Optimize tool summarization (#7938)

Co-authored-by: Michael Neale <michael.neale@gmail.com>
Co-authored-by: Douwe Osinga <douwe@squareup.com>

crates/goose/src/agents/agent.rs

@@ -363,15 +363,29 @@ impl Agent {
             self.tool_inspection_manager.apply_tool_annotations(&tools);
         }
 
+        let tool_call_cut_off = match Config::global().get_param::<usize>("GOOSE_TOOL_CALL_CUTOFF")
+        {
+            Ok(v) => v,
+            Err(_) => {
+                let context_limit = self
+                    .provider()
+                    .await
+                    .map(|p| p.get_model_config().context_limit())
+                    .unwrap_or(crate::model::DEFAULT_CONTEXT_LIMIT);
+                let compaction_threshold = Config::global()
+                    .get_param::<f64>("GOOSE_AUTO_COMPACT_THRESHOLD")
+                    .unwrap_or(crate::context_mgmt::DEFAULT_COMPACTION_THRESHOLD);
+                crate::context_mgmt::compute_tool_call_cutoff(context_limit, compaction_threshold)
+            }
+        };
+
         Ok(ReplyContext {
             conversation,
             tools,
             toolshim_tools,
             system_prompt,
             goose_mode,
-            tool_call_cut_off: Config::global()
-                .get_param::<usize>("GOOSE_TOOL_CALL_CUTOFF")
-                .unwrap_or(10),
+            tool_call_cut_off,
             initial_messages,
         })
     }
@@ -1127,6 +1141,15 @@ impl Agent {
             });
         }
 
+        // Count tool calls present before this reply — everything added during
+        // the reply loop is part of the current turn and should not be summarized.
+        let pre_turn_tool_count = conversation
+            .messages()
+            .iter()
+            .flat_map(|m| m.content.iter())
+            .filter(|c| matches!(c, MessageContent::ToolRequest(_)))
+            .count();
+
         let working_dir = session.working_dir.clone();
         let reply_stream_span = tracing::info_span!(target: "goose::agents::agent", "reply_stream", session.id = %session_config.id);
         let inner = Box::pin(async_stream::try_stream! {
@@ -1162,13 +1185,6 @@ impl Agent {
                     break;
                 }
 
-                let tool_pair_summarization_task = crate::context_mgmt::maybe_summarize_tool_pair(
-                    self.provider().await?,
-                    session_config.id.clone(),
-                    conversation.clone(),
-                    tool_call_cut_off,
-                );
-
                 let conversation_with_moim = super::moim::inject_moim(
                     &session_config.id,
                     conversation.clone(),
@@ -1185,6 +1201,20 @@ impl Agent {
                     &toolshim_tools,
                 ).await?;
 
+                let current_turn_tool_count = conversation.messages().iter()
+                    .flat_map(|m| m.content.iter())
+                    .filter(|c| matches!(c, MessageContent::ToolRequest(_)))
+                    .count()
+                    .saturating_sub(pre_turn_tool_count);
+
+                let tool_pair_summarization_task = crate::context_mgmt::maybe_summarize_tool_pairs(
+                    self.provider().await?,
+                    session_config.id.clone(),
+                    conversation.clone(),
+                    tool_call_cut_off,
+                    current_turn_tool_count,
+                );
+
                 let mut no_tools_called = true;
                 let mut messages_to_add = Conversation::default();
                 let mut tools_updated = false;
@@ -1641,34 +1671,40 @@ impl Agent {
                     }
                 }
 
-                if let Ok(Some((summary_msg, tool_id))) = tool_pair_summarization_task.await {
+                if is_token_cancelled(&cancel_token) {
+                    tool_pair_summarization_task.abort();
+                }
+
+                if let Ok(summaries) = tool_pair_summarization_task.await {
                     let mut updated_messages = conversation.messages().clone();
 
-                    let matching: Vec<&mut Message> = updated_messages
-                        .iter_mut()
-                        .filter(|msg| {
-                            msg.id.is_some() && msg.content.iter().any(|c| match c {
-                                MessageContent::ToolRequest(req) => req.id == tool_id,
-                                MessageContent::ToolResponse(resp) => resp.id == tool_id,
-                                _ => false,
+                    for (summary_msg, tool_id) in summaries {
+                        let matching: Vec<&mut Message> = updated_messages
+                            .iter_mut()
+                            .filter(|msg| {
+                                msg.id.is_some() && msg.content.iter().any(|c| match c {
+                                    MessageContent::ToolRequest(req) => req.id == tool_id,
+                                    MessageContent::ToolResponse(resp) => resp.id == tool_id,
+                                    _ => false,
+                                })
                             })
-                        })
-                        .collect();
+                            .collect();
 
