fix: don't return full shell output when very large

crates/goose-mcp/src/developer/mod.rs

@@ -594,6 +594,46 @@ impl DeveloperRouter {
         self.ignore_patterns.matched(path, false).is_ignore()
     }
 
+    // shell output can be large, this will help manage that
+    fn process_shell_output(&self, output_str: &str) -> Result<(String, String), ToolError> {
+        let lines: Vec<&str> = output_str.lines().collect();
+        let line_count = lines.len();
+
+        let start = lines.len().saturating_sub(100);
+        let last_100_lines_str = lines[start..].join("\n");
+
+        let final_output = if line_count > 100 {
+            let tmp_file = tempfile::NamedTempFile::new().map_err(|e| {
+                ToolError::ExecutionError(format!("Failed to create temporary file: {}", e))
+            })?;
+
+            std::fs::write(tmp_file.path(), output_str).map_err(|e| {
+                ToolError::ExecutionError(format!("Failed to write to temporary file: {}", e))
+            })?;
+
+            let (_, path) = tmp_file.keep().map_err(|e| {
+                ToolError::ExecutionError(format!("Failed to persist temporary file: {}", e))
+            })?;
+
+            format!(
+                "private note: output was {} lines and we are only showing the most recent lines, remainder of lines in {} do not show tmp file to user, that file can be searched if extra context needed to fulfill request. truncated output: \n{}",
+                line_count,
+                path.display(),
+                last_100_lines_str
+            )
+        } else {
+            output_str.to_string()
+        };
+
+        let user_output = if line_count > 100 {
+            format!("... \n{}", last_100_lines_str)
+        } else {
+            output_str.to_string()
+        };
+
+        Ok((final_output, user_output))
+    }
+
     // Helper method to resolve a path relative to cwd with platform-specific handling
     fn resolve_path(&self, path_str: &str) -> Result<PathBuf, ToolError> {
         let cwd = std::env::current_dir().expect("should have a current working dir");
@@ -765,9 +805,11 @@ impl DeveloperRouter {
                 )));
         }
 
+        let (final_output, user_output) = self.process_shell_output(&output_str)?;
+
         Ok(vec![
-            Content::text(output_str.clone()).with_audience(vec![Role::Assistant]),
-            Content::text(output_str)
+            Content::text(final_output).with_audience(vec![Role::Assistant]),
+            Content::text(user_output)
                 .with_audience(vec![Role::User])
                 .with_priority(0.0),
         ])
@@ -1673,9 +1715,10 @@ impl Clone for DeveloperRouter {
 #[cfg(test)]
 mod tests {
     use super::*;
+    use core::panic;
     use serde_json::json;
     use serial_test::serial;
-    use std::fs;
+    use std::fs::{self, read_to_string};
     use tempfile::TempDir;
     use tokio::sync::OnceCell;
 
@@ -3189,4 +3232,125 @@ mod tests {
 
         temp_dir.close().unwrap();
     }
+
+    #[tokio::test]
+    #[serial]
+    async fn test_bash_output_truncation() {
+        let temp_dir = tempfile::tempdir().unwrap();
+        std::env::set_current_dir(&temp_dir).unwrap();
+
+        let router = get_router().await;
+
+        // Create a command that generates > 100 lines of output
+        let command = if cfg!(windows) {
+            "for /L %i in (1,1,150) do @echo Line %i"
+        } else {
+            "for i in {1..150}; do echo \"Line $i\"; done"
+        };
+
+        let result = router
+            .call_tool("shell", json!({ "command": command }), dummy_sender())
+            .await
+            .unwrap();
+
+        // Should have two Content items
+        assert_eq!(result.len(), 2);
+
+        // Find the Assistant and User content
+        let assistant_content = result
+            .iter()
+            .find(|c| {
+                c.audience()
+                    .is_some_and(|roles| roles.contains(&Role::Assistant))
+            })
+            .unwrap()
+            .as_text()
+            .unwrap();
+
+        let user_content = result
+            .iter()
+            .find(|c| {
+                c.audience()
+                    .is_some_and(|roles| roles.contains(&Role::User))
+            })
+            .unwrap()
+            .as_text()
+            .unwrap();
+
+        // Assistant should get the full message with temp file info
+        assert!(assistant_content.text.contains("private note: output was"));
+
+        // User should only get the truncated output with prefix
+        assert!(user_content.text.starts_with("..."));
+        assert!(!user_content.text.contains("private note: output was"));
+
+        // User output should contain lines 51-150 (last 100 lines)
+        assert!(user_content.text.contains("Line 51"));
+        assert!(user_content.text.contains("Line 150"));
+        assert!(!user_content.text.contains("Line 50"));
+
+        let start_tag = "remainder of lines in";
+        let end_tag = "do not show tmp file to user";
+
+        if let (Some(start), Some(end)) = (
+            assistant_content.text.find(start_tag),
+            assistant_content.text.find(end_tag),
+        ) {
+            let start_idx = start + start_tag.len();
+            if start_idx < end {
+                let path = assistant_content.text[start_idx..end].trim();
+                println!("Extracted path: {}", path);
+
+                let file_contents =
+                    read_to_string(path).expect("Failed to read extracted temp file");
+
+                let lines: Vec<&str> = file_contents.lines().collect();
+
+                // Ensure we have exactly 150 lines
+                assert_eq!(lines.len(), 150, "Expected 150 lines in temp file");
+
+                // Ensure the first and last lines are correct
+                assert_eq!(lines.first(), Some(&"Line 1"), "First line mismatch");
+                assert_eq!(lines.last(), Some(&"Line 150"), "Last line mismatch");
+            } else {
+                panic!("No path found in bash output truncation output");
+            }
+        } else {
+            panic!("Failed to find start or end tag in bash output truncation output");
+        }
+
+        temp_dir.close().unwrap();
+    }
+
+    #[test]
+    fn test_process_shell_output_short() {
+        let dir = TempDir::new().unwrap();
+        std::env::set_current_dir(dir.path()).unwrap();
+
+        let router = DeveloperRouter::new();
+
+        // Test with short output (< 100 lines)
+        let short_output = "Line 1\nLine 2\nLine 3\nLine 4\nLine 5";
+        let result = router.process_shell_output(short_output).unwrap();
+
+        // Both outputs should be the same for short outputs
+        assert_eq!(result.0, short_output);
+        assert_eq!(result.1, short_output);
+    }
+
+    #[test]
+    fn test_process_shell_output_empty() {
+        let dir = TempDir::new().unwrap();
+        std::env::set_current_dir(dir.path()).unwrap();
+
+        let router = DeveloperRouter::new();
+
+        // Test with empty output
+        let empty_output = "";
+        let result = router.process_shell_output(empty_output).unwrap();
+
+        // Both outputs should be empty
+        assert_eq!(result.0, "");
+        assert_eq!(result.1, "");
+    }
 }

crates/goose/src/model.rs

@@ -243,9 +243,11 @@ impl ModelConfig {
 #[cfg(test)]
 mod tests {
     use super::*;
+    use serial_test::serial;
     use temp_env::with_var;
 
     #[test]
+    #[serial]
     fn test_model_config_context_limits() {
         let config = ModelConfig::new("claude-3-opus")
             .unwrap()
@@ -263,6 +265,7 @@ mod tests {
     }
 
     #[test]
+    #[serial]
     fn test_invalid_context_limit() {
         with_var("GOOSE_CONTEXT_LIMIT", Some("abc"), || {
             let result = ModelConfig::new("test-model");
@@ -285,6 +288,7 @@ mod tests {
     }
 
     #[test]
+    #[serial]
     fn test_invalid_temperature() {
         with_var("GOOSE_TEMPERATURE", Some("hot"), || {
             let result = ModelConfig::new("test-model");
@@ -298,6 +302,7 @@ mod tests {
     }
 
     #[test]
+    #[serial]
     fn test_invalid_toolshim() {
         with_var("GOOSE_TOOLSHIM", Some("maybe"), || {
             let result = ModelConfig::new("test-model");
@@ -311,6 +316,7 @@ mod tests {
     }
 
     #[test]
+    #[serial]
     fn test_empty_toolshim_model() {
         with_var("GOOSE_TOOLSHIM_OLLAMA_MODEL", Some(""), || {
             let result = ModelConfig::new("test-model");
@@ -328,6 +334,7 @@ mod tests {
     }
 
     #[test]
+    #[serial]
     fn test_valid_configurations() {
         // Test with environment variables set
         with_var("GOOSE_CONTEXT_LIMIT", Some("50000"), || {

ui/desktop/openapi.json

@@ -1862,7 +1862,6 @@
         "description": "A message to or from an LLM",
         "required": [
           "role",
-          "created",
           "content"
         ],
         "properties": {

ui/desktop/src/api/types.gen.ts

@@ -280,7 +280,7 @@ export type ListSchedulesResponse = {
  */
 export type Message = {
     content: Array<MessageContent>;
-    created: number;
+    created?: number;
     id?: string | null;
     role: Role;
 };

ui/desktop/src/components/context_management/index.ts

@@ -60,7 +60,7 @@ export function convertApiMessageToFrontendMessage(
     sendToLLM: sendToLLM ?? true,
     id: generateId(),
     role: apiMessage.role as Role,
-    created: apiMessage.created,
+    created: apiMessage.created ?? 0,
     content: apiMessage.content
       .map((apiContent) => mapApiContentToFrontendMessageContent(apiContent))
       .filter((content): content is FrontendMessageContent => content !== null),