fix(tui): provide fallback response when subagent produces no text output (#294)

When a subagent (Task tool) executes tool calls but the LLM produces no
textual response (only tool calls), the parent agent previously received
an empty response with just metadata. This caused confusion as the
subagent appeared to have done nothing.

This fix adds a meaningful fallback message when final_content is empty:
- If no tools were executed AND no content: indicates potential issue
- If tools were executed but no content: provides a summary with
  success/failure counts and the original task description

This ensures the parent agent always receives actionable feedback from
subagents, preventing silent failures where tools ran but no report was
generated.

Fixes the issue where research subagents could complete tool execution
(TodoWrite, LS, Read, Batch) but return nothing to the orchestrating
agent.

Co-authored-by: Droid Agent <droid@factory.ai>

Author: echobt

cortex-tui/src/runner/event_loop.rs

@@ -844,6 +844,41 @@ impl EventLoop {
                 tool_calls_executed.join("\n")
             };
 
+            // CRITICAL: Handle case where LLM produced no text output
+            // This can happen when the LLM only makes tool calls without explanation
+            // In this case, provide a meaningful summary so the parent agent gets feedback
+            let effective_content = if final_content.trim().is_empty() {
+                if tool_calls_executed.is_empty() {
+                    // No tools and no content - something went wrong
+                    format!(
+                        "The {} subagent completed but produced no output or tool calls. \
+                         This may indicate an issue with the task or model response.",
+                        subagent_type
+                    )
+                } else {
+                    // Tools executed but no text explanation - provide summary
+                    let success_count = tool_calls_executed
+                        .iter()
+                        .filter(|s| s.contains("success"))
+                        .count();
+                    let error_count = tool_calls_executed
+                        .iter()
+                        .filter(|s| s.contains("error") || s.contains("failed"))
+                        .count();
+                    format!(
+                        "The {} subagent completed {} tool call(s) ({} successful, {} failed) \
+                         but did not provide a textual summary. Task: {}",
+                        subagent_type,
+                        tool_calls_executed.len(),
+                        success_count,
+                        error_count,
+                        description
+                    )
+                }
+            } else {
+                final_content
+            };
+
             let output = format!(
                 "{}\n\n\
                  Tools executed:\n{}\n\n\
@@ -852,7 +887,7 @@ impl EventLoop {
                  agent_type: {}\n\
                  description: {}\n\
                  </task_metadata>",
-                final_content, tools_summary, id, subagent_type, description
+                effective_content, tools_summary, id, subagent_type, description
             );
 
             let duration = started_at.elapsed();