🤖 Clear agent status on stream start and encourage usage

Changes:
1. Clear agentStatus when new stream starts (unlike todos which persist)
   - Added comment explaining intentional difference from TODO behavior
   - Rationale: Status represents current activity, should reset per stream
   - Todos represent pending work, so they persist until completion

2. Updated tool description to encourage active usage:
   - "ALWAYS set a status at the start of your response"
   - "Update it as you work through different phases"
   - Added more examples of good status messages
   - Emphasizes keeping status current during work

3. Added test verifying status clears on new stream start

All tests passing (6 tests) ✅

Generated with `cmux`

Author: ammar-agent

src/utils/messages/StreamingMessageAggregator.status.test.ts

@@ -198,5 +198,64 @@ describe("StreamingMessageAggregator - Agent Status", () => {
     // Status should remain undefined
     expect(aggregator.getAgentStatus()).toBeUndefined();
   });
+
+  it("should clear agent status on stream-start (different from TODO behavior)", () => {
+    const aggregator = new StreamingMessageAggregator("2024-01-01T00:00:00.000Z");
+
+    // Start first stream and set status
+    aggregator.handleStreamStart({
+      type: "stream-start",
+      workspaceId: "workspace1",
+      messageId: "msg1",
+      model: "test-model",
+      historySequence: 1,
+    });
+
+    aggregator.handleToolCallStart({
+      type: "tool-call-start",
+      workspaceId: "workspace1",
+      messageId: "msg1",
+      toolCallId: "tool1",
+      toolName: "status_set",
+      args: { emoji: "🔍", message: "First task" },
+      tokens: 10,
+      timestamp: Date.now(),
+    });
+
+    aggregator.handleToolCallEnd({
+      type: "tool-call-end",
+      workspaceId: "workspace1",
+      messageId: "msg1",
+      toolCallId: "tool1",
+      toolName: "status_set",
+      result: { success: true, emoji: "🔍", message: "First task" },
+    });
+
+    expect(aggregator.getAgentStatus()?.message).toBe("First task");
+
+    // End first stream
+    aggregator.handleStreamEnd({
+      type: "stream-end",
+      workspaceId: "workspace1",
+      messageId: "msg1",
+      metadata: { model: "test-model" },
+      parts: [],
+    });
+
+    // Status persists after stream ends
+    expect(aggregator.getAgentStatus()?.message).toBe("First task");
+
+    // Start a NEW stream - status should be cleared
+    aggregator.handleStreamStart({
+      type: "stream-start",
+      workspaceId: "workspace1",
+      messageId: "msg2",
+      model: "test-model",
+      historySequence: 2,
+    });
+
+    // Status should be cleared on new stream start
+    expect(aggregator.getAgentStatus()).toBeUndefined();
+  });
 });
 

src/utils/tools/toolDefinitions.ts

@@ -184,8 +184,10 @@ export const TOOL_DEFINITIONS = {
   status_set: {
     description:
       "Set a status indicator to show what the agent is currently doing. " +
-      "The emoji will appear left of the streaming indicator, and the message will show on hover. " +
-      "Use this to communicate ongoing activities (e.g., '🔍 Analyzing code', '📝 Writing tests').",
+      "The emoji appears left of the streaming indicator, and the message shows on hover. " +
+      "IMPORTANT: Always set a status at the start of each response and update it as your work progresses. " +
+      "The status is cleared when a new stream starts, so you must set it again for each response. " +
+      "Use this to communicate ongoing activities (e.g., '🔍 Analyzing code', '📝 Writing tests', '🔧 Refactoring logic').",
     schema: z
       .object({
         emoji: z.string().describe("A single emoji character representing the current activity"),