-                    if matching.len() == 2 {
-                        for msg in matching {
-                            let id = msg.id.as_ref().unwrap();
-                            msg.metadata = msg.metadata.with_agent_invisible();
-                            SessionManager::update_message_metadata(&session_config.id, id, |metadata| {
-                                metadata.with_agent_invisible()
-                            }).await?;
+                        if matching.len() == 2 {
+                            for msg in matching {
+                                let id = msg.id.as_ref().unwrap();
+                                msg.metadata = msg.metadata.with_agent_invisible();
+                                SessionManager::update_message_metadata(&session_config.id, id, |metadata| {
+                                    metadata.with_agent_invisible()
+                                }).await?;
+                            }
+                            messages_to_add.push(summary_msg);
+                        } else {
+                            warn!("Expected a tool request/reply pair, but found {} matching messages",
+                                matching.len());
                         }
-                        conversation = Conversation::new_unvalidated(updated_messages);
-                        messages_to_add.push(summary_msg);
-                    } else {
-                        warn!("Expected a tool request/reply pair, but found {} matching messages",
-                            matching.len());
                     }
+                    conversation = Conversation::new_unvalidated(updated_messages);
                 }
 
                 for msg in &messages_to_add {

crates/goose/src/context_mgmt/mod.rs

@@ -18,10 +18,13 @@ use tracing::log::warn;
 
 pub const DEFAULT_COMPACTION_THRESHOLD: f64 = 0.8;
 
-/// Feature flag to enable/disable tool pair summarization.
-/// Set to `false` to disable summarizing old tool call/response pairs.
-/// TODO: Re-enable once tool summarization stability issues are resolved.
-const ENABLE_TOOL_PAIR_SUMMARIZATION: bool = false;
+const TOOLCALL_SUMMARIZATION_BATCH_SIZE: usize = 10;
+
+fn tool_pair_summarization_enabled() -> bool {
+    Config::global()
+        .get_param::<bool>("GOOSE_TOOL_PAIR_SUMMARIZATION")
+        .unwrap_or(true)
+}
 
 const CONVERSATION_CONTINUATION_TEXT: &str =
     "Your context was compacted. The previous message contains a summary of the conversation so far.
@@ -418,13 +421,24 @@ fn format_message_for_compacting(msg: &Message) -> String {
     }
 }
 
-/// Find the id of a tool call to summarize. We only do this if we have more than
-/// cutoff tool calls that aren't summarized yet
-pub fn tool_id_to_summarize(conversation: &Conversation, cutoff: usize) -> Option<String> {
+pub fn compute_tool_call_cutoff(context_limit: usize, compaction_threshold: f64) -> usize {
+    let threshold = if compaction_threshold > 0.0 && compaction_threshold <= 1.0 {
+        compaction_threshold
+    } else {
+        DEFAULT_COMPACTION_THRESHOLD
+    };
+    let effective_limit = (context_limit as f64 * threshold) as usize;
+    (3 * effective_limit / 20_000).clamp(10, 500)
+}
+
+pub fn tool_ids_to_summarize(
+    conversation: &Conversation,
+    cutoff: usize,
+    protect_last_n: usize,
+) -> Vec<String> {
     let messages = conversation.messages();
 
-    let mut tool_call_count = 0;
-    let mut first_tool_call_id = None;
+    let mut tool_call_ids: Vec<String> = Vec::new();
 
     for msg in messages.iter() {
         if !msg.is_agent_visible() {
@@ -433,17 +447,21 @@ pub fn tool_id_to_summarize(conversation: &Conversation, cutoff: usize) -> Optio
 
