📈 Add token usage reporting from LLM responses

Extract detailed usage statistics from PromptResponse to log input,
output, and total token counts during agent execution. This enables
better visibility into API consumption for cost monitoring and
performance analysis.

Uses extended_details() to get the full response including usage
metrics rather than just the output text.

Author: hyperb1iss

src/agents/iris.rs

@@ -4,7 +4,7 @@
 //! and multi-turn execution using Rig. One agent to rule them all! ✨
 
 use anyhow::Result;
-use rig::agent::{Agent, AgentBuilder as RigAgentBuilder};
+use rig::agent::{Agent, AgentBuilder as RigAgentBuilder, PromptResponse};
 use rig::client::builder::DynClientBuilder;
 use rig::completion::{CompletionModel, Prompt};
 use schemars::JsonSchema;
@@ -551,11 +551,28 @@ Guidelines:
             "LLM request",
             "Sending prompt to agent with multi_turn(50)",
         );
-        let response = agent.prompt(&full_prompt).multi_turn(50).await?;
+        let prompt_response: PromptResponse = agent
+            .prompt(&full_prompt)
+            .multi_turn(50)
+            .extended_details()
+            .await?;
 
         timer.finish();
 
-        debug::debug_llm_response(&response, std::time::Duration::from_secs(0), None);
+        // Extract usage stats for debug output
+        let usage = &prompt_response.total_usage;
+        debug::debug_context_management(
+            "Token usage",
+            &format!(
+                "input: {} | output: {} | total: {}",
+                usage.input_tokens, usage.output_tokens, usage.total_tokens
+            ),
+        );
+
+        let response = &prompt_response.output;
+        #[allow(clippy::cast_possible_truncation, clippy::as_conversions)]
+        let total_tokens_usize = usage.total_tokens as usize;
+        debug::debug_llm_response(response, std::time::Duration::from_secs(0), Some(total_tokens_usize));
 
         // Update status - synthesis phase
         crate::iris_status_dynamic!(
@@ -566,7 +583,7 @@ Guidelines:
         );
 
         // Extract and parse JSON from the response
-        let json_str = extract_json_from_response(&response)?;
+        let json_str = extract_json_from_response(response)?;
         let sanitized_json = sanitize_json_response(&json_str);
         let sanitized_ref = sanitized_json.as_ref();