eval: Port to agent2 (#40704)

Release Notes:

- N/A

crates/eval/src/examples/add_arg_to_trait_method.rs

@@ -27,14 +27,12 @@ impl Example for AddArgToTraitMethod {
 
     async fn conversation(&self, cx: &mut ExampleContext) -> Result<()> {
         const FILENAME: &str = "assistant_tool.rs";
-        cx.push_user_message(format!(
+        let _ = cx.prompt(format!(
             r#"
             Add a `window: Option<gpui::AnyWindowHandle>` argument to the `Tool::run` trait method in {FILENAME},
             and update all the implementations of the trait and call sites accordingly.
             "#
-        ));
-
-        let _ = cx.run_to_end().await?;
+        )).await?;
 
         // Adds ignored argument to all but `batch_tool`
 

crates/eval/src/examples/code_block_citations.rs

@@ -29,16 +29,19 @@ impl Example for CodeBlockCitations {
 
     async fn conversation(&self, cx: &mut ExampleContext) -> Result<()> {
         const FILENAME: &str = "assistant_tool.rs";
-        cx.push_user_message(format!(
-            r#"
-            Show me the method bodies of all the methods of the `Tool` trait in {FILENAME}.
-
-            Please show each method in a separate code snippet.
-            "#
-        ));
 
         // Verify that the messages all have the correct formatting.
-        let texts: Vec<String> = cx.run_to_end().await?.texts().collect();
+        let texts: Vec<String> = cx
+            .prompt(format!(
+                r#"
+                Show me the method bodies of all the methods of the `Tool` trait in {FILENAME}.
+
+                Please show each method in a separate code snippet.
+                "#
+            ))
+            .await?
+            .texts()
+            .collect();
         let closing_fence = format!("\n{FENCE}");
 
         for text in texts.iter() {

crates/eval/src/examples/comment_translation.rs

@@ -22,30 +22,26 @@ impl Example for CommentTranslation {
     }
 
     async fn conversation(&self, cx: &mut ExampleContext) -> Result<()> {
-        cx.push_user_message(r#"
-            Edit the following files and translate all their comments to italian, in this exact order:
+        let response = cx.prompt(
+            r#"
+                Edit the following files and translate all their comments to italian, in this exact order:
 
-            - font-kit/src/family.rs
-            - font-kit/src/canvas.rs
-            - font-kit/src/error.rs
-        "#);
-        cx.run_to_end().await?;
+                - font-kit/src/family.rs
+                - font-kit/src/canvas.rs
+                - font-kit/src/error.rs
+            "#
+        ).await?;
 
         let mut create_or_overwrite_count = 0;
-        cx.agent_thread().read_with(cx, |thread, cx| {
-            for message in thread.messages() {
-                for tool_use in thread.tool_uses_for_message(message.id, cx) {
-                    if tool_use.name == "edit_file" {
-                        let input: EditFileToolInput = serde_json::from_value(tool_use.input)?;
-                        if !matches!(input.mode, EditFileMode::Edit) {
-                            create_or_overwrite_count += 1;
-                        }
-                    }
+        for tool_call in response.tool_calls() {
+            if tool_call.name == "edit_file" {
+                let input = tool_call.parse_input::<EditFileToolInput>()?;
+                if !matches!(input.mode, EditFileMode::Edit) {
+                    create_or_overwrite_count += 1;
                 }
             }
+        }
 
-            anyhow::Ok(())
-        })??;
         cx.assert_eq(create_or_overwrite_count, 0, "no_creation_or_overwrite")?;
 
         Ok(())

crates/eval/src/examples/file_change_notification.rs

@@ -48,8 +48,8 @@ impl Example for FileChangeNotificationExample {
         })?;
 
         // Start conversation (specific message is not important)
-        cx.push_user_message("Find all files in this repo");
-        cx.run_turn().await?;
+        cx.prompt_with_max_turns("Find all files in this repo", 1)
+            .await?;
 
         // Edit the README buffer - the model should get a notification on next turn
         buffer.update(cx, |buffer, cx| {
@@ -58,7 +58,7 @@ impl Example for FileChangeNotificationExample {
 
         // Run for some more turns.
         // The model shouldn't thank us for letting it know about the file change.
-        cx.run_turns(3).await?;
+        cx.proceed_with_max_turns(3).await?;
 
         Ok(())
     }

crates/eval/src/examples/file_search.rs

@@ -25,18 +25,19 @@ impl Example for FileSearchExample {
 
     async fn conversation(&self, cx: &mut ExampleContext) -> Result<()> {
         const FILENAME: &str = "find_replace_file_tool.rs";
-        cx.push_user_message(format!(
-                r#"
+
+        let prompt = format!(
+            r#"
         Look at the `{FILENAME}`. I want to implement a card for it. The card should implement the `Render` trait.
 
