feat(goose-acp): enable parallel sessions with isolated agent state (#6392)

Signed-off-by: Adrian Cole <adrian@tetrate.io>

crates/goose-bench/Cargo.toml

@@ -11,21 +11,21 @@ description.workspace = true
 workspace = true
 
 [dependencies]
-anyhow = "1.0"
+anyhow = { workspace = true }
 paste = "1.0"
 ctor = "0.2.7"
 goose = { path = "../goose" }
 rmcp = { workspace = true }
 async-trait = "0.1.89"
 chrono = { version = "0.4", features = ["serde"] }
-serde_json = "1.0"
+serde_json = { workspace = true }
 serde = { version = "1.0", features = ["derive"] }
-tracing = "0.1"
+tracing = { workspace = true }
 tracing-subscriber = { version = "0.3", features = ["registry"] }
-tokio = { version = "1.43", features = ["full"] }
+tokio = { workspace = true }
 include_dir = "0.7.4"
 once_cell = "1.19"
-regex = "1.11.1"
+regex = { workspace = true }
 dotenvy = "0.15.7"
 
 [target.'cfg(target_os = "windows")'.dependencies]

crates/goose-bench/src/bench_session.rs

@@ -1,6 +1,7 @@
 use async_trait::async_trait;
 use chrono::{DateTime, Utc};
 use goose::conversation::Conversation;
+use goose::session::session_manager::Session;
 
 use serde::{Deserialize, Serialize};
 use std::sync::Arc;
@@ -19,7 +20,7 @@ pub trait BenchBaseSession: Send + Sync {
     async fn headless(&mut self, message: String) -> anyhow::Result<()>;
     fn message_history(&self) -> Conversation;
     fn get_total_token_usage(&self) -> anyhow::Result<Option<i32>>;
-    fn get_session_id(&self) -> anyhow::Result<String>;
+    async fn get_session(&self) -> anyhow::Result<Session>;
 }
 // struct for managing agent-session-access. to be passed to evals for benchmarking
 pub struct BenchAgent {
@@ -52,7 +53,7 @@ impl BenchAgent {
         self.session.get_total_token_usage().ok().flatten()
     }
 
-    pub(crate) fn get_session_id(&self) -> anyhow::Result<String> {
-        self.session.get_session_id()
+    pub(crate) async fn get_session(&self) -> anyhow::Result<Session> {
+        self.session.get_session().await
     }
 }

crates/goose-bench/src/runners/eval_runner.rs

@@ -5,7 +5,6 @@ use crate::eval_suites::{EvaluationSuite, ExtensionRequirements};
 use crate::reporting::EvaluationResult;
 use crate::utilities::await_process_exits;
 use anyhow::{bail, Context, Result};
-use goose::session::SessionManager;
 use std::env;
 use std::fs;
 use std::future::Future;
@@ -156,8 +155,7 @@ impl EvalRunner {
                 .canonicalize()
                 .context("Failed to canonicalize current directory path")?;
 
-            let session_id = agent.get_session_id()?.to_string();
-            let session = SessionManager::get_session(&session_id, true).await?;
+            let session = agent.get_session().await?;
 
             let session_json = serde_json::to_string_pretty(&session)
                 .context("Failed to serialize session to JSON")?;