nothumanallowed 13.5.91 → 13.5.93

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "nothumanallowed",
-  "version": "13.5.91",
+  "version": "13.5.93",
   "description": "NotHumanAllowed — 38 AI agents, 80 tools, Studio (visual agentic workflows). Email, calendar, browser automation, screen capture, canvas, cron/heartbeat, Alexandria E2E messaging, GitHub, Notion, Slack, voice chat, free AI (Liara), 28 languages. Zero-dependency CLI.",
   "type": "module",
   "bin": {

package/src/commands/ui.mjs

@@ -3856,7 +3856,7 @@ ${completedHeadings ? `## SECTIONS ALREADY WRITTEN (headings only):\n${completed
           return;
         }
         try {
-          const result = await callLLM(config, body.system, body.user, { max_tokens: body.max_tokens || 4096 });
+          const result = await callLLM(config, body.system, body.user, { max_tokens: body.max_tokens || 8192, temperature: 0.15 });
           sendJSON(res, 200, { text: result });
         } catch (e) {
           sendJSON(res, 500, { error: e.message });

package/src/services/llm.mjs

@@ -7,14 +7,15 @@
 
 // ── Providers ──────────────────────────────────────────────────────────────
 
-export async function callAnthropic(apiKey, model, systemPrompt, userMessage, stream = false) {
+export async function callAnthropic(apiKey, model, systemPrompt, userMessage, stream = false, opts = {}) {
   const body = {
     model: model || 'claude-sonnet-4-20250514',
-    max_tokens: 8192,
+    max_tokens: opts.max_tokens || 8192,
     system: systemPrompt,
     messages: [{ role: 'user', content: userMessage }],
     stream,
   };
+  if (opts.temperature !== undefined) body.temperature = opts.temperature;
   const res = await fetch('https://api.anthropic.com/v1/messages', {
     method: 'POST',
     headers: {
@@ -33,16 +34,17 @@ export async function callAnthropic(apiKey, model, systemPrompt, userMessage, st
   return data.content?.[0]?.text || '';
 }
 
-export async function callOpenAI(apiKey, model, systemPrompt, userMessage, stream = false) {
+export async function callOpenAI(apiKey, model, systemPrompt, userMessage, stream = false, opts = {}) {
   const body = {
     model: model || 'gpt-4o',
-    max_tokens: 8192,
+    max_tokens: opts.max_tokens || 8192,
     messages: [
       { role: 'system', content: systemPrompt },
       { role: 'user', content: userMessage },
     ],
     stream,
   };
+  if (opts.temperature !== undefined) body.temperature = opts.temperature;
   const res = await fetch('https://api.openai.com/v1/chat/completions', {
     method: 'POST',
     headers: {
@@ -60,13 +62,15 @@ export async function callOpenAI(apiKey, model, systemPrompt, userMessage, strea
   return data.choices?.[0]?.message?.content || '';
 }
 
-export async function callGemini(apiKey, model, systemPrompt, userMessage, _stream = false) {
+export async function callGemini(apiKey, model, systemPrompt, userMessage, _stream = false, opts = {}) {
   const m = model || 'gemini-2.5-pro-preview-05-06';
   const url = `https://generativelanguage.googleapis.com/v1beta/models/${m}:generateContent?key=${apiKey}`;
+  const generationConfig = { maxOutputTokens: opts.max_tokens || 8192 };
+  if (opts.temperature !== undefined) generationConfig.temperature = opts.temperature;
   const body = {
     system_instruction: { parts: [{ text: systemPrompt }] },
     contents: [{ parts: [{ text: userMessage }] }],
-    generationConfig: { maxOutputTokens: 8192 },
+    generationConfig,
   };
   const res = await fetch(url, {
     method: 'POST',
@@ -81,16 +85,17 @@ export async function callGemini(apiKey, model, systemPrompt, userMessage, _stre
   return data.candidates?.[0]?.content?.parts?.[0]?.text || '';
 }
 
