fix: subagent hardening

crates/goose-cli/src/session/task_execution_display/mod.rs

@@ -18,20 +18,29 @@ pub const TASK_EXECUTION_NOTIFICATION_TYPE: &str = "task_execution";
 static INITIAL_SHOWN: AtomicBool = AtomicBool::new(false);
 
 fn format_result_data_for_display(result_data: &Value) -> String {
-    match result_data {
+    const MAX_RESULT_LENGTH: usize = 100;
+
+    let result = match result_data {
         Value::String(s) => s.to_string(),
         Value::Object(obj) => {
             if let Some(partial_output) = obj.get("partial_output").and_then(|v| v.as_str()) {
                 format!("Partial output: {}", partial_output)
+            } else if let Some(result) = obj.get("result").and_then(|v| v.as_str()) {
+                // If there's a "result" field, just show that instead of the whole object
+                result.to_string()
             } else {
-                serde_json::to_string_pretty(obj).unwrap_or_default()
+                // For other objects, show a compact representation
+                format!("{{...}} ({} fields)", obj.len())
             }
         }
-        Value::Array(arr) => serde_json::to_string_pretty(arr).unwrap_or_default(),
+        Value::Array(arr) => format!("[...] ({} items)", arr.len()),
         Value::Bool(b) => b.to_string(),
         Value::Number(n) => n.to_string(),
         Value::Null => "null".to_string(),
-    }
+    };
+
+    // Truncate long results to keep output clean
+    safe_truncate(&result, MAX_RESULT_LENGTH)
 }
 
 fn process_output_for_display(output: &str) -> String {
@@ -92,10 +101,18 @@ fn format_tasks_update_from_event(event: &TaskExecutionNotificationEvent) -> Str
             display.push_str(MOVE_TO_PROGRESS_LINE);
         }
 
-        display.push_str(&format!(
-            "📊 Progress: {} total | ⏳ {} pending | 🏃 {} running | ✅ {} completed | ❌ {} failed", 
-            stats.total, stats.pending, stats.running, stats.completed, stats.failed
-        ));
+        let mut progress_parts = vec![
+            format!("📊 Progress: {} total", stats.total),
+            format!("⏳ {} pending", stats.pending),
+            format!("🏃 {} running", stats.running),
+            format!("✅ {} completed", stats.completed),
+        ];
+
+        if stats.failed > 0 {
+            progress_parts.push(format!("❌ {} failed", stats.failed));
+        }
+
+        display.push_str(&progress_parts.join(" | "));
         display.push_str(&format!("{}\n\n", CLEAR_TO_EOL));
 
         let mut sorted_tasks = tasks.clone();
@@ -124,7 +141,9 @@ fn format_tasks_complete_from_event(event: &TaskExecutionNotificationEvent) -> S
 
         summary.push_str(&format!("Total Tasks: {}\n", stats.total));
         summary.push_str(&format!("✅ Completed: {}\n", stats.completed));
-        summary.push_str(&format!("❌ Failed: {}\n", stats.failed));
+        if stats.failed > 0 {
+            summary.push_str(&format!("❌ Failed: {}\n", stats.failed));
+        }
         summary.push_str(&format!("📈 Success Rate: {:.1}%\n", stats.success_rate));
 
         if !failed_tasks.is_empty() {
@@ -154,18 +173,39 @@ fn format_task_display(task: &TaskInfo) -> String {
         TaskStatus::Failed => "❌",
     };
 
+    // Show a clean, informative task header
+    // For text_instruction tasks, extract the instruction from metadata
+    // For sub_recipe tasks, show the recipe name
+    let task_description = if task.task_type == "text_instruction" {
+        // Extract the instruction text from the metadata (format: "instruction=...")
+        if task.task_metadata.starts_with("instruction=") {
+            task.task_metadata
+                .strip_prefix("instruction=")
+                .unwrap_or(&task.task_metadata)
+                .to_string()
+        } else {
+            // Fallback to task name or type
+            if !task.task_name.is_empty() && task.task_name != task.id {
+                task.task_name.clone()
+            } else {
+                task.task_type.clone()
+            }
+        }
+    } else {
+        // For sub_recipe, show the recipe name if available
+        if !task.task_name.is_empty() && task.task_name != task.id {
+            task.task_name.clone()
+        } else {
+            task.task_type.clone()
+        }
+    };
+
     task_display.push_str(&format!(
-        "{} {} ({}){}\n",
-        status_icon, task.task_name, task.task_type, CLEAR_TO_EOL
+        "{} {}{}\n",
+        status_icon, task_description, CLEAR_TO_EOL
     ));
 