         for content in &msg.content {
             if let MessageContent::ToolRequest(req) = content {
-                if first_tool_call_id.is_none() {
-                    first_tool_call_id = Some(req.id.clone());
-                }
-                tool_call_count += 1;
-                if tool_call_count > cutoff {
-                    return first_tool_call_id;
-                }
+                tool_call_ids.push(req.id.clone());
             }
         }
     }
-    None
+
+    // Never summarize the last N tool calls (current turn)
+    let eligible = tool_call_ids.len().saturating_sub(protect_last_n);
+    if eligible <= cutoff + TOOLCALL_SUMMARIZATION_BATCH_SIZE {
+        return Vec::new();
+    }
+
+    tool_call_ids
+        .into_iter()
+        .take(TOOLCALL_SUMMARIZATION_BATCH_SIZE)
+        .collect()
 }
 
 pub async fn summarize_tool_call(
@@ -482,17 +500,16 @@ pub async fn summarize_tool_call(
     let summarization_request = vec![user_message];
 
     let system_prompt = indoc! {r#"
-                Your task is to summarize a tool call & response pair to save tokens
+                Your task is to summarize a tool call & response pair to save tokens.
 
-                reply with a single message that describe what happened. Typically a toolcall
-                is asks for something using a bunch of parameters and then the result is also some
+                Reply with a single message that describes what happened. Typically a tool call
+                asks for something using a bunch of parameters and then the result is also some
                 structured output. So the tool might ask to look up something on github and the
                 reply might be a json document. So you could reply with something like:
 
                 "A call to github was made to get the project status"
 
                 if that is what it was.
-
             "#};
 
     let (mut response, _) = provider
@@ -506,31 +523,30 @@ pub async fn summarize_tool_call(
     Ok(response.with_generated_id())
 }
 
-pub fn maybe_summarize_tool_pair(
+pub fn maybe_summarize_tool_pairs(
     provider: Arc<dyn Provider>,
     session_id: String,
     conversation: Conversation,
     cutoff: usize,
-) -> JoinHandle<Option<(Message, String)>> {
+    protect_last_n: usize,
+) -> JoinHandle<Vec<(Message, String)>> {
     tokio::spawn(async move {
-        // Tool pair summarization is currently disabled via feature flag.
-        // See ENABLE_TOOL_PAIR_SUMMARIZATION constant above.
-        if !ENABLE_TOOL_PAIR_SUMMARIZATION {
-            return None;
+        if !tool_pair_summarization_enabled() || provider.manages_own_context() {
+            return Vec::new();
         }
 
-        if let Some(tool_id) = tool_id_to_summarize(&conversation, cutoff) {
+        let tool_ids = tool_ids_to_summarize(&conversation, cutoff, protect_last_n);
+        let mut results = Vec::new();
+        for tool_id in tool_ids {
             match summarize_tool_call(provider.as_ref(), &session_id, &conversation, &tool_id).await
             {
-                Ok(summary) => Some((summary, tool_id)),
+                Ok(summary) => results.push((summary, tool_id)),
                 Err(e) => {
                     warn!("Failed to summarize tool pair: {}", e);
-                    None
                 }
             }
-        } else {
-            None
         }
+        results
     })
 }
 
@@ -544,6 +560,30 @@ mod tests {
     use async_trait::async_trait;
     use rmcp::model::{AnnotateAble, CallToolRequestParams, RawContent, Tool};
 
+    fn create_tool_pair(
+        call_id: &str,
+        response_id: &str,
+        tool_name: &str,
+        response_text: &str,
+    ) -> Vec<Message> {
+        vec![
+            Message::assistant()
+                .with_tool_request(
+                    call_id,
+                    Ok(CallToolRequestParams::new(tool_name.to_string())),
+                )
+                .with_id(call_id),
+            Message::user()
+                .with_tool_response(
+                    call_id,
+                    Ok(rmcp::model::CallToolResult::success(vec![
+                        RawContent::text(response_text).no_annotation(),
+                    ])),
+                )
+                .with_id(response_id),
+        ]
+    }
+
     struct MockProvider {
         message: Message,
         config: ModelConfig,
@@ -677,125 +717,86 @@ mod tests {
         );
     }
 
