@maintainabilityai/research-runner 0.1.14 → 0.1.16

package/dist/runner/nodes/hackernews-search.js

@@ -40,7 +40,10 @@ async function runHackerNewsSearch(opts) {
                     fromQuery: query,
                     title: r.title,
                     url,
-                    content: '', // HN search returns no abstract
+                    // Show HN / Ask HN posts carry a self-post body. URL-only
+                    // submissions have no body — leave empty; the synth agent
+                    // can read the linked URL if it needs more context.
+                    content: r.storyText,
                     score: pointsToScore(r.points),
                     publishedDate: r.createdAt || undefined,
                     authors: r.author ? [r.author] : undefined,

package/dist/search/hackernews-client.d.ts

@@ -15,6 +15,8 @@ export interface HackerNewsResult {
     points: number;
     numComments: number;
     createdAt: string;
+    /** Self-post body for Show HN / Ask HN; empty for URL submissions. */
+    storyText: string;
 }
 export interface HackerNewsSearchOpts {
     query: string;

package/dist/search/hackernews-client.js

@@ -39,6 +39,29 @@ async function hackerNewsSearch(opts) {
         points: h.points ?? 0,
         numComments: h.num_comments ?? 0,
         createdAt: h.created_at ?? '',
+        // Algolia returns the self-post body for Show HN / Ask HN. URL-only
+        // submissions have this empty. Strip basic HTML tags so the excerpt
+        // is readable in the issue comment.
+        storyText: stripBasicHtml(h.story_text ?? '').slice(0, 2000),
     })).filter(r => r.objectId && r.title);
     return { query: opts.query, results, responseBytes: Buffer.byteLength(rawText, 'utf8'), httpStatus };
 }
+/**
+ * HN Algolia returns Show HN / Ask HN bodies with light HTML
+ * (`<p>`, `<i>`, etc.). Strip tags + decode the common entities so
+ * the excerpt blockquote in the issue body reads cleanly. No need for
+ * a full HTML parser — these posts are plain text with a sprinkle of
+ * inline tags.
+ */
+function stripBasicHtml(s) {
+    return s
+        .replace(/<\/?(?:p|br|i|b|em|strong|code|pre|a|ul|ol|li)[^>]*>/gi, ' ')
+        .replace(/&nbsp;/g, ' ')
+        .replace(/&amp;/g, '&')
+        .replace(/&lt;/g, '<')
+        .replace(/&gt;/g, '>')
+        .replace(/&quot;/g, '"')
+        .replace(/&#x27;/g, "'")
+        .replace(/\s+/g, ' ')
+        .trim();
+}

package/dist/search/uspto-client.js

@@ -71,7 +71,10 @@ async function usptoSearch(opts) {
             patentNumber: num,
             title: meta.inventionTitle ?? '',
             abstract: '',
-            url: num ? `https://patents.google.com/patent/US${num}` : '',
+            // earliestPublicationNumber already carries the `US` prefix (e.g.
+            // `US20260064729A1`); patentNumber is plain digits. Avoid the double
+            // `USUS…` URL we were producing.
+            url: num ? `https://patents.google.com/patent/${num.startsWith('US') ? num : `US${num}`}` : '',
             grantedAt: meta.grantDate || meta.filingDate || meta.effectiveFilingDate || '',
             inventors: meta.firstInventorName ? [meta.firstInventorName] : [],
             _xmlUri: xmlUri,
@@ -80,16 +83,33 @@ async function usptoSearch(opts) {
     // Stage 2: parallel best-effort abstract fetch. The full-text XML carries
     // the <abstract> element; we regex it out rather than parsing the whole
     // document (the XML is large and we only want the abstract).
+    //
+    // Telemetry: count how many we attempted and how many succeeded so the
+    // archeologist progress log can surface "uspto abstracts: 3/5" instead of
+    // silently shipping empty `>` blockquotes to the synth agent.
+    let attempted = 0;
+    let succeeded = 0;
+    let missingUri = 0;
+    const failureCauses = [];
     await Promise.all(stage1.map(async (r) => {
         if (!r._xmlUri) {
+            missingUri += 1;
             return;
         }
+        attempted += 1;
         try {
             const xmlRes = await fetchImpl(r._xmlUri, {
                 method: 'GET',
-                headers: { 'X-API-Key': opts.apiKey, accept: 'application/xml' },
+                // The signed XML URI lives on a CDN/storage host, not api.uspto.gov.
+                // It accepts unauthenticated GETs; sending X-API-Key actually causes
+                // some hosts to 403. Use a vanilla request with a polite User-Agent.
+                headers: {
+                    accept: 'application/xml,text/xml,*/*',
+                    'user-agent': 'maintainabilityai-research-runner/1.0',
+                },
             });
             if (!xmlRes.ok) {
+                failureCauses.push(`http${xmlRes.status}`);
                 return;
             }
             const xml = await xmlRes.text();
@@ -97,10 +117,20 @@ async function usptoSearch(opts) {
             if (m) {
                 const stripped = m[1].replace(/<[^>]+>/g, ' ').replace(/\s+/g, ' ').trim();
                 r.abstract = stripped.slice(0, 1000);
+                succeeded += 1;
             }
+            else {
+                failureCauses.push('no-abstract-tag');
+            }
+        }
+        catch (err) {
+            failureCauses.push(err instanceof Error ? err.name : 'unknown');
         }
-        catch { /* ignore — best-effort */ }
     }));
+    if (records.length > 0 && process.env.RESEARCH_RUNNER_QUIET !== '1') {
+        process.stderr.write(`[research-runner] uspto abstracts: ${succeeded}/${attempted} fetched ` +
+            `(${missingUri} record(s) had no XML URI; failures: ${failureCauses.join(',') || 'none'})\n`);
+    }
     // Drop the internal _xmlUri marker before returning.
     const results = stage1.map(({ _xmlUri: _ignored, ...rest }) => rest);
     return {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@maintainabilityai/research-runner",
-  "version": "0.1.14",
+  "version": "0.1.16",
   "description": "Research + PRD agent runner — orchestrates the Archeologist and PRD pipelines for the MaintainabilityAI governance mesh",
   "license": "MIT",
   "author": "MaintainabilityAI",