         The card should show a diff. It should be a beautifully presented diff. The card "box" should look like what we show for
         markdown codeblocks (look at `MarkdownElement`). I want to see a red background for lines that were deleted and a green
         background for lines that were added. We should have a div per diff line.
         "#
-        ));
+        );
 
-        let response = cx.run_turn().await?;
-        let tool_use = response.expect_tool("find_path", cx)?;
+        let response = cx.prompt_with_max_turns(prompt, 1).await?;
+        let tool_use = response.expect_tool_call("find_path", cx)?;
         let input = tool_use.parse_input::<FindPathToolInput>()?;
 
         let glob = input.glob;

crates/eval/src/examples/grep_params_escapement.rs

@@ -1,3 +1,4 @@
+use agent::GrepToolInput;
 use agent_settings::AgentProfileId;
 use anyhow::Result;
 use async_trait::async_trait;
@@ -35,9 +36,9 @@ impl Example for GrepParamsEscapementExample {
     }
 
     async fn conversation(&self, cx: &mut ExampleContext) -> Result<()> {
-        // cx.push_user_message("How does the precedence/specificity work with Keymap contexts? I am seeing that `MessageEditor > Editor` is lower precendence than `Editor` which is surprising to me, but might be how it works");
-        cx.push_user_message("Search for files containing the characters `>` or `<`");
-        let response = cx.run_turns(2).await?;
+        let response = cx
+            .prompt_with_max_turns("Search for files containing the characters `>` or `<`", 2)
+            .await?;
         let grep_input = response
             .find_tool_call("grep")
             .and_then(|tool_use| tool_use.parse_input::<GrepToolInput>().ok());

crates/eval/src/examples/mod.rs

@@ -144,9 +144,8 @@ impl Example for DeclarativeExample {
     }
 
     async fn conversation(&self, cx: &mut ExampleContext) -> Result<()> {
-        cx.push_user_message(&self.prompt);
         let max_turns = self.metadata.max_turns.unwrap_or(1000);
-        let _ = cx.run_turns(max_turns).await;
+        let _ = cx.prompt_with_max_turns(&self.prompt, max_turns).await;
         Ok(())
     }
 

crates/eval/src/examples/overwrite_file.rs

@@ -1,3 +1,4 @@
+use agent::{EditFileMode, EditFileToolInput};
 use agent_settings::AgentProfileId;
 use anyhow::Result;
 use async_trait::async_trait;
@@ -35,17 +36,14 @@ impl Example for FileOverwriteExample {
     }
 
     async fn conversation(&self, cx: &mut ExampleContext) -> Result<()> {
-        let response = cx.run_turns(1).await?;
-        let file_overwritten = if let Some(tool_use) = response.find_tool_call("edit_file") {
-            let input = tool_use.parse_input::<EditFileToolInput>()?;
-            match input.mode {
-                EditFileMode::Edit => false,
-                EditFileMode::Create | EditFileMode::Overwrite => {
-                    input.path.ends_with("src/language_model_selector.rs")
-                }
+        let response = cx.proceed_with_max_turns(1).await?;
+        let tool_use = response.expect_tool_call("edit_file", cx)?;
+        let input = tool_use.parse_input::<EditFileToolInput>()?;
+        let file_overwritten = match input.mode {
+            EditFileMode::Edit => false,
+            EditFileMode::Create | EditFileMode::Overwrite => {
+                input.path.ends_with("src/language_model_selector.rs")
             }
-        } else {
-            false
         };
 
         cx.assert(!file_overwritten, "File should be edited, not overwritten")

crates/eval/src/examples/planets.rs

@@ -23,20 +23,19 @@ impl Example for Planets {
     }
 
     async fn conversation(&self, cx: &mut ExampleContext) -> Result<()> {
-        cx.push_user_message(
-            r#"
+        let response = cx
+            .prompt(
+                r#"
             Make a plain JavaScript web page which renders an animated 3D solar system.
             Let me drag to rotate the camera around.
             Do not use npm.
-            "#
-            .to_string(),
-        );
-
-        let response = cx.run_to_end().await?;
+            "#,
+            )
+            .await?;
         let mut open_tool_uses = 0;
         let mut terminal_tool_uses = 0;
 
-        for tool_use in response.tool_uses() {
+        for tool_use in response.tool_calls() {
             if tool_use.name == OpenTool::name() {
                 open_tool_uses += 1;
             } else if tool_use.name == TerminalTool::name() {