-export async function callDeepSeek(apiKey, model, systemPrompt, userMessage, stream = false) {
+export async function callDeepSeek(apiKey, model, systemPrompt, userMessage, stream = false, opts = {}) {
   const body = {
     model: model || 'deepseek-chat',
-    max_tokens: 8192,
+    max_tokens: opts.max_tokens || 8192,
     messages: [
       { role: 'system', content: systemPrompt },
       { role: 'user', content: userMessage },
     ],
     stream,
   };
+  if (opts.temperature !== undefined) body.temperature = opts.temperature;
   const res = await fetch('https://api.deepseek.com/v1/chat/completions', {
     method: 'POST',
     headers: {
@@ -108,16 +113,17 @@ export async function callDeepSeek(apiKey, model, systemPrompt, userMessage, str
   return data.choices?.[0]?.message?.content || '';
 }
 
-export async function callGrok(apiKey, model, systemPrompt, userMessage, stream = false) {
+export async function callGrok(apiKey, model, systemPrompt, userMessage, stream = false, opts = {}) {
   const body = {
     model: model || 'grok-3-latest',
-    max_tokens: 8192,
+    max_tokens: opts.max_tokens || 8192,
     messages: [
       { role: 'system', content: systemPrompt },
       { role: 'user', content: userMessage },
     ],
     stream,
   };
+  if (opts.temperature !== undefined) body.temperature = opts.temperature;
   const res = await fetch('https://api.x.ai/v1/chat/completions', {
     method: 'POST',
     headers: {
@@ -135,16 +141,17 @@ export async function callGrok(apiKey, model, systemPrompt, userMessage, stream
   return data.choices?.[0]?.message?.content || '';
 }
 
-export async function callMistral(apiKey, model, systemPrompt, userMessage, stream = false) {
+export async function callMistral(apiKey, model, systemPrompt, userMessage, stream = false, opts = {}) {
   const body = {
     model: model || 'mistral-large-latest',
-    max_tokens: 8192,
+    max_tokens: opts.max_tokens || 8192,
     messages: [
       { role: 'system', content: systemPrompt },
       { role: 'user', content: userMessage },
     ],
     stream,
   };
+  if (opts.temperature !== undefined) body.temperature = opts.temperature;
   const res = await fetch('https://api.mistral.ai/v1/chat/completions', {
     method: 'POST',
     headers: {
@@ -162,13 +169,14 @@ export async function callMistral(apiKey, model, systemPrompt, userMessage, stre
   return data.choices?.[0]?.message?.content || '';
 }
 
-export async function callCohere(apiKey, model, systemPrompt, userMessage, _stream = false) {
+export async function callCohere(apiKey, model, systemPrompt, userMessage, _stream = false, opts = {}) {
   const body = {
     model: model || 'command-r-plus',
-    max_tokens: 8192,
+    max_tokens: opts.max_tokens || 8192,
     preamble: systemPrompt,
     message: userMessage,
   };
+  if (opts.temperature !== undefined) body.temperature = opts.temperature;
   const res = await fetch('https://api.cohere.ai/v1/chat', {
     method: 'POST',
     headers: {
@@ -236,7 +244,7 @@ export async function streamSSE(res, format) {
  * NHA Free (Liara) — free LLM tier, no API key required.
  * Qwen3 32B on Hetzner RTX 6000 Pro 96GB. Supports thinking mode.
  */
-export async function callNHA(apiKey, model, systemPrompt, userMessage, stream = false) {
+export async function callNHA(apiKey, model, systemPrompt, userMessage, stream = false, opts = {}) {
   // Read thinking preference from config
   let thinkingEnabled = false; // OFF by default for speed
   try {
@@ -266,7 +274,7 @@ export async function callNHA(apiKey, model, systemPrompt, userMessage, stream =
 
   const body = {
     model: model || '/opt/models/qwen3-32b',
-    max_tokens: thinkingEnabled ? 16384 : 8192,
+    max_tokens: opts.max_tokens || (thinkingEnabled ? 16384 : 8192),
     messages: [
       { role: 'system', content: sanitizeForSentinel(systemPrompt) },
       { role: 'user', content: sanitizeForSentinel(userMessage) },
@@ -274,6 +282,7 @@ export async function callNHA(apiKey, model, systemPrompt, userMessage, stream =
     stream,
     chat_template_kwargs: { enable_thinking: thinkingEnabled },
   };
+  if (opts.temperature !== undefined) body.temperature = opts.temperature;
   // Route through NHA server proxy (SENTINEL protection) instead of direct to Hetzner
   const res = await fetch('https://nothumanallowed.com/api/v1/liara/chat', {
     method: 'POST',
@@ -338,7 +347,7 @@ export async function callLLM(config, systemPrompt, userMessage, opts = {}) {
   const callFn = getProviderCall(provider);
   if (!callFn) throw new Error(`Unknown provider: ${provider}`);
 