-    #[tokio::test]
-    async fn test_tool_pair_summarization_workflow() {
-        fn create_tool_pair(
-            call_id: &str,
-            response_id: &str,
-            tool_name: &str,
-            response_text: &str,
-        ) -> Vec<Message> {
-            vec![
-                Message::assistant()
-                    .with_tool_request(
-                        call_id,
-                        Ok(CallToolRequestParams::new(tool_name.to_string())),
-                    )
-                    .with_id(call_id),
-                Message::user()
-                    .with_tool_response(
-                        call_id,
-                        Ok(rmcp::model::CallToolResult::success(vec![
-                            RawContent::text(response_text).no_annotation(),
-                        ])),
-                    )
-                    .with_id(response_id),
-            ]
+    #[test]
+    fn test_compute_tool_call_cutoff_scales_with_context() {
+        // Default threshold (0.8)
+        assert_eq!(compute_tool_call_cutoff(128_000, 0.8), 15); // 102K effective
+        assert_eq!(compute_tool_call_cutoff(200_000, 0.8), 24); // 160K effective
+        assert_eq!(compute_tool_call_cutoff(1_000_000, 0.8), 120); // 800K effective
+                                                                   // Clamp at minimum
+        assert_eq!(compute_tool_call_cutoff(50_000, 0.8), 10);
+        assert_eq!(compute_tool_call_cutoff(10_000, 0.8), 10);
+        // Clamp at maximum (500)
+        assert_eq!(compute_tool_call_cutoff(10_000_000, 0.8), 500);
+        // Lower compaction threshold means earlier summarization
+        assert_eq!(compute_tool_call_cutoff(200_000, 0.3), 10); // 60K effective
+        assert_eq!(compute_tool_call_cutoff(1_000_000, 0.5), 75); // 500K effective
+                                                                  // Invalid threshold falls back to default 0.8
+        assert_eq!(compute_tool_call_cutoff(200_000, 0.0), 24); // falls back to 0.8
+        assert_eq!(compute_tool_call_cutoff(200_000, -1.0), 24); // falls back to 0.8
+    }
+
+    #[test]
+    fn test_tool_ids_to_summarize_triggers_at_cutoff_plus_batch() {
+        // cutoff=5, so we need >5+10=15 to trigger. 15 exactly should NOT trigger.
+        let mut messages = vec![Message::user().with_text("hello")];
+        for i in 0..15 {
+            messages.extend(create_tool_pair(
+                &format!("call{}", i),
+                &format!("resp{}", i),
+                "read_file",
+                "content",
+            ));
         }
+        let conversation = Conversation::new_unvalidated(messages);
+        let result = tool_ids_to_summarize(&conversation, 5, 0);
+        assert!(result.is_empty(), "Exactly cutoff+batch should not trigger");
 
-        let summary_response = Message::assistant()
-            .with_text("Tool call to list files and response with file listing");
-        let provider = MockProvider::new(summary_response, 1000);
-
-        let mut messages = vec![Message::user().with_text("list files").with_id("msg_1")];
-        messages.extend(create_tool_pair(
-            "call1",
-            "response1",
-            "shell",
-            "file1.txt\nfile2.txt",
-        ));
-        messages.extend(create_tool_pair(
-            "call2",
-            "response2",
-            "read_file",
-            "content of file1",
-        ));
-        messages.extend(create_tool_pair(
-            "call3",
-            "response3",
-            "read_file",
-            "content of file2",
-        ));
+        // 16 tool calls: now exceeds cutoff+10, should return a batch of 10
+        let mut messages = vec![Message::user().with_text("hello")];
+        for i in 0..16 {
+            messages.extend(create_tool_pair(
+                &format!("call{}", i),
+                &format!("resp{}", i),
+                "read_file",
+                "content",
+            ));
+        }
+        let conversation = Conversation::new_unvalidated(messages);
+        let result = tool_ids_to_summarize(&conversation, 5, 0);
+        assert_eq!(result.len(), TOOLCALL_SUMMARIZATION_BATCH_SIZE);
+        assert_eq!(result[0], "call0");
+        assert_eq!(result[9], "call9");
+    }
 
+    #[test]
+    fn test_tool_ids_to_summarize_protects_current_turn() {
+        // 20 tool pairs, cutoff=2 → 20 > 12, would normally trigger
+        let mut messages = vec![Message::user().with_text("hello")];
+        for i in 0..20 {
+            messages.extend(create_tool_pair(
+                &format!("call{}", i),
+                &format!("resp{}", i),
+                "read_file",
+                "content",
+            ));
+        }
         let conversation = Conversation::new_unvalidated(messages);
 