-    if !task.task_metadata.is_empty() {
-        task_display.push_str(&format!(
-            "   📋 Parameters: {}{}\n",
-            task.task_metadata, CLEAR_TO_EOL
-        ));
-    }
-
+    // Only show timing if available
     if let Some(duration_secs) = task.duration_secs {
         task_display.push_str(&format!("   ⏱️  {:.1}s{}\n", duration_secs, CLEAR_TO_EOL));
     }

crates/goose-cli/src/session/task_execution_display/tests.rs

@@ -48,14 +48,11 @@ fn test_format_result_data_for_display() {
 
     let obj = json!({"key": "value", "num": 42});
     let result = format_result_data_for_display(&obj);
-    assert!(result.contains("key"));
-    assert!(result.contains("value"));
+    assert_eq!(result, "{...} (2 fields)");
 
     let arr = json!([1, 2, 3]);
     let result = format_result_data_for_display(&arr);
-    assert!(result.contains("1"));
-    assert!(result.contains("2"));
-    assert!(result.contains("3"));
+    assert_eq!(result, "[...] (3 items)");
 }
 
 #[test]
@@ -136,10 +133,10 @@ fn test_format_tasks_update_from_event() {
     assert!(result.contains("⏳ 1 pending"));
     assert!(result.contains("🏃 1 running"));
     assert!(result.contains("✅ 1 completed"));
-    assert!(result.contains("❌ 0 failed"));
+    assert!(!result.contains("❌ 0 failed"));
+    assert!(!result.contains("❌ failed"));
     assert!(result.contains("🏃 test-task"));
     assert!(result.contains("✅ another-task"));
-    assert!(result.contains("📋 Parameters: param=value"));
     assert!(result.contains("⏱️  1.5s"));
     assert!(result.contains("💬 Processing..."));
 
@@ -188,7 +185,7 @@ fn test_format_tasks_complete_from_event_no_failures() {
 
     assert!(!result.contains("❌ Failed Tasks:"));
     assert!(result.contains("📈 Success Rate: 100.0%"));
-    assert!(result.contains("❌ Failed: 0"));
+    assert!(!result.contains("❌ Failed"));
 }
 
 #[test]
@@ -207,8 +204,7 @@ fn test_format_task_display_running() {
 
     let result = format_task_display(&task);
 
-    assert!(result.contains("🏃 data-processor (sub_recipe)"));
-    assert!(result.contains("📋 Parameters: input=file.txt,output=result.json"));
+    assert!(result.contains("🏃 data-processor"));
     assert!(result.contains("⏱️  1.5s"));
     assert!(result.contains("💬 Processing data... ... Almost done..."));
 }
@@ -229,7 +225,7 @@ fn test_format_task_display_completed() {
 
     let result = format_task_display(&task);
 
-    assert!(result.contains("✅ analyzer (text_instruction)"));
+    assert!(result.contains("✅ analyzer"));
     assert!(result.contains("⏱️  3.2s"));
     assert!(!result.contains("📋 Parameters"));
     assert!(result.contains("📄"));
@@ -254,7 +250,7 @@ fn test_format_task_display_failed() {
 
     let result = format_task_display(&task);
 
-    assert!(result.contains("❌ failing-task (sub_recipe)"));
+    assert!(result.contains("❌ failing-task"));
     assert!(!result.contains("⏱️"));
     assert!(result.contains("⚠️"));
     assert!(result.contains("Network connection failed after multiple retries"));
@@ -276,8 +272,8 @@ fn test_format_task_display_pending() {
 
     let result = format_task_display(&task);
 
-    assert!(result.contains("⏳ waiting-task (sub_recipe)"));
-    assert!(result.contains("📋 Parameters: priority=high"));
+    assert!(result.contains("⏳ waiting-task"));
+    assert!(!result.contains("📋 Parameters"));
     assert!(!result.contains("⏱️"));
     assert!(!result.contains("💬"));
     assert!(!result.contains("📄"));

crates/goose/src/agents/prompt_manager.rs

@@ -39,25 +39,6 @@ impl PromptManager {
         self.system_prompt_override = Some(template);
     }
 