-  return callFn(apiKey, model, systemPrompt, userMessage, false);
+  return callFn(apiKey, model, systemPrompt, userMessage, false, opts);
 }
 
 /**

package/src/services/web-ui.mjs

@@ -7987,15 +7987,43 @@ async function wcGenerate() {
   _wcGenStartTime = Date.now();
   _wcTokIn = 0; _wcTokOut = 0;  // reset global counters for this generation run
 
-  // Helper: generate one file
+  // CSS files that need two-pass generation (too long for one call)
+  var WC_CSS_SPLIT = {
+    'public/css/base.css': [
+      'PART 1 of 2. Generate the FIRST HALF of public/css/base.css. Include: (1) all CSS custom properties / design tokens (colors, spacing, font sizes, shadows, radii, transitions, z-index scale, dark/light mode via prefers-color-scheme data-theme), (2) CSS reset (*, box-sizing, margin, padding), (3) base typography (body, h1-h6, p, a, code, pre, blockquote), (4) utility classes (flex, grid helpers, spacing, text alignment, visibility, truncation). End the file at a natural boundary (closing brace). Do NOT generate components. Output raw CSS only.',
+      'PART 2 of 2. Continue (do NOT repeat) public/css/base.css from where part 1 ended. Generate: (5) layout helpers (.container, .grid, .col-*, .stack, .cluster, .sidebar-layout), (6) responsive breakpoint utilities (768px, 480px), (7) animation keyframes (@keyframes fadeIn, slideUp, pulse, spin), (8) scrollbar styling, (9) selection styles, (10) print styles. Output raw CSS only, starting directly from where part 1 ended — no repetition.'
+    ],
+    'public/css/components.css': [
+      'PART 1 of 2. Generate the FIRST HALF of public/css/components.css using strict BEM. Include components: (1) .btn (--primary, --secondary, --danger, --ghost, --sm, --lg, disabled state, loading state with spinner), (2) .form (.form__group, .form__label, .form__input, .form__textarea, .form__select, .form__error, .form__hint, .form__input--invalid, focus states), (3) .card (.card__header, .card__body, .card__footer, .card--interactive hover/active), (4) .badge (--success, --error, --warning, --info, --neutral), (5) .alert (--success, --error, --warning, --info with icon space). Output raw CSS only.',
+      'PART 2 of 2. Continue (do NOT repeat) public/css/components.css from where part 1 ended. Include components: (6) .nav (.nav__brand, .nav__links, .nav__link, .nav__link--active, .nav__toggle mobile hamburger, .nav--sticky), (7) .modal (.modal__overlay, .modal__content, .modal__header, .modal__body, .modal__footer, open/close transition), (8) .spinner (sizes: sm/md/lg, colors), (9) .dropdown (.dropdown__menu, .dropdown__item, open state), (10) .avatar (.avatar--sm/md/lg, .avatar--initials), (11) .progress (.progress__bar, animated fill), (12) .table (.table__head, .table__row, .table__cell, striped, hover). Output raw CSS only, starting directly from where part 1 ended.'
+    ]
+  };
+
+  // Helper: strip markdown fences from LLM output
+  function wcStripFences(content) {
+    var _nl2 = String.fromCharCode(10);
+    var _fence = String.fromCharCode(96,96,96);
+    var lines = content.split(_nl2);
+    if (lines.length > 0 && lines[0].indexOf(_fence) === 0) lines.shift();
+    if (lines.length > 0 && lines[lines.length-1].trim() === _fence) lines.pop();
+    return lines.join(_nl2).trim();
+  }
+
+  // Helper: generate one file (with two-pass split for large CSS files)
   async function wcGenOneFile(fp, signal) {
     var _nl2 = String.fromCharCode(10);
+    var splitPrompts = WC_CSS_SPLIT[fp.name];
+    if (splitPrompts) {
+      // Two-pass generation: call LLM twice and concatenate
+      var part1 = await wcCallLLM(sysPreamble, splitPrompts[0] + _nl2 + _nl2 + 'File: ' + fp.name, signal, fp.lang, 8192);
+      part1 = wcStripFences(part1);
+      if (signal && signal.aborted) return part1;
+      var part2 = await wcCallLLM(sysPreamble, splitPrompts[1] + _nl2 + _nl2 + 'File: ' + fp.name, signal, fp.lang, 8192);
+      part2 = wcStripFences(part2);
+      return part1 + _nl2 + _nl2 + part2;
+    }
     var content = await wcCallLLM(sysPreamble, fp.prompt + _nl2 + _nl2 + 'File to generate: ' + fp.name, signal, fp.lang);
-    var _fence = String.fromCharCode(96,96,96);
-    var wcLines = content.split(_nl2);
-    if (wcLines.length > 0 && wcLines[0].indexOf(_fence) === 0) wcLines.shift();
-    if (wcLines.length > 0 && wcLines[wcLines.length-1].trim() === _fence) wcLines.pop();
-    return wcLines.join(_nl2).trim();
+    return wcStripFences(content);
   }
 
