feat: add additional goosebench evals (#1571)

Co-authored-by: Alice Hau <alice.a.hau@gmail.com>

crates/goose-cli/src/commands/bench.rs

@@ -56,6 +56,20 @@ impl BenchAgent for BenchSession {
         let errors = self.errors.lock().await;
         errors.clone()
     }
+
+    async fn get_token_usage(&self) -> Option<i32> {
+        // Get token usage from the provider
+        if let Ok(usage) = self.session.get_usage().await {
+            // Sum up total tokens across all usage entries
+            let total_tokens = usage
+                .iter()
+                .map(|u| u.usage.total_tokens.unwrap_or(0))
+                .sum();
+            Some(total_tokens)
+        } else {
+            None
+        }
+    }
 }
 
 // Wrapper struct to implement BenchAgent for Arc<Mutex<BenchSession>>
@@ -72,6 +86,11 @@ impl BenchAgent for BenchAgentWrapper {
         let session = self.0.lock().await;
         session.get_errors().await
     }
+
+    async fn get_token_usage(&self) -> Option<i32> {
+        let session = self.0.lock().await;
+        session.get_token_usage().await
+    }
 }
 
 async fn run_eval(

crates/goose-cli/src/session/mod.rs

@@ -16,6 +16,7 @@ use goose::agents::extension::{Envs, ExtensionConfig};
 use goose::agents::{Agent, SessionConfig};
 use goose::config::Config;
 use goose::message::{Message, MessageContent};
+use goose::providers::base::ProviderUsage;
 use goose::session;
 use mcp_core::handler::ToolError;
 use mcp_core::prompt::PromptMessage;
@@ -643,4 +644,9 @@ impl Session {
     pub fn message_history(&self) -> Vec<Message> {
         self.messages.clone()
     }
+
+    /// Get the token usage from the agent
+    pub async fn get_usage(&self) -> Result<Vec<ProviderUsage>> {
+        Ok(self.agent.usage().await)
+    }
 }