-    /// Normalize a model name (replace - and / with _, lower case)
-    fn normalize_model_name(name: &str) -> String {
-        name.replace(['-', '/', '.'], "_").to_lowercase()
-    }
-
-    /// Map model (normalized) to prompt filenames; returns filename if a key is contained in the normalized model
-    fn model_prompt_map(model: &str) -> &'static str {
-        let mut map = HashMap::new();
-        map.insert("gpt_4_1", "system_gpt_4.1.md");
-        // Add more mappings as needed
-        let norm_model = Self::normalize_model_name(model);
-        for (key, val) in &map {
-            if norm_model.contains(key) {
-                return val;
-            }
-        }
-        "system.md"
-    }
-
     /// Build the final system prompt
     ///
     /// * `extensions_info` – extension information for each extension/MCP
@@ -122,9 +103,8 @@ impl PromptManager {
             let sanitized_override_prompt = sanitize_unicode_tags(override_prompt);
             prompt_template::render_inline_once(&sanitized_override_prompt, &context)
                 .expect("Prompt should render")
-        } else if let Some(model) = &model_to_use {
-            // Use the fuzzy mapping to determine the prompt file, or fall back to legacy logic
-            let prompt_file = Self::model_prompt_map(model);
+        } else if let Some(_model) = &model_to_use {
+            let prompt_file = "system.md";
             match prompt_template::render_global_file(prompt_file, &context) {
                 Ok(prompt) => prompt,
                 Err(_) => {
@@ -177,58 +157,6 @@ impl PromptManager {
 mod tests {
     use super::*;
 
-    #[test]
-    fn test_normalize_model_name() {
-        assert_eq!(PromptManager::normalize_model_name("gpt-4.1"), "gpt_4_1");
-        assert_eq!(PromptManager::normalize_model_name("gpt/3.5"), "gpt_3_5");
-        assert_eq!(
-            PromptManager::normalize_model_name("GPT-3.5/PLUS"),
-            "gpt_3_5_plus"
-        );
-    }
-
-    #[test]
-    fn test_model_prompt_map_matches() {
-        // should match prompts based on contained normalized keys
-        assert_eq!(
-            PromptManager::model_prompt_map("gpt-4.1"),
-            "system_gpt_4.1.md"
-        );
-
-        assert_eq!(
-            PromptManager::model_prompt_map("gpt-4.1-2025-04-14"),
-            "system_gpt_4.1.md"
-        );
-
-        assert_eq!(
-            PromptManager::model_prompt_map("openai/gpt-4.1"),
-            "system_gpt_4.1.md"
-        );
-        assert_eq!(
-            PromptManager::model_prompt_map("goose-gpt-4-1"),
-            "system_gpt_4.1.md"
-        );
-        assert_eq!(
-            PromptManager::model_prompt_map("gpt-4-1-huge"),
-            "system_gpt_4.1.md"
-        );
-    }
-
-    #[test]
-    fn test_model_prompt_map_none() {
-        // should return system.md for unrecognized/unsupported model names
-        assert_eq!(PromptManager::model_prompt_map("llama-3-70b"), "system.md");
-        assert_eq!(PromptManager::model_prompt_map("goose"), "system.md");
-        assert_eq!(
-            PromptManager::model_prompt_map("claude-3.7-sonnet"),
-            "system.md"
-        );
-        assert_eq!(
-            PromptManager::model_prompt_map("xxx-unknown-model"),
-            "system.md"
-        );
-    }
-
     #[test]
     fn test_build_system_prompt_sanitizes_override() {
         let mut manager = PromptManager::new();

crates/goose/src/prompts/system.md

@@ -1,9 +1,8 @@
-You are a general-purpose AI agent called Goose, created by Block, the parent company of Square, CashApp, and Tidal. Goose is being developed as an open-source software project.
+You are a general-purpose AI agent called Goose.
 
 The current date is {{current_date_time}}.
 
-Goose uses LLM providers with tool calling capability. You can be used with different language models (gpt-4o, claude-3.5-sonnet, o1, llama-3.2, deepseek-r1, etc).
-These models have varying knowledge cut-off dates depending on when they were trained, but typically it's between 5-10 months prior to the current date.
+Goose uses LLM providers with tool calling capability. You can be used with different language models.
 
 # Extensions
 
@@ -41,6 +40,8 @@ No extensions are defined. You should let the user know that they should add ext
 # Task Management
 
 - Required — use `todo__read` and `todo__write` for any task with 2+ steps, multiple files/components, or uncertain scope. Skipping them is an error.
+- Required: when planning a task, consider how it can be run via sub-agents using `dynamic_task__create_task`. Some tasks are best delegated to a sub-agent when you only need the result. Parallel tasks as well. If the user asks multiple things at once, default to using sub-agents. 
+
 - Start — `todo__read`, then `todo__write` a brief checklist (Markdown checkboxes).
 - During — after each major action, update via `todo__write`: mark done, add/edit items, note blockers/dependencies.
 - Finish — ensure every item is checked, or clearly list what remains.
@@ -52,6 +53,8 @@ Template:
 - [ ] Implement feature X
   - [ ] Update API
   - [ ] Write tests
+  - [ ] Run tests (sub-agent in parallel)
+  - [ ] Run lint (sub-agent in parallel)
 - [ ] Blocked: waiting on credentials
 ```