   wcStartGenTimer();
@@ -8094,7 +8122,8 @@ async function wcAutoRepair(filePlan, sysPreamble) {
   if (filePlan) filePlan.forEach(function(fp){ planMap[fp.name] = fp; });
 
   var _nl3 = String.fromCharCode(10);
-  var sysBase = sysPreamble || ('You are an expert full-stack engineer. Output ONLY the complete corrected file content. No explanations, no markdown fences.');
+  // Use compact system prompt for repair to avoid exceeding Liara context window
+  var sysBase = 'You are an expert full-stack engineer. Output ONLY the complete corrected file content. No explanations, no markdown fences, no preamble. Raw file content only.';
 
   for (var ri = 0; ri < toFix.length; ri++) {
     var broken = toFix[ri];
@@ -8120,7 +8149,7 @@ async function wcAutoRepair(filePlan, sysPreamble) {
           (broken.content.length > 800 ? broken.content.slice(0, 400) + _nl3 + '...' + _nl3 + broken.content.slice(-400) : broken.content) + _nl3 + _nl3 +
           'Output the COMPLETE corrected file from the beginning.';
       }
-      var fixed = await wcCallLLM(fixSys, fixUser, null, broken.lang || plan && plan.lang);
+      var fixed = await wcCallLLM(fixSys, fixUser, null, broken.lang || plan && plan.lang, 8192);
       var _fence3 = String.fromCharCode(96,96,96);
       var fixLines = fixed.split(_nl3);
       if (fixLines.length > 0 && fixLines[0].indexOf(_fence3) === 0) fixLines.shift();
@@ -8230,11 +8259,11 @@ function wcIsTruncated(content, lang) {
   return false;
 }
 
-async function wcCallLLMRaw(sys, user, signal) {
+async function wcCallLLMRaw(sys, user, signal, maxTok) {
   var fetchOpts = {
     method: 'POST',
     headers: {'Content-Type':'application/json'},
-    body: JSON.stringify({system: sys, user: user, max_tokens: 8192})
+    body: JSON.stringify({system: sys, user: user, max_tokens: maxTok || 16384})
   };
   if (signal) fetchOpts.signal = signal;
   for (var attempt = 0; attempt < 3; attempt++) {
@@ -8262,8 +8291,8 @@ async function wcCallLLMRaw(sys, user, signal) {
   }
 }
 
-async function wcCallLLM(sys, user, signal, lang) {
-  var content = await wcCallLLMRaw(sys, user, signal);
+async function wcCallLLM(sys, user, signal, lang, maxTok) {
+  var content = await wcCallLLMRaw(sys, user, signal, maxTok);
   // Continuation loop: if response is truncated, ask model to continue
   var maxContinuations = 2;
   for (var ci = 0; ci < maxContinuations; ci++) {
@@ -8272,7 +8301,7 @@ async function wcCallLLM(sys, user, signal, lang) {
     var continuePrompt = 'Continue generating the file EXACTLY from where you stopped. Do not repeat anything already written. Output ONLY the remaining code, starting from the next character after where you stopped.' +
       String.fromCharCode(10) + String.fromCharCode(10) + 'The file so far ends with:' +
       String.fromCharCode(10) + content.slice(-300);
-    var continuation = await wcCallLLMRaw(sys, continuePrompt, signal);
+    var continuation = await wcCallLLMRaw(sys, continuePrompt, signal, maxTok);
     if (!continuation || continuation.trim().length < 5) break;
     content = content + String.fromCharCode(10) + continuation;
   }