🤖 fix: improve status_set tool description and display logic

[ammar-agent]

Overview

Fixes two critical bugs with the status_set tool and refines completion status examples.

Problems Fixed

1. Validation failures showed as 'completed' ✓ instead of 'failed' ✗

When status_set validation failed (e.g., invalid emoji), the tool displayed 'completed' status even though it failed. This made users think validation was silently failing, especially when the status didn't appear in the sidebar.

Root cause: Tool status determination only checked part.state === 'output-available' (meaning the tool returned a result), but didn't check whether that result indicated success or failure via result.success.

2. Status didn't persist across page reloads or workspace switches

Workspaces with successful status_set calls would lose their status after reload because loadHistoricalMessages() didn't reconstruct derived state from tool results.

Root cause: Derived state (agentStatus, currentTodos) was only updated during live streaming events, not when loading historical messages.

Solutions

1. Show 'failed' status for validation errors

• Enhanced status determination to check result.success for tools returning { success: boolean }
• Extracted hasSuccessResult() and hasFailureResult() helpers to eliminate duplication
• Display error message in StatusSetToolCall UI when validation fails

2. Shared tool result processing

Extracted processToolResult() as single source of truth for updating derived state. Called from:

1. handleToolCallEnd() - live streaming events
2. loadHistoricalMessages() - historical message loading

This ensures agentStatus and currentTodos are reconstructed uniformly whether processing live events or historical snapshots.

3. Refined completion status examples

Replaced generic examples with outcome-specific ones that show variety:

• ✅ Success: 'PR checks pass and ready to merge'
• ❌ Failure: 'CreateWorkspace Tests failed'
• ⚠️ Warning: 'Encountered serious issue with design'

Encourages agents to communicate actual outcomes, not just completion.

Implementation Details

Status determination (StreamingMessageAggregator.ts:772-789):

if (part.state === "output-available") {
  status = hasFailureResult(part.output) ? "failed" : "completed";
}

Shared result processing (StreamingMessageAggregator.ts:530-552):

private processToolResult(toolName: string, input: unknown, output: unknown): void {
  if (toolName === "todo_write" && hasSuccessResult(output)) {
    // Update todos...
  }
  if (toolName === "status_set" && hasSuccessResult(output)) {
    // Update agentStatus...
  }
}

Historical message processing (StreamingMessageAggregator.ts:199-213):

loadHistoricalMessages(messages: CmuxMessage[]): void {
  for (const message of messages) {
    this.messages.set(message.id, message);
    if (message.role === "assistant") {
      for (const part of message.parts) {
        if (isDynamicToolPart(part) && part.state === "output-available") {
          this.processToolResult(part.toolName, part.input, part.output);
        }
      }
    }
  }
}

Testing

• Added 5 new unit tests for status persistence and validation display
• All 842 tests pass
• Typecheck and lint pass

New tests:

1. Show 'failed' status when validation fails
2. Show 'completed' status when validation succeeds
3. Reconstruct agentStatus from historical messages
4. Use most recent status when loading multiple messages
5. Don't reconstruct from failed status_set

Before/After

Validation Failure Display

Before:

status_set({ emoji: "not-emoji", message: "test" })
→ Shows: completed ✓
→ Sidebar: No status appears
→ User: "Is validation failing silently?" 🤔

After:

status_set({ emoji: "not-emoji", message: "test" })
→ Shows: failed ✗ (emoji must be a single emoji character)
→ Sidebar: No status appears (correct - validation failed)
→ User: Clear feedback! ✨

Status Persistence

Before:

1. Agent sets status: "🔍 Analyzing code"
2. Reload page or switch workspace
3. Status disappears (even though it was successful)

After:

1. Agent sets status: "🔍 Analyzing code"
2. Reload page or switch workspace
3. Status persists ✅

Architecture Benefits

1. Single source of truth: One method processes tool results, regardless of source
2. Leverages existing data: Tool parts already contain everything we need
3. No special reconstruction logic: Just process parts uniformly
4. Easier to extend: Adding new derived state only requires updating processToolResult()
5. Matches architecture: Historical messages are complete snapshots

Generated with cmux

[ammario]

Depot runners are having issues per their status, so force-merging.