prior-cli 1.2.9 → 1.3.1

package/bin/prior.js

@@ -606,8 +606,9 @@ async function startChat(opts = {}) {
 
   // Conversation history (for agent mode, keeps full multi-turn context)
   const chatHistory = [];
-  let currentModel = opts.model || null;
-  let uncensored   = opts.uncensored || false;
+  let currentModel          = opts.model || null;
+  let uncensored            = opts.uncensored || false;
+  let _currentAbortController = null;
 
   // ── Live slash-command suggestions ──────────────────────────
   let clearSuggestions = () => {};
@@ -657,6 +658,11 @@ async function startChat(opts = {}) {
 
       if (_suggTimer) { clearTimeout(_suggTimer); _suggTimer = null; }
 
+      if (key.name === 'escape' && _currentAbortController) {
+        _currentAbortController.abort();
+        return;
+      }
+
       if (key.name === 'return' || key.name === 'enter' || (key.ctrl && key.name === 'c')) {
         clearSuggestions();
         return;
@@ -989,6 +995,7 @@ Keep it under 350 words. Write prior.md now.`;
             return approved;
           };
 
+          _currentAbortController = new AbortController();
           await runAgent({
             messages:       [...chatHistory, { role: 'user', content: input }],
             model:          currentModel,
@@ -996,6 +1003,7 @@ Keep it under 350 words. Write prior.md now.`;
             cwd:            process.cwd(),
             projectContext,
             confirm,
+            signal:         _currentAbortController.signal,
             send: ev => {
               switch (ev.type) {
 
@@ -1003,6 +1011,12 @@ Keep it under 350 words. Write prior.md now.`;
                   spinStart('thinking…');
                   break;
 
+                case 'cancelled':
+                  spinStop();
+                  process.stdout.write('\n');
+                  console.log(c.muted('  ✗ Cancelled'));
+                  break;
+
                 case 'tool_start':
                   spinStop();
                   _progressStarted = false;
@@ -1042,6 +1056,7 @@ Keep it under 350 words. Write prior.md now.`;
 
                 case 'text': {
                   spinStop();
+                  if (!ev.content) break;
                   const rendered  = renderMarkdown(ev.content);
                   const thinkTime = elapsed(Date.now() - _thinkStart);
                   console.log(c.brand('  Prior  ') + c.muted(`·  ${timeNow()}  ·  ${thinkTime}`));
@@ -1065,7 +1080,11 @@ Keep it under 350 words. Write prior.md now.`;
           });
         } catch (err) {
           spinStop();
-          console.error(c.err(`  ✗ ${err.message}`));
+          if (err.name !== 'AbortError') {
+            console.error(c.err(`  ✗ ${err.message}`));
+          }
+        } finally {
+          _currentAbortController = null;
         }
 
         chatHistory.push({ role: 'user', content: input });

package/lib/agent.js

@@ -10,12 +10,13 @@ const MAX_ITER   = 14;
 
 // ── Single inference call (server just runs Ollama + returns) ─
 
-async function infer(messages, model, token, { cwd, uncensored, projectContext } = {}) {
+async function infer(messages, model, token, { cwd, uncensored, projectContext } = {}, signal) {
   const res = await fetch(`${CLI_BASE}/api/infer`, {
     method:  'POST',
     headers: { 'Content-Type': 'application/json' },
     body:    JSON.stringify({ messages, model, token, cwd, uncensored, projectContext }),
     timeout: 120000,
+    signal,
   });
   if (!res.ok) {
     const err = await res.json().catch(() => ({}));
@@ -150,7 +151,7 @@ function stripToolTags(text) {
 
 const CONFIRM_TOOLS = new Set(['run_command', 'file_delete', 'file_write']);
 
-async function runAgent({ messages, model, uncensored, cwd, projectContext, send, confirm }) {
+async function runAgent({ messages, model, uncensored, cwd, projectContext, send, confirm, signal }) {
   const token   = getToken();
   const history = [...messages];
 
@@ -159,13 +160,24 @@ async function runAgent({ messages, model, uncensored, cwd, projectContext, send
 
   for (let iter = 0; iter < MAX_ITER; iter++) {
 
+    if (signal?.aborted) {
+      send({ type: 'cancelled' });
+      send({ type: 'done' });
+      return;
+    }
+
     send({ type: 'thinking' });
 
     let result;
     try {
-      result = await infer(history, model || 'qwen3.5:4b', token, { cwd, uncensored, projectContext });
+      result = await infer(history, model || 'qwen3.5:4b', token, { cwd, uncensored, projectContext }, signal);
     } catch (err) {
       await trackTokenUsage(token, totalPromptTokens, totalCompletionTokens);
+      if (err.name === 'AbortError' || signal?.aborted) {
+        send({ type: 'cancelled' });
+        send({ type: 'done' });
+        return;
+      }
       send({ type: 'error', message: err.message });
       send({ type: 'done' });
       return;
@@ -186,8 +198,15 @@ async function runAgent({ messages, model, uncensored, cwd, projectContext, send
 
     // ── No tool calls → final answer ──────────────────────────
     if (calls.length === 0) {
+      const finalText = stripToolTags(cleaned);
+      if (!finalText && iter < MAX_ITER - 1) {
+        // Model returned blank (all think tags, no actual output) — nudge once
+        history.push({ role: 'assistant', content: raw });
+        history.push({ role: 'user', content: '(Your response was empty. Please write your reply.)' });
+        continue;
+      }
       await trackTokenUsage(token, totalPromptTokens, totalCompletionTokens);
-      send({ type: 'text', content: stripToolTags(cleaned) });
+      send({ type: 'text', content: finalText });
       send({ type: 'done' });
       return;
     }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "prior-cli",
-  "version": "1.2.9",
+  "version": "1.3.1",
   "description": "Prior Network AI — command-line interface",
   "bin": {
     "prior": "bin/prior.js"