prior-cli 1.4.0 → 1.4.2

package/bin/prior.js

@@ -408,7 +408,7 @@ function renderToolDone(name, summary, preview) {
     const more    = lines.length - toShow.length;
     if (toShow.length > 0) {
       drawBox([
-        ...toShow.map(l => ({ text: l.slice(0, 80), dim: true })),
+        ...toShow.map(l => ({ text: l.replace(/\r/g, '').slice(0, 80), dim: true })),
         ...(more > 0 ? [{ text: `… ${more} more line${more !== 1 ? 's' : ''}`, dim: true }] : []),
       ]);
     }
@@ -927,7 +927,7 @@ Keep it under 350 words. Write prior.md now.`;
                       renderToolStart(ev.name, ev.args);
                       spinStart('working…');
                       break;
-                    case 'tool_done':  spinStop(); renderToolDone(ev.name, ev.summary); break;
+                    case 'tool_done':  spinStop(); renderToolDone(ev.name, ev.summary, ev.preview); break;
                     case 'tool_error': spinStop(); renderToolError(ev.name, ev.error); break;
                     case 'text':
                       spinStop();
@@ -1023,10 +1023,9 @@ Keep it under 350 words. Write prior.md now.`;
           console.log(c.brand('  ◈') + c.dim(`  ${label} attached`));
         }
 
-        let responseText      = '';
-        let _progressStarted  = false;
-        let _streamStarted    = false; // true after first text_chunk
-        const _thinkStart     = Date.now();
+        let responseText     = '';
+        let _progressStarted = false;
+        const _thinkStart    = Date.now();
 
         spinStart('thinking…');
 
@@ -1059,42 +1058,13 @@ Keep it under 350 words. Write prior.md now.`;
                   spinStart('thinking…');
                   break;
 
-                case 'waiting':
-                  spinStart(`waiting for Ollama… (${ev.attempt}/${ev.max})`);
-                  break;
-
                 case 'cancelled':
                   spinStop();
-                  if (_streamStarted) process.stdout.write('\n');
                   console.log(c.muted('  ✗ Cancelled'));
                   break;
 
-                // ── Streaming text events ──────────────────────
-                case 'stream_start': {
-                  spinStop();
-                  const thinkTime = elapsed(Date.now() - _thinkStart);
-                  console.log(c.brand('  Prior  ') + c.muted(`·  ${timeNow()}  ·  ${thinkTime}`));
-                  console.log('');
-                  process.stdout.write('  ');
-                  _streamStarted = true;
-                  break;
-                }
-
-                case 'text_chunk':
-                  if (ev.content) {
-                    process.stdout.write(ev.content);
-                    responseText += ev.content;
-                  }
-                  break;
-
-                case 'stream_end':
-                  process.stdout.write('\n');
-                  _streamStarted = false;
-                  break;
-
                 case 'tool_start':
                   spinStop();
-                  if (_streamStarted) { process.stdout.write('\n'); _streamStarted = false; }
                   _progressStarted = false;
                   renderToolStart(ev.name, ev.args);
                   if (!CONFIRM_TOOLS.has(ev.name)) spinStart('working…');
@@ -1148,7 +1118,6 @@ Keep it under 350 words. Write prior.md now.`;
 
                 case 'error':
                   spinStop();
-                  if (_streamStarted) process.stdout.write('\n');
                   console.error(c.err(`  ✗ ${ev.message}`));
                   break;
               }

package/lib/agent.js

@@ -8,9 +8,9 @@ const CLI_BASE  = 'https://prior.ngrok.app/cli-backend';
 const PRIOR_BASE = 'https://prior.ngrok.app';
 const MAX_ITER   = 14;
 
-// ── Single inference call — reads NDJSON stream from backend ──
+// ── Single inference call ─────────────────────────────────────
 
-async function infer(messages, model, token, { cwd, uncensored, projectContext, images } = {}, signal, onChunk) {
+async function infer(messages, model, token, { cwd, uncensored, projectContext, images } = {}, signal) {
   const res = await fetch(`${CLI_BASE}/api/infer`, {
     method:  'POST',
     headers: { 'Content-Type': 'application/json' },
@@ -18,49 +18,11 @@ async function infer(messages, model, token, { cwd, uncensored, projectContext,
     timeout: 120000,
     signal,
   });
-
   if (!res.ok) {
     const err = await res.json().catch(() => ({}));
     throw new Error(err.error || `Server error: HTTP ${res.status}`);
   }
-
-  let content          = '';
-  let promptTokens     = 0;
-  let completionTokens = 0;
-
-  await new Promise((resolve, reject) => {
-    let buf = '';
-    res.body.on('data', rawChunk => {
-      if (signal?.aborted) {
-        res.body.destroy();
-        return reject(Object.assign(new Error('AbortError'), { name: 'AbortError' }));
-      }
-      buf += rawChunk.toString();
-      const lines = buf.split('\n');
-      buf = lines.pop();
-      for (const line of lines) {
-        if (!line.trim()) continue;
-        try {
-          const data = JSON.parse(line);
-          if (data.type === 'chunk') {
-            if (onChunk) onChunk(data.content);
-          } else if (data.type === 'done') {
-            content          = data.content || '';
-            promptTokens     = data.promptTokens     || 0;
-            completionTokens = data.completionTokens || 0;
-          } else if (data.type === 'waiting') {
-            if (onChunk) onChunk(null, { type: 'waiting', attempt: data.attempt, max: data.max });
-          } else if (data.type === 'error') {
-            reject(new Error(data.message));
-          }
-        } catch { /* skip malformed line */ }
-      }
-    });
-    res.body.on('end', resolve);
-    res.body.on('error', reject);
-  });
-
-  return { content, promptTokens, completionTokens };
+  return await res.json();
 }
 
