@absolutejs/rag 0.22.6 → 0.22.8

package/CHANGELOG.md

@@ -6,6 +6,19 @@ This file is generated by `absolute-changelog` from the entries in
 `changelog/`. Edit an entry, not this file — and add new ones under
 `changelog/unreleased/`.
 
+## 0.22.8 — 2026-10-28
+
+### Fixed
+
+- **Allow bounded research to follow validated links discovered in fetched pages, without treating unread links as evidence or admitting invented URLs.** (`createResearch`)
+
+## 0.22.7 — 2026-10-28
+
+### Fixed
+
+- **Separate structured research tasks from literal search queries** (`createResearch`, `researchPlugin`, `ResearchInput`)
+  Task mode plans concise initial searches within the existing round budget and does not treat quoted JSON fields as required source phrases. Exact-phrase search mode retains identity filtering.
+
 ## 0.22.6 — 2026-10-22
 
 ### Changed

package/changelog.json

@@ -2,6 +2,35 @@
 	"contract": 1,
 	"name": "@absolutejs/rag",
 	"releases": [
+		{
+			"changes": [
+				{
+					"kind": "fixed",
+					"summary": "Allow bounded research to follow validated links discovered in fetched pages, without treating unread links as evidence or admitting invented URLs.",
+					"symbols": [
+						"createResearch"
+					]
+				}
+			],
+			"date": "2026-10-28",
+			"version": "0.22.8"
+		},
+		{
+			"changes": [
+				{
+					"detail": "Task mode plans concise initial searches within the existing round budget and does not treat quoted JSON fields as required source phrases. Exact-phrase search mode retains identity filtering.",
+					"kind": "fixed",
+					"summary": "Separate structured research tasks from literal search queries",
+					"symbols": [
+						"createResearch",
+						"researchPlugin",
+						"ResearchInput"
+					]
+				}
+			],
+			"date": "2026-10-28",
+			"version": "0.22.7"
+		},
 		{
 			"changes": [
 				{

package/dist/research/index.js

@@ -121450,9 +121450,10 @@ ${prompt}`,
       return generated.object;
     };
     const normalize = (text) => text.normalize("NFKC").toLowerCase().replace(/\s+/gu, " ").trim();
-    const requiredPhrases = input.requiredPhrases ?? [...input.query.matchAll(/"([^"\n]+)"/gu)].map((match) => match[1]);
+    const requiredPhrases = input.requiredPhrases ?? (input.queryMode === "task" ? [] : [...input.query.matchAll(/"([^"\n]+)"/gu)].map((match) => match[1]));
     const urls = new Set;
     const queries = new Set;
+    const discoