stepId.
+        // Group messages by their persisted stepId. Only assistant messages
+        // are surfaced in the tracker — the orchestrator persists each step's
+        // prompt as a `user`-role row with the step's id (RunWorkflowUseCase
+        // step 2+) and that row is an internal orchestration detail, not
+        // something the user typed or the agent produced. Including it would
+        // inflate every step's badge count by 1 and dump the raw prompt into
+        // the expanded body.
         const byStep = new Map();
         for (const m of messages) {
             const sid = m.stepId;
             if (!sid)
                 continue;
+            if (m.role !== InteractiveMessageRole.assistant)
+                continue;
             const list = byStep.get(sid) ?? [];
             list.push(m);
             byStep.set(sid, list);
@@ -447,13 +468,47 @@ export function useChatRuntime(featureId, worktreePath, options) {
             toolMessages: byStep.get(s.id) ?? [],
         }));
         const allDone = steps.length > 0 && steps.every((s) => s.status === 'done');
+        // Live status string for the running step card. Order of fallback
+        // mirrors the old flat-thread streaming bubble: tool / status log
+        // first (most informative — "Reading file X"), then a short prefix
+        // of the streaming reply text, then the generic "Thinking…" while
+        // we wait for the first chunk. Truncated to one line so it fits
+        // inline next to the spinner.
+        const trimmedStream = backendStreamingText?.trim() ?? '';
+        const trimmedLocal = streamingText?.trim() ?? '';
+        const streamPreview = (trimmedLocal || trimmedStream).split('\n')[0]?.slice(0, 80) ?? '';
+        let liveStatus = null;
+        if (allDone) {
+            liveStatus = null;
+        }
+        else if (statusLog) {
+            liveStatus = statusLog;
+        }
+        else if (streamPreview) {
+            liveStatus = streamPreview;
+        }
+        else if (backendTurnStatus === 'processing' ||
+            awaitingResponse ||
+            sessionStatus === 'booting') {
+            liveStatus = sessionStatus === 'booting' ? 'Waking up…' : 'Thinking…';
+        }
         return {
             hasPlan: true,
             steps,
             activeStepId: workflow.currentStepId,
             allDone,
+            liveStatus,
         };
-    }, [chatState?.workflow, messages]);
+    }, [
+        chatState?.workflow,
+        messages,
+        statusLog,
+        streamingText,
+        backendStreamingText,
+        backendTurnStatus,
+        awaitingResponse,
+        sessionStatus,
+    ]);
     // ── Initial request — the VERY first stepless user message.
     //    The application-creation flow persists it via the
     //    orchestrator's first `sendMessage.execute` BEFORE
@@ -470,18 +525,40 @@ export function useChatRuntime(featureId, worktreePath, options) {
     }, [stepProgress.hasPlan, messages]);
     const threadMessages = useMemo(() => {
         const hasPlan = stepProgress.hasPlan;
-        // When the workflow is active: ONLY show messages that have no
-        // stepId — those are out-of-band conversations (typed by the
-        // user into the composer after the workflow finishes, or
-        // ambient assistant replies). Every workflow-driven message
-        // lives inside the tracker card of its owning step, so we
-        // never want it rendered as a flat bubble above the tracker.
+        // When a workflow is active and STILL RUNNING: hide stepless
+        // assistant messages from the flat thread. These exist because of
+        // a small race window in the orchestrator — the agent can start
+        // streaming and persist a chunk before `setActiveStep` is called,
+        // leaving an assistant row with no stepId that would otherwise
+        // render as a stray bubble below the tracker. While the workflow
+        // is running, the running step's own card is the only legal
+        // surface for in-progress agent output.
         //
-        // ALSO filter out `initialRequestMessage` — the host page
-        // renders it manually above the tracker, so letting it appear
-        // again in the flat thread would duplicate it.
+        // Once the workflow is DONE (`allDone`), the orchestrator has
+        // called its final `clearActiveStep`. From this point on, every
+        // new assistant message is a legitimate follow-up reply to a
+        // user-typed prompt, and MUST be visible in the flat thread —
+        // otherwise the chat appears to ignore the user. The previous
+        // version of this filter dropped them too and that's what looked
+        // like "agent is ignoring me" after the build completed.
+        //
+        // The user's very first message (`initialRequestMessage`) is
+        // always filtered out — the host page pins it above the tracker
+        // via `<InitialRequestBubble>` so leaving it in the flat thread
+        // would duplicate it.
+        const workflowRunning = hasPlan && !stepProgress.allDone;
         const sourceMessages = hasPlan
-            ? messages.filter((m) => !m.stepId && m.id !== initialRequestMessage?.id)
+            ? messages.filter((m) => {
+                if (m.id === initialRequestMessage?.id)
+                    return false;
+                if (m.stepId)
+                    return false; // step-tagged messages live inside their card
+                if (workflowRunning && m.role === InteractiveMessageRole.assistant) {
+                    // Race-window leftover — see comment above.
+                    return false;
+                }
+                return true;
+            })
             : messages;
         const chatMessages = sourceMessages.map(toThreadMessage);
         // Merge debug bubbles into the timeline by timestamp
@@ -573,6 +650,7 @@ export function useChatRuntime(featureId, worktreePath, options) {
         options?.debugMode,
         debugEvents,
         stepProgress.hasPlan,
+        stepProgress.allDone,
     ]);
     // ── Status info for typing indicator ──────────────────────────────────
     const status = useMemo(() => {
@@ -639,10 +717,23 @@ export function useChatRuntime(featureId, worktreePath, options) {
         void queryClient.invalidateQueries({ queryKey: chatQueryKey(featureId) });
     }, [featureId, queryClient]);
     // ── Build assistant-ui runtime ──────────────────────────────────────────
+    // While a workflow is RUNNING, assistant-ui's `isRunning=true` would
+    // auto-inject an empty assistant placeholder bubble at the bottom of
+    // the thread (its built-in "thinking" indicator). The step tracker
+    // already conveys progress via the running step's spinner + live
+    // status, so that placeholder would just float as a hollow bubble.
+    // Force `isRunning=false` for the duration of the workflow so
+    // assistant-ui keeps its hands off the viewport.
+    //
+    // Once the workflow is DONE, hand control back to assistant-ui so
+    // its thinking indicator works normally for follow-up replies the
+    // user types into the composer after the build is finished.
+    const workflowInFlight = stepProgress.hasPlan && !stepProgress.allDone;
+    const runtimeIsRunning = workflowInFlight ? false : isRunning;
     const runtime = useExternalStoreRuntime({
         messages: threadMessages,
         convertMessage: useCallback((msg) => msg, []),
-        isRunning,
+        isRunning: runtimeIsRunning,
         onNew,
         onCancel: useCallback(async () => {
             setStreamingText('');

package/dist/src/presentation/web/components/features/chat/workflow-placeholder.d.ts

@@ -2,7 +2,7 @@
  * Client-side mirror of the application-creation workflow's
  * user-facing step titles + descriptions. Used to render a
  * placeholder tracker BEFORE the backend orchestrator has written
- * the real `workflow_steps` rows — the user sees all 9 pending
+ * the real `workflow_steps` rows — the user sees all 8 pending
  * cards the instant they land on the application page, instead
  * of a blank area that suddenly pops into existence a few
  * hundred milliseconds later.

package/dist/src/presentation/web/components/features/chat/workflow-placeholder.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"workflow-placeholder.d.ts","sourceRoot":"","sources":["../../../../../../../src/presentation/web/components/features/chat/workflow-placeholder.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAEH,MAAM,WAAW,eAAe;IAC9B,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;IACd,WAAW,EAAE,MAAM,CAAC;CACrB;AAED,eAAO,MAAM,sCAAsC,EAAE,eAAe,EA2BnE,CAAC"}
+{"version":3,"file":"workflow-placeholder.d.ts","sourceRoot":"","sources":["../../../../../../../src/presentation/web/components/features/chat/workflow-placeholder.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAEH,MAAM,WAAW,eAAe;IAC9B,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;IACd,WAAW,EAAE,MAAM,CAAC;CACrB;AAED,eAAO,MAAM,sCAAsC,EAAE,eAAe,EAyBnE,CAAC"}

package/dist/src/presentation/web/components/features/chat/workflow-placeholder.js

@@ -2,7 +2,7 @@
  * Client-side mirror of the application-creation workflow's
  * user-facing step titles + descriptions. Used to render a
  * placeholder tracker BEFORE the backend orchestrator has written
- * the real `workflow_steps` rows — the user sees all 9 pending
+ * the real `workflow_steps` rows — the user sees all 8 pending
  * cards the instant they land on the application page, instead
  * of a blank area that suddenly pops into existence a few
  * hundred milliseconds later.
@@ -23,12 +23,10 @@ export const APPLICATION_CREATION_PLACEHOLDER_STEPS = [
         title: 'Installing design tools',
         description: 'Adding Tailwind and essentials',
     },
-    { stepKey: 'plan', title: 'Sketching the app', description: 'Planning screens and data' },
-    { stepKey: 'components', title: 'Building the pieces', description: 'Creating reusable parts' },
     {
-        stepKey: 'content',
-        title: 'Adding realistic content',
-        description: 'Writing copy and sample data',
+        stepKey: 'components',
+        title: 'Building the pieces',
+        description: 'Designing and creating reusable parts',
     },
     { stepKey: 'wire', title: 'Connecting everything', description: 'Wiring navigation and forms' },
     {