zames_pro 2.96.4 → 2.96.5

package/CHANGELOG.md

@@ -7,6 +7,16 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [2.96.5] - 2026-11-06
+
+### Fixed
+
+- **web:** tolerate whitespace around = in a DuckDuckGo result attribute
+- **web:** recognise x-zip and x-msdownload as binary content
+- **web:** drop the page head and keep <pre> code intact when converting HTML
+- **web:** share one in-flight headless context between concurrent fetches
+- **scheduler:** never scan cron forever on a non-finite fromMs
+
 ## [2.96.4] - 2026-11-06
 
 ### Fixed

package/dist/scheduler.js

@@ -10,6 +10,9 @@ export function parseInterval(text) {
     const n = Number(m[1]);
     if (!Number.isFinite(n) || n <= 0)
         return null;
+    // `n * mult` can overflow to Infinity for a huge digit string; the range
+    // check below then compares Infinity > MAX_TIMER_MS and rejects it, but only
+    // AFTER the multiplication — keep the guard explicit for both.
     const unit = (m[2] || 's').toLowerCase();
     const mult = unit.startsWith('h')
         ? 3_600_000
@@ -163,9 +166,15 @@ export function nextCronTime(expr, fromMs) {
     // 02:45 job created after the first 02:45 fired the NEXT DAY. A fixed
     // 60-second step visits every real minute exactly once; the field checks
     // still use the LOCAL getters, so the wall-clock match is unchanged.
+    // A non-finite/huge `fromMs` (a clock past 8.64e15, a hand-edited job) made
+    // `limit` Infinity and the scan walked one minute at a time FOREVER.
+    if (!Number.isFinite(fromMs) || Math.abs(fromMs) > 8.64e15)
+        return null;
     const step = 60_000;
     let t = Math.floor(fromMs / step) * step + step;
     const limit = fromMs + 4 * 366 * 24 * 60 * 60 * 1000;
+    if (!Number.isFinite(limit))
+        return null;
     while (t <= limit) {
         const d = new Date(t);
         if (c.minute.has(d.getMinutes()) &&

package/dist/web.js

@@ -37,6 +37,7 @@ export function htmlToText(html) {
     // document to the next.
     pendingHighSurrogate = null;
     // Remove blocks we don't need
+    const parkedPre = [];
     let s = html
         // A tag that never CLOSES (a truncated body, or HTML the regex cannot
         // pair) used to leak its raw JS/CSS into the text: also drop everything
@@ -47,6 +48,10 @@ export function htmlToText(html) {
         .replace(/<style\b[\s\S]*$/i, '')
         .replace(/<noscript[\s\S]*?<\/noscript>/gi, '')
         .replace(/<noscript\b[\s\S]*$/i, '')
+        // The <head> (title, meta) is page CHROME, not content: left in place the
+        // title glued itself to the first paragraph (`My pageBody`).
+        .replace(/<head[\s\S]*?<\/head>/gi, '')
+        .replace(/<title[\s\S]*?<\/title>/gi, '')
         .replace(/<svg[\s\S]*?<\/svg>/gi, '')
         .replace(/<svg\b[\s\S]*$/i, '')
         .replace(/<!--[\s\S]*?-->/g, '');
@@ -60,11 +65,18 @@ export function htmlToText(html) {
     // `<td>42</td>` became `Name42` — two independent values fused into one, and
     // the model reported a number that never appeared on the page.
     s = s
+        // A `<pre>` block is CODE: a bare `<` inside it (`if (a < b)`) is text, not
+        // the start of a tag, so the generic strip below would eat the rest of the
+        // line. Park the block, then restore it after the tag removal.
+        .replace(/<pre\b[^>]*>([\s\S]*?)<\/pre>/gi, (_m, code) => '\u0000PRE' + (parkedPre.push(code) - 1) + 'PRE\u0000')
         .replace(/<\/(h[1-6]|p|div|li|tr|section|article)>/gi, '\n')
         .replace(/<\/(td|th)>/gi, ' ')
         .replace(/<br\s*\/?>/gi, '\n')
         .replace(/<li[^>]*>/gi, '- ')
         .replace(/<[^>]+>/g, '');
+    // A parked `<pre>` block comes back AFTER the tag strip, with only its own
+    // tags removed and its text kept verbatim.
+    s = s.replace(/\u0000PRE(\d+)PRE\u0000/g, (_m, i) => (parkedPre[Number(i)] ?? '').replace(/<[^>]+>/g, ''));
     // HTML entities — the basic ones
     s = s
         .replace(/&nbsp;/g, ' ')
@@ -406,7 +418,7 @@ export function isBinaryContentType(contentType) {
     if (/\+(json|xml)$/.test(ct))
         return false;
     if (/^(image|audio|video|font|model)\//.test(ct) ||
-        /^application\/(pdf|zip|gzip|x-gzip|x-tar|octet-stream|msword|vnd\.|x-7z|x-rar|x-bzip|x-protobuf|wasm)/.test(ct)) {
+        /^application\/(pdf|zip|gzip|x-gzip|x-tar|octet-stream|msword|vnd\.|x-7z|x-rar|x-bzip|x-zip|x-msdownload|x-protobuf|wasm|epub\+zip)/.test(ct)) {
         return true;
     }
     return false;
@@ -503,13 +515,32 @@ async function getHeadless() {
 // tested without launching a real browser.
 export function createReusable(factory, dispose) {
     let cached = null;
+    // The FACTORY is async: checking `cached === null` and then awaiting it let
+    // two concurrent get() calls BOTH run the factory (two Chromium contexts,
+    // the first leaked). The in-flight promise is cached, so every caller waits
+    // for the SAME value.
+    let inflight = null;
     return {
         async get() {
-            if (cached === null)
-                cached = await factory();
-            return cached;
+            if (cached !== null)
+                return cached;
+            if (!inflight) {
+                inflight = factory().then((v) => {
+                    cached = v;
+                    inflight = null;
+                    return v;
+                }, (e) => {
+                    inflight = null;
+                    throw e;
+                });
+            }
+            return inflight;
         },
         async reset() {
+            if (inflight) {
+                // A failed factory must not leave a poisoned promise behind.
+                await inflight.catch(() => { });
+            }
             if (cached === null)
                 return;
             const v = cached;
@@ -673,9 +704,11 @@ export function parseDuckDuckGoResults(body, limit) {
     // through a proxy does), so each quote style is accepted rather than only
     // `class="…"`. Rejecting them made a valid page parse to zero results and
     // WebSearch answer "the output format may have changed".
-    const anchorRe = /<a[^>]*class=(?:"([^"]*)"|'([^']*)')[^>]*>([\s\S]*?)<\/a>/gi;
+    // HTML allows whitespace around the `=` of an attribute (`class = "x"`);
+    // requiring it to touch both sides made such a page parse to ZERO results.
+    const anchorRe = /<a[^>]*class\s*=\s*(?:"([^"]*)"|'([^']*)')[^>]*>([\s\S]*?)<\/a>/gi;
     const hrefOf = (tag) => {
-        const m = tag.match(/href=(?:"([^"]*)"|'([^']*)')/i);
+        const m = tag.match(/href\s*=\s*(?:"([^"]*)"|'([^']*)')/i);
         return m ? (m[1] ?? m[2] ?? '') : '';
     };
     let cur = null;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "zames_pro",
-  "version": "2.96.4",
+  "version": "2.96.5",
   "description": "Terminal coding agent that drives a real browser to chat with DeepSeek, Qwen or GLM — no API key required. Reads and edits files, runs commands, commits to git, like Claude Code / Codex CLI.",
   "type": "module",
   "bin": {