@ekkos/cli 1.5.0 → 1.5.1

package/dist/commands/dashboard.js

@@ -599,7 +599,8 @@ function buildDashboardData(params) {
         model: params.model,
         turnCount: turns.length,
         totalCost,
-        totalTokens: totalInput + totalCacheRead + totalCacheCreate + totalOutput,
+        // Unique content tokens: excludes cache_read (same prefix re-billed per turn)
+        totalTokens: totalInput + totalCacheCreate + totalOutput,
         totalInput,
         totalCacheRead,
         totalCacheCreate,
@@ -1915,7 +1916,12 @@ async function launchDashboard(initialSessionName, initialJsonlPath, refreshMs,
         if (!d)
             return;
         try {
-            const totalTokensM = ((d.totalInput + d.totalCacheRead + d.totalCacheCreate + d.totalOutput) / 1000000).toFixed(2);
+            // Show unique content tokens: output (model-generated) + cache_create (new content cached)
+            // + uncached input. Excludes cache_read (same prefix re-billed each turn — NOT unique content).
+            // Previously showed cumulative API total (input+read+create+output) which inflated by ~10-20x
+            // because the same prefix was counted on every tool round-trip.
+            const uniqueTokens = d.totalOutput + d.totalCacheCreate + d.totalInput;
+            const totalTokensM = (uniqueTokens / 1000000).toFixed(2);
             const totalSavings = d.turns.reduce((s, t) => s + t.savings, 0);
             // Model routing breakdown (uses routedModel for actual model counts)
             const opusCount = d.turns.filter(t => t.routedModel.includes('opus')).length;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ekkos/cli",
-  "version": "1.5.0",
+  "version": "1.5.1",
   "description": "ekkOS memory CLI — persistent memory for AI coding assistants (Claude Code, Gemini, Cursor, Windsurf)",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",