Include EditAgent's raw output when inspecting thread (#30337)

This allows us to debug the raw edits that were generated when people
report feedback, when running evals and when opening the thread as
Markdown.

Release Notes:

- Improved debug output for agent threads.

crates/agent/src/thread.rs

@@ -2487,6 +2487,13 @@ impl Thread {
 
                 writeln!(markdown, "**\n")?;
                 writeln!(markdown, "{}", tool_result.content)?;
+                if let Some(output) = tool_result.output.as_ref() {
+                    writeln!(
+                        markdown,
+                        "\n\nDebug Output:\n\n```json\n{}\n```\n",
+                        serde_json::to_string_pretty(output)?
+                    )?;
+                }
             }
         }
 

crates/assistant_tools/src/edit_agent.rs

@@ -20,7 +20,8 @@ use language_model::{
     LanguageModelToolChoice, MessageContent, Role,
 };
 use project::{AgentLocation, Project};
-use serde::Serialize;
+use schemars::JsonSchema;
+use serde::{Deserialize, Serialize};
 use std::{cmp, iter, mem, ops::Range, path::PathBuf, sync::Arc, task::Poll};
 use streaming_diff::{CharOperation, StreamingDiff};
 
@@ -50,10 +51,10 @@ pub enum EditAgentOutputEvent {
     OldTextNotFound(SharedString),
 }
 
-#[derive(Clone, Debug)]
+#[derive(Clone, Debug, Serialize, Deserialize, JsonSchema)]
 pub struct EditAgentOutput {
-    pub _raw_edits: String,
+    pub raw_edits: String,
-    pub _parser_metrics: EditParserMetrics,
+    pub parser_metrics: EditParserMetrics,
 }
 
 #[derive(Clone)]
@@ -186,8 +187,8 @@ impl EditAgent {
         }
 
         Ok(EditAgentOutput {
-            _raw_edits: raw_edits,
+            raw_edits,
-            _parser_metrics: EditParserMetrics::default(),
+            parser_metrics: EditParserMetrics::default(),
         })
     }
 
@@ -426,8 +427,8 @@ impl EditAgent {
                 }
             }
             Ok(EditAgentOutput {
-                _raw_edits: raw_edits,
+                raw_edits,
-                _parser_metrics: parser.finish(),
+                parser_metrics: parser.finish(),
             })
         });
         (output, rx)

crates/assistant_tools/src/edit_agent/edit_parser.rs

@@ -1,4 +1,6 @@
 use derive_more::{Add, AddAssign};
+use schemars::JsonSchema;
+use serde::{Deserialize, Serialize};
 use smallvec::SmallVec;
 use std::{cmp, mem, ops::Range};
 
@@ -13,7 +15,9 @@ pub enum EditParserEvent {
     NewTextChunk { chunk: String, done: bool },
 }
 
-#[derive(Clone, Debug, Default, PartialEq, Eq, Add, AddAssign)]
+#[derive(
+    Clone, Debug, Default, PartialEq, Eq, Add, AddAssign, Serialize, Deserialize, JsonSchema,
+)]
 pub struct EditParserMetrics {
     pub tags: usize,
     pub mismatched_tags: usize,

crates/assistant_tools/src/edit_agent/evals.rs

@@ -1116,7 +1116,7 @@ fn eval(iterations: usize, expected_pass_ratio: f32, mut eval: EvalInput) {
     while let Ok(output) = rx.recv() {
         match output {
             Ok(output) => {
-                cumulative_parser_metrics += output.sample.edit_output._parser_metrics.clone();
+                cumulative_parser_metrics += output.sample.edit_output.parser_metrics.clone();
                 eval_outputs.push(output.clone());
                 if output.assertion.score < 80 {
                     failed_count += 1;
@@ -1197,9 +1197,9 @@ impl Display for EvalOutput {
         writeln!(
             f,
             "Parser Metrics:\n{:#?}",
-            self.sample.edit_output._parser_metrics
+            self.sample.edit_output.parser_metrics
         )?;
-        writeln!(f, "Raw Edits:\n{}", self.sample.edit_output._raw_edits)?;
+        writeln!(f, "Raw Edits:\n{}", self.sample.edit_output.raw_edits)?;
         Ok(())
     }
 }

crates/assistant_tools/src/edit_file_tool.rs

@@ -1,6 +1,6 @@
 use crate::{
     Templates,
-    edit_agent::{EditAgent, EditAgentOutputEvent},
+    edit_agent::{EditAgent, EditAgentOutput, EditAgentOutputEvent},
     schema::json_schema_for,
 };
 use anyhow::{Result, anyhow};
@@ -88,6 +88,7 @@ pub struct EditFileToolOutput {
     pub original_path: PathBuf,
     pub new_text: String,
     pub old_text: String,
+    pub raw_output: Option<EditAgentOutput>,
 }
 
 #[derive(Debug, Serialize, Deserialize, JsonSchema)]
@@ -248,7 +249,7 @@ impl Tool for EditFileTool {
                     EditAgentOutputEvent::OldTextNotFound(_) => hallucinated_old_text = true,
                 }
             }
-            output.await?;
+            let agent_output = output.await?;
 
             project
                 .update(cx, |project, cx| project.save_buffer(buffer.clone(), cx))?
@@ -268,6 +269,7 @@ impl Tool for EditFileTool {
                 original_path: project_path.path.to_path_buf(),
                 new_text: new_text.clone(),
                 old_text: old_text.clone(),
+                raw_output: Some(agent_output),
             };
 
             if let Some(card) = card_clone {

crates/eval/src/instance.rs

@@ -965,6 +965,15 @@ impl RequestMarkdown {
                             messages.push_str("**ERROR:**\n");
                         }
                         messages.push_str(&format!("{}\n\n", tool_result.content));
+
+                        if let Some(output) = tool_result.output.as_ref() {
+                            writeln!(
+                                messages,
+                                "**Debug Output**:\n\n```json\n{}\n```\n",
+                                serde_json::to_string_pretty(output).unwrap()
+                            )
+                            .unwrap();
+                        }
                     }
                 }
             }