-        let result = tool_id_to_summarize(&conversation, 2);
+        // No protection: 20 eligible, 20 > 12 → batch of 10
+        let result = tool_ids_to_summarize(&conversation, 2, 0);
+        assert_eq!(result.len(), TOOLCALL_SUMMARIZATION_BATCH_SIZE);
+
+        // Protect last 8: 12 eligible, 12 <= 12 → nothing
+        let result = tool_ids_to_summarize(&conversation, 2, 8);
         assert!(
-            result.is_some(),
-            "Should return a pair to summarize when tool calls exceed cutoff"
+            result.is_empty(),
+            "Should not summarize when protected count leaves eligible <= cutoff + batch"
         );
 
-        let tool_call_id = result.unwrap();
-        assert_eq!(tool_call_id, "call1");
-
-        let summary = summarize_tool_call(&provider, "test-session", &conversation, &tool_call_id)
-            .await
-            .unwrap();
-
-        assert_eq!(summary.role, Role::User);
-        assert!(summary.metadata.agent_visible);
-        assert!(!summary.metadata.user_visible);
-
-        let mut updated_messages = conversation.messages().clone();
-        for msg in updated_messages.iter_mut() {
-            let has_matching_content = msg.content.iter().any(|c| match c {
-                MessageContent::ToolRequest(req) => req.id == tool_call_id,
-                MessageContent::ToolResponse(resp) => resp.id == tool_call_id,
-                _ => false,
-            });
-
-            if has_matching_content {
-                msg.metadata = msg.metadata.with_agent_invisible();
-            }
-        }
-
-        updated_messages.push(summary);
-
-        let updated_conversation = Conversation::new_unvalidated(updated_messages);
-        let messages = updated_conversation.messages();
-
-        let call1_msg = messages
-            .iter()
-            .find(|m| m.id.as_deref() == Some("call1"))
-            .unwrap();
-        assert!(
-            !call1_msg.is_agent_visible(),
-            "Original call should not be agent visible"
-        );
-
-        let response1_msg = messages
-            .iter()
-            .find(|m| m.id.as_deref() == Some("response1"))
-            .unwrap();
-        assert!(
-            !response1_msg.is_agent_visible(),
-            "Original response should not be agent visible"
-        );
-
-        let summary_msg = messages
-            .iter()
-            .find(|m| {
-                m.metadata.agent_visible
-                    && !m.metadata.user_visible
-                    && m.as_concat_text().contains("Tool call")
-            })
-            .unwrap();
-        assert!(
-            !summary_msg.is_user_visible(),
-            "Summary should not be user visible"
-        );
-
-        let result = tool_id_to_summarize(&updated_conversation, 3);
-        assert!(result.is_none(), "Nothing left to summarize");
+        // Protect last 7: 13 eligible, 13 > 12 → batch of 10
+        let result = tool_ids_to_summarize(&conversation, 2, 7);
+        assert_eq!(result.len(), TOOLCALL_SUMMARIZATION_BATCH_SIZE);
+        assert_eq!(result[0], "call0");
     }
 }

crates/goose/src/model.rs

@@ -5,7 +5,7 @@ use std::collections::HashMap;
 use thiserror::Error;
 use utoipa::ToSchema;
 
-const DEFAULT_CONTEXT_LIMIT: usize = 128_000;
+pub const DEFAULT_CONTEXT_LIMIT: usize = 128_000;
 
 #[derive(Debug, Clone, Deserialize)]
 struct PredefinedModel {

crates/goose/src/providers/base.rs

@@ -622,6 +622,14 @@ pub trait Provider: Send + Sync {
         false
     }
 
+    /// Whether the provider manages its own conversation context (e.g. CLI
+    /// wrappers like Claude Code or Gemini CLI). When true, goose-side
+    /// context management such as tool-pair summarization is skipped because
+    /// the provider's internal state is the source of truth.
+    fn manages_own_context(&self) -> bool {
+        false
+    }
+
     async fn supports_cache_control(&self) -> bool {
         false
     }

crates/goose/src/providers/claude_code.rs

@@ -637,6 +637,10 @@ impl Provider for ClaudeCodeProvider {
         &self.name
     }
 
+    fn manages_own_context(&self) -> bool {
+        true
+    }
+
     fn get_model_config(&self) -> ModelConfig {
         self.model.clone()
     }

crates/goose/src/providers/gemini_cli.rs

@@ -185,6 +185,10 @@ impl Provider for GeminiCliProvider {
         &self.name
     }
 