 // ── Token usage tracking ──────────────────────────────────────
@@ -252,57 +214,12 @@ async function runAgent({ messages, model, uncensored, cwd, projectContext, imag
 
     send({ type: 'thinking' });
 
-    // ── Per-iteration streaming state ─────────────────────────
-    // After </think> we buffer LOOK_SIZE chars to detect tool calls before
-    // deciding whether to stream text live or stay in buffered mode.
-    let thinkBuf  = '';    // chunks buffered while inside <think>…</think>
-    let thinkDone = false;
-    let lookBuf   = '';    // first N chars of actual response (after think)
-    let streaming = false; // true once we've committed to live-streaming text
-    const LOOK_SIZE = 60;
-
-    function tryStartStreaming() {
-      if (streaming) return;
-      const trimmed = lookBuf.replace(/^[\s\n]+/, '');
-      // If the response starts with a tool tag, keep buffered (no live text)
-      if (/^<(?:tool|write|append|docx)[\s>{"[]/.test(trimmed)) return;
-      streaming = true;
-      send({ type: 'stream_start' });
-      if (trimmed) send({ type: 'text_chunk', content: trimmed });
-      lookBuf = '';
-    }
-
-    const onChunk = (raw, meta) => {
-      if (meta?.type === 'waiting') { send({ type: 'waiting', attempt: meta.attempt, max: meta.max }); return; }
-      if (!raw) return;
-
-      if (!thinkDone) {
-        thinkBuf += raw;
-        const idx = thinkBuf.indexOf('</think>');
-        if (idx !== -1) {
-          thinkDone = true;
-          lookBuf   = thinkBuf.slice(idx + 8).replace(/^[\s\n]+/, '');
-          thinkBuf  = '';
-          if (lookBuf.length >= LOOK_SIZE) tryStartStreaming();
-        }
-        return;
-      }
-
-      if (!streaming) {
-        lookBuf += raw;
-        if (lookBuf.length >= LOOK_SIZE) tryStartStreaming();
-        return;
-      }
-
-      send({ type: 'text_chunk', content: raw });
-    };
-
     const iterImages = pendingImages;
     pendingImages = null;
 
     let result;
     try {
-      result = await infer(history, model || 'qwen3.5:4b', token, { cwd, uncensored, projectContext, images: iterImages }, signal, onChunk);
+      result = await infer(history, model || 'qwen3.5:4b', token, { cwd, uncensored, projectContext, images: iterImages }, signal);
     } catch (err) {
       await trackTokenUsage(token, totalPromptTokens, totalCompletionTokens);
       if (err.name === 'AbortError' || signal?.aborted) { send({ type: 'cancelled' }); send({ type: 'done' }); return; }
@@ -311,10 +228,6 @@ async function runAgent({ messages, model, uncensored, cwd, projectContext, imag
       return;
     }
 
-    // Flush look-ahead if stream ended before LOOK_SIZE was reached
-    if (thinkDone && !streaming && lookBuf) tryStartStreaming();
-    if (streaming) send({ type: 'stream_end' });
-
     totalPromptTokens     += result.promptTokens     || 0;
     totalCompletionTokens += result.completionTokens || 0;
 
@@ -337,16 +250,14 @@ async function runAgent({ messages, model, uncensored, cwd, projectContext, imag
         continue;
       }
       await trackTokenUsage(token, totalPromptTokens, totalCompletionTokens);
-      if (!streaming) send({ type: 'text', content: finalText }); // already shown if streaming
+      send({ type: 'text', content: finalText });
       send({ type: 'done' });
       return;
     }
 
     // ── Text before first tool call ───────────────────────────
-    if (!streaming) {
-      const textBefore = stripToolTags(cleaned.slice(0, calls[0].offset)).trim();
-      if (textBefore) send({ type: 'text', content: textBefore });
-    }
+    const textBefore = stripToolTags(cleaned.slice(0, calls[0].offset)).trim();
+    if (textBefore) send({ type: 'text', content: textBefore });
 
     history.push({ role: 'assistant', content: raw });
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "prior-cli",
-  "version": "1.4.0",
+  "version": "1.4.2",
   "description": "Prior Network AI — command-line interface",
   "bin": {
     "prior": "bin/prior.js"