@tiens.nguyen/gonext-local-worker 1.0.301 → 1.0.302

package/gonext-repl.mjs

@@ -1383,11 +1383,15 @@ async function runAgentTurn(history) {
     // > the "almost done" heartbeat > the bare fallback.
     const max = Math.max(24, (process.stdout.columns || 80) - 14);
     const fit = (s) => (s.length > max ? s.slice(0, max - 1) + "…" : s);
+    // Task #104: global (user + coding-URL) OUTPUT-token total, ↓ = generated. Add THIS
+    // turn's in-progress output (latestOutputTokens) to the persisted global so the live
+    // count climbs in real time and matches the web; it's reset to 0 once the turn's total
+    // is POSTed into globalOutputTokens (so it's never double-counted).
+    const liveOutputTotal = globalOutputTokens + latestOutputTokens;
     const tokLabel =
       dim(` · ${fmtTokens(latestCodeTokens)} tok`) +
       orange(` · ${fmtTokens(sessionMaxCodeTokens)} max`) +
-      // Task #104: global (user + coding-URL) lifetime OUTPUT-token total, ↓ = generated.
-      (globalOutputTokens > 0 ? cyan(` · ↓ ${fmtTokens(globalOutputTokens)}`) : "");
+      (liveOutputTotal > 0 ? cyan(` · ↓ ${fmtTokens(liveOutputTotal)}`) : "");
     // Task #96: the thought line is "clickable" only when the fuller thought has MORE text than
     // fits on line 1 (a live command / bare "Thinking" / a thought that already fits has nothing
     // to reveal). Line 1 (what's happening now): a live command > the model's current Thought >
@@ -1765,6 +1769,9 @@ async function runAgentTurn(history) {
           outputTokensPosted = true;
           void addOutputTokens(resolve(process.cwd()), latestOutputTokens).then((t) => {
             if (t && Number.isFinite(t.globalTotal)) globalOutputTokens = t.globalTotal;
+            // This turn's total is now folded into globalOutputTokens — zero the live
+            // per-turn tally so the ↓ (global + latest) doesn't double-count it.
+            latestOutputTokens = 0;
           });
         }
         // The job can flip to "completed" between polls with authoritative resultText that
@@ -1802,7 +1809,14 @@ async function runAgentTurn(history) {
           }
         }
         clearStatus();
-        return { text: answerFrom(text), shown: answerShownLive };
+        // Task #104: expose this turn's code-model INPUT + OUTPUT token totals so the caller
+        // can print a persistent footer at the bottom of the screen.
+        return {
+          text: answerFrom(text),
+          shown: answerShownLive,
+          inputTokens: latestCodeTokens,
+          outputTokens: latestOutputTokens,
+        };
       }
       if (text.length > shownChars) {
         if (seenRaw.length < 4096) {
@@ -2112,7 +2126,8 @@ async function main() {
     }
     history.push({ role: "user", content: taskLine });
     try {
-      const { text: answer, shown, cancelled } = await runAgentTurn(history);
+      const { text: answer, shown, cancelled, inputTokens, outputTokens } =
+        await runAgentTurn(history);
       if (answer) {
         history.push({ role: "assistant", content: answer });
         // Persist after every successful turn (not just on clean exit) — an ungraceful
@@ -2122,6 +2137,16 @@ async function main() {
         // A plain-reply answer already streamed live in the loop above — printing it
         // again here would just duplicate it; a blank line is enough for spacing.
         console.log(shown ? "" : `\n\n${answer}\n`);
+        // Task #104: persistent per-turn token footer at the bottom of the screen — the
+        // last turn's code-model INPUT + OUTPUT tokens, kept visible above the next prompt.
+        // Only for turns that actually used the code model (plain-reply greetings = 0/0).
+        if ((inputTokens ?? 0) > 0 || (outputTokens ?? 0) > 0) {
+          console.log(
+            dim("  tokens ·  in ") +
+              fmtTokens(inputTokens ?? 0) +
+              cyan(`  ·  ↓ out ${fmtTokens(outputTokens ?? 0)}`)
+          );
+        }
       } else {
         history.pop(); // aborted/failed/cancelled turn — don't poison the next request
         // A turn that ended WITHOUT an answer (Ctrl+C cancel, or an abort/failure) must not

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tiens.nguyen/gonext-local-worker",
-  "version": "1.0.301",
+  "version": "1.0.302",
   "description": "Polls GoNext cloud API for async local LLM jobs and runs them against Ollama/OpenAI-compatible servers on this Mac",
   "type": "module",
   "license": "MIT",