+    fn manages_own_context(&self) -> bool {
+        true
+    }
+
     fn get_model_config(&self) -> ModelConfig {
         self.model.clone()
     }

crates/goose/tests/agent.rs

@@ -495,6 +495,234 @@ mod tests {
         }
     }
 
+    #[cfg(test)]
+    mod tool_pair_summarization_tests {
+        use super::*;
+        use async_trait::async_trait;
+        use goose::agents::SessionConfig;
+        use goose::config::base::Config;
+        use goose::config::GooseMode;
+        use goose::conversation::message::Message;
+        use goose::model::ModelConfig;
+        use goose::providers::base::{
+            stream_from_single_message, MessageStream, Provider, ProviderDef, ProviderMetadata,
+            ProviderUsage, Usage,
+        };
+        use goose::providers::errors::ProviderError;
+        use goose::session::session_manager::SessionType;
+        use rmcp::model::{AnnotateAble, CallToolRequestParams, CallToolResult, RawContent, Tool};
+        use std::path::PathBuf;
+        use std::sync::atomic::{AtomicUsize, Ordering};
+        use std::sync::Arc;
+
+        /// Mock provider that returns text for the main reply and summaries for
+        /// summarization calls. Distinguishes by checking if tools are empty
+        /// (summarization calls pass no tools).
+        struct SummarizationTestProvider {
+            summary_count: AtomicUsize,
+        }
+
+        impl SummarizationTestProvider {
+            fn new() -> Self {
+                Self {
+                    summary_count: AtomicUsize::new(0),
+                }
+            }
+        }
+
+        impl ProviderDef for SummarizationTestProvider {
+            type Provider = Self;
+
+            fn metadata() -> ProviderMetadata {
+                ProviderMetadata {
+                    name: "mock-summarization".to_string(),
+                    display_name: "Mock Summarization Provider".to_string(),
+                    description: "Mock provider for summarization tests".to_string(),
+                    default_model: "mock-model".to_string(),
+                    known_models: vec![],
+                    model_doc_link: "".to_string(),
+                    config_keys: vec![],
+                    setup_steps: vec![],
+                }
+            }
+
+            fn from_env(
+                _model: ModelConfig,
+                _extensions: Vec<goose::config::ExtensionConfig>,
+            ) -> futures::future::BoxFuture<'static, anyhow::Result<Self>> {
+                Box::pin(async { Ok(Self::new()) })
+            }
+        }
+
+        #[async_trait]
+        impl Provider for SummarizationTestProvider {
+            async fn stream(
+                &self,
+                _model_config: &ModelConfig,
+                _session_id: &str,
+                _system_prompt: &str,
+                _messages: &[Message],
+                tools: &[Tool],
+            ) -> Result<MessageStream, ProviderError> {
+                let message = if tools.is_empty() {
+                    // Summarization call — return a unique summary
+                    let n = self.summary_count.fetch_add(1, Ordering::SeqCst);
+                    Message::assistant().with_text(format!("Summary of tool call #{}", n))
+                } else {
+                    // Main agent reply — return plain text so the loop exits
+                    Message::assistant().with_text("Done processing.")
+                };
+
+                let usage = ProviderUsage::new(
+                    "mock-model".to_string(),
+                    Usage::new(Some(10), Some(5), Some(15)),
+                );
+                Ok(stream_from_single_message(message, usage))
+            }
+
+            fn get_model_config(&self) -> ModelConfig {
+                ModelConfig::new("mock-model").unwrap()
+            }
+
+            fn get_name(&self) -> &str {
+                "mock-summarization"
+            }
+        }
+
+        /// Test that batch tool pair summarization preserves all summaries.
+        ///
+        /// Pre-populates a session with enough tool call/response pairs to trigger
+        /// batch summarization, runs agent.reply(), then verifies:
+        /// - All 10 summaries are present in the final conversation
+        /// - The original tool pairs are marked invisible
+        #[tokio::test]
+        async fn test_batch_summarization_preserves_all_summaries() -> Result<()> {
+            // Set a low cutoff so we don't need hundreds of tool pairs.
+            // cutoff=2 means we need >2+10=12 visible tool pairs to trigger.
+            Config::global()
+                .set_param("GOOSE_TOOL_CALL_CUTOFF", 2)
+                .unwrap();
+
+            let agent = Agent::new();
+            let session_manager = agent.config.session_manager.clone();
