refactor(agent): use hive_claude::TokenUsage directly, drop the duplicate

hive-ag3nt/src/events.rs

@@ -12,6 +12,7 @@ use std::path::{Path, PathBuf};
 use std::sync::atomic::{AtomicBool, AtomicI64, AtomicU64, Ordering};
 use std::sync::{Arc, Mutex};
 
+use hive_claude::TokenUsage;
 use rusqlite::{Connection, params};
 use serde::{Deserialize, Serialize};
 use tokio::sync::broadcast;
@@ -432,28 +433,6 @@ impl EventStore {
     }
 }
 
-/// Token usage emitted by claude in the final `result` stream-json event.
-/// All counts are in tokens. `None` fields mean the server didn't report them.
-#[derive(Debug, Clone, Copy, Default, Serialize, Deserialize, PartialEq, Eq)]
-pub struct TokenUsage {
-    pub input_tokens: u64,
-    pub output_tokens: u64,
-    pub cache_read_input_tokens: u64,
-    pub cache_creation_input_tokens: u64,
-}
-
-impl TokenUsage {
-    /// Total context consumed this turn (input + cache reads + cache writes).
-    /// This is the per-inference context footprint that counts against the
-    /// model's `contextWindow` limit. Tracked from the last `assistant` event
-    /// in the stream-json (per-inference usage, not the cumulative `result`
-    /// event which sums across all inferences in a tool-heavy turn and can
-    /// far exceed the per-inference window).
-    #[must_use]
-    pub fn context_tokens(&self) -> u64 {
-        self.input_tokens + self.cache_read_input_tokens + self.cache_creation_input_tokens
-    }
-}
 
 /// Authoritative turn-loop state. The harness owns it; the web UI
 /// reads via `/api/state` and renders. Lives alongside the bus