zames_pro 2.81.14 → 2.81.15

package/CHANGELOG.md

@@ -7,6 +7,12 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [2.81.15] - 2026-10-21
+
+### Fixed
+
+- **web:** a search result without a snippet no longer borrows the next one
+
 ## [2.81.14] - 2026-10-21
 
 ### Fixed

package/dist/web.js

@@ -645,33 +645,47 @@ export function createWebTools() {
 // wraps links in a /l/?uddg=... redirect which must be unwrapped.
 export function parseDuckDuckGoResults(body, limit) {
     const results = [];
-    const re = /<a[^>]*class="[^"]*result__a[^"]*"[^>]*href="([^"]+)"[^>]*>([\s\S]*?)<\/a>[\s\S]*?(?:<a[^>]*class="[^"]*result__snippet[^"]*"[^>]*>([\s\S]*?)<\/a>)?/gi;
+    // Parse the page ROW BY ROW: each `result__a` title opens a result, and a
+    // `result__snippet` belongs to the result opened most recently. The old
+    // single regex made the snippet group OPTIONAL with an unbounded gap, so a
+    // result without a snippet picked up the FOLLOWING result's snippet (wrong
+    // text under a URL). A row scan ties every snippet to its own result.
+    const anchorRe = /<a[^>]*class="[^"]*result__(a|snippet)[^"]*"[^>]*>([\s\S]*?)<\/a>/gi;
+    const hrefOf = (tag) => {
+        const m = tag.match(/href="([^"]*)"/i);
+        return m ? m[1] : '';
+    };
+    let cur = null;
     let m;
-    while ((m = re.exec(body)) && results.length < limit) {
-        let href = m[1];
-        const uddgMatch = href.match(/[?&]uddg=([^&]+)/);
-        if (uddgMatch) {
-            // decodeURIComponent THROWS on a malformed percent escape (`%ZZ`, a lone
-            // `%`). One broken href in the page must not abort the whole search: keep
-            // the raw href, which the http(s) check below then rejects.
-            try {
-                href = decodeURIComponent(uddgMatch[1]);
+    while ((m = anchorRe.exec(body))) {
+        const kind = m[1].toLowerCase();
+        if (kind === 'a') {
+            if (cur && cur.title && /^https?:\/\//i.test(cur.url)) {
+                results.push(cur);
+                if (results.length >= limit)
+                    return results;
             }
-            catch {
-                href = uddgMatch[1];
+            cur = { title: htmlToText(m[2] || '').trim(), url: '', snippet: '' };
+            // The href lives in the OPENING tag, which m[0] starts with.
+            let href = hrefOf(m[0].slice(0, m[0].indexOf('>') + 1));
+            const uddg = href.match(/[?&]uddg=([^&]+)/);
+            if (uddg) {
+                try {
+                    href = decodeURIComponent(uddg[1]);
+                }
+                catch {
+                    href = uddg[1];
+                }
             }
-        }
-        const title = htmlToText(m[2] || '').trim();
-        const snippet = htmlToText(m[3] || '').trim();
-        // Only an absolute http(s) link is a real result. The page can carry a
-        // `javascript:`/`data:` href (malicious or broken markup) or a
-        // site-relative one; handing either to the model as a result URL is wrong
-        // (a `javascript:` URL looks fetchable, a relative one is not).
-        if (!title || !/^https?:\/\//i.test(href))
+            cur.url = href;
             continue;
-        results.push({ title, url: href, snippet });
+        }
+        if (cur && !cur.snippet)
+            cur.snippet = htmlToText(m[2] || '').trim();
     }
-    return results;
+    if (cur && cur.title && /^https?:\/\//i.test(cur.url))
+        results.push(cur);
+    return results.slice(0, limit);
 }
 function formatResult(status, url, text, limit) {
     let out = text;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "zames_pro",
-  "version": "2.81.14",
+  "version": "2.81.15",
   "description": "Terminal coding agent that drives a real browser to chat with DeepSeek, Qwen and other web LLMs — no API key required. Reads and edits files, runs commands, commits to git, like Claude Code / Codex CLI.",
   "type": "module",
   "bin": {