@tiens.nguyen/gonext-local-worker 1.0.300 → 1.0.302
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/gonext-repl.mjs +29 -4
- package/package.json +1 -1
package/gonext-repl.mjs
CHANGED
|
@@ -1383,11 +1383,15 @@ async function runAgentTurn(history) {
|
|
|
1383
1383
|
// > the "almost done" heartbeat > the bare fallback.
|
|
1384
1384
|
const max = Math.max(24, (process.stdout.columns || 80) - 14);
|
|
1385
1385
|
const fit = (s) => (s.length > max ? s.slice(0, max - 1) + "…" : s);
|
|
1386
|
+
// Task #104: global (user + coding-URL) OUTPUT-token total, ↓ = generated. Add THIS
|
|
1387
|
+
// turn's in-progress output (latestOutputTokens) to the persisted global so the live
|
|
1388
|
+
// count climbs in real time and matches the web; it's reset to 0 once the turn's total
|
|
1389
|
+
// is POSTed into globalOutputTokens (so it's never double-counted).
|
|
1390
|
+
const liveOutputTotal = globalOutputTokens + latestOutputTokens;
|
|
1386
1391
|
const tokLabel =
|
|
1387
1392
|
dim(` · ${fmtTokens(latestCodeTokens)} tok`) +
|
|
1388
1393
|
orange(` · ${fmtTokens(sessionMaxCodeTokens)} max`) +
|
|
1389
|
-
|
|
1390
|
-
(globalOutputTokens > 0 ? cyan(` · ↓ ${fmtTokens(globalOutputTokens)}`) : "");
|
|
1394
|
+
(liveOutputTotal > 0 ? cyan(` · ↓ ${fmtTokens(liveOutputTotal)}`) : "");
|
|
1391
1395
|
// Task #96: the thought line is "clickable" only when the fuller thought has MORE text than
|
|
1392
1396
|
// fits on line 1 (a live command / bare "Thinking" / a thought that already fits has nothing
|
|
1393
1397
|
// to reveal). Line 1 (what's happening now): a live command > the model's current Thought >
|
|
@@ -1765,6 +1769,9 @@ async function runAgentTurn(history) {
|
|
|
1765
1769
|
outputTokensPosted = true;
|
|
1766
1770
|
void addOutputTokens(resolve(process.cwd()), latestOutputTokens).then((t) => {
|
|
1767
1771
|
if (t && Number.isFinite(t.globalTotal)) globalOutputTokens = t.globalTotal;
|
|
1772
|
+
// This turn's total is now folded into globalOutputTokens — zero the live
|
|
1773
|
+
// per-turn tally so the ↓ (global + latest) doesn't double-count it.
|
|
1774
|
+
latestOutputTokens = 0;
|
|
1768
1775
|
});
|
|
1769
1776
|
}
|
|
1770
1777
|
// The job can flip to "completed" between polls with authoritative resultText that
|
|
@@ -1802,7 +1809,14 @@ async function runAgentTurn(history) {
|
|
|
1802
1809
|
}
|
|
1803
1810
|
}
|
|
1804
1811
|
clearStatus();
|
|
1805
|
-
|
|
1812
|
+
// Task #104: expose this turn's code-model INPUT + OUTPUT token totals so the caller
|
|
1813
|
+
// can print a persistent footer at the bottom of the screen.
|
|
1814
|
+
return {
|
|
1815
|
+
text: answerFrom(text),
|
|
1816
|
+
shown: answerShownLive,
|
|
1817
|
+
inputTokens: latestCodeTokens,
|
|
1818
|
+
outputTokens: latestOutputTokens,
|
|
1819
|
+
};
|
|
1806
1820
|
}
|
|
1807
1821
|
if (text.length > shownChars) {
|
|
1808
1822
|
if (seenRaw.length < 4096) {
|
|
@@ -2112,7 +2126,8 @@ async function main() {
|
|
|
2112
2126
|
}
|
|
2113
2127
|
history.push({ role: "user", content: taskLine });
|
|
2114
2128
|
try {
|
|
2115
|
-
const { text: answer, shown, cancelled } =
|
|
2129
|
+
const { text: answer, shown, cancelled, inputTokens, outputTokens } =
|
|
2130
|
+
await runAgentTurn(history);
|
|
2116
2131
|
if (answer) {
|
|
2117
2132
|
history.push({ role: "assistant", content: answer });
|
|
2118
2133
|
// Persist after every successful turn (not just on clean exit) — an ungraceful
|
|
@@ -2122,6 +2137,16 @@ async function main() {
|
|
|
2122
2137
|
// A plain-reply answer already streamed live in the loop above — printing it
|
|
2123
2138
|
// again here would just duplicate it; a blank line is enough for spacing.
|
|
2124
2139
|
console.log(shown ? "" : `\n\n${answer}\n`);
|
|
2140
|
+
// Task #104: persistent per-turn token footer at the bottom of the screen — the
|
|
2141
|
+
// last turn's code-model INPUT + OUTPUT tokens, kept visible above the next prompt.
|
|
2142
|
+
// Only for turns that actually used the code model (plain-reply greetings = 0/0).
|
|
2143
|
+
if ((inputTokens ?? 0) > 0 || (outputTokens ?? 0) > 0) {
|
|
2144
|
+
console.log(
|
|
2145
|
+
dim(" tokens · in ") +
|
|
2146
|
+
fmtTokens(inputTokens ?? 0) +
|
|
2147
|
+
cyan(` · ↓ out ${fmtTokens(outputTokens ?? 0)}`)
|
|
2148
|
+
);
|
|
2149
|
+
}
|
|
2125
2150
|
} else {
|
|
2126
2151
|
history.pop(); // aborted/failed/cancelled turn — don't poison the next request
|
|
2127
2152
|
// A turn that ended WITHOUT an answer (Ctrl+C cancel, or an abort/failure) must not
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@tiens.nguyen/gonext-local-worker",
|
|
3
|
-
"version": "1.0.
|
|
3
|
+
"version": "1.0.302",
|
|
4
4
|
"description": "Polls GoNext cloud API for async local LLM jobs and runs them against Ollama/OpenAI-compatible servers on this Mac",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"license": "MIT",
|