+            let provider = Arc::new(SummarizationTestProvider::new());
+
+            let session = session_manager
+                .create_session(
+                    PathBuf::from("."),
+                    "summarization-test".to_string(),
+                    SessionType::Hidden,
+                    GooseMode::Auto,
+                )
+                .await?;
+
+            agent.update_provider(provider, &session.id).await?;
+
+            // Pre-populate: start with a user message, then 13 tool call/response pairs
+            // (need > cutoff + 10 = 12 to trigger batch summarization)
+            let initial_msg = Message::user().with_text("help me read some files");
+            session_manager
+                .add_message(&session.id, &initial_msg)
+                .await?;
+
+            for i in 0..13 {
+                let call_id = format!("precall_{}", i);
+                let req_msg = Message::assistant()
+                    .with_tool_request(&call_id, Ok(CallToolRequestParams::new("read_file")))
+                    .with_generated_id();
+                session_manager.add_message(&session.id, &req_msg).await?;
+
+                let resp_msg = Message::user()
+                    .with_tool_response(
+                        &call_id,
+                        Ok(CallToolResult::success(vec![RawContent::text(format!(
+                            "content of file {}",
+                            i
+                        ))
+                        .no_annotation()])),
+                    )
+                    .with_generated_id();
+                session_manager.add_message(&session.id, &resp_msg).await?;
+            }
+
+            // Send a user message to trigger the reply loop
+            let user_message = Message::user().with_text("summarize what you found");
+
+            let session_config = SessionConfig {
+                id: session.id.clone(),
+                schedule_id: None,
+                max_turns: Some(1),
+                retry_config: None,
+            };
+
+            let reply_stream = agent.reply(user_message, session_config, None).await?;
+            tokio::pin!(reply_stream);
+
+            // Drain the stream
+            while let Some(event) = reply_stream.next().await {
+                match event {
+                    Ok(AgentEvent::Message(_)) => {}
+                    Ok(_) => {}
+                    Err(e) => return Err(e),
+                }
+            }
+
+            // Load the final session and inspect the conversation
+            let final_session = session_manager.get_session(&session.id, true).await?;
+            let conversation = final_session
+                .conversation
+                .expect("Session should have a conversation");
+            let messages = conversation.messages();
+
+            // Count summaries: messages that are agent-visible, not user-visible,
+            // and contain our summary text pattern
+            let summaries: Vec<&Message> = messages
+                .iter()
+                .filter(|m| {
+                    m.metadata.agent_visible
+                        && !m.metadata.user_visible
+                        && m.as_concat_text().starts_with("Summary of tool call #")
+                })
+                .collect();
+
+            assert_eq!(
+                summaries.len(),
+                10,
+                "Expected 10 summaries (one full batch), got {}. Summary texts: {:?}",
+                summaries.len(),
+                summaries
+                    .iter()
+                    .map(|m| m.as_concat_text())
+                    .collect::<Vec<_>>()
+            );
+
+            // Verify each summary is unique
+            let summary_texts: std::collections::HashSet<String> =
+                summaries.iter().map(|m| m.as_concat_text()).collect();
+            assert_eq!(summary_texts.len(), 10, "All 10 summaries should be unique");
+
+            // Count invisible tool pairs: original pairs that were summarized
+            // should have agent_visible=false
+            let invisible_tool_msgs: Vec<&Message> = messages
+                .iter()
+                .filter(|m| !m.metadata.agent_visible && (m.is_tool_call() || m.is_tool_response()))
+                .collect();
+
+            // Each summarized pair = 2 invisible messages (request + response)
+            assert_eq!(
+                invisible_tool_msgs.len(),
+                20, // 10 pairs × 2 messages
+                "Expected 20 invisible tool messages (10 summarized pairs), got {}",
+                invisible_tool_msgs.len()
+            );
+
+            // Clean up the config override
+            Config::global().delete("GOOSE_TOOL_CALL_CUTOFF").unwrap();
+
+            Ok(())
+        }
+    }
+
     #[cfg(test)]
     mod extension_manager_tests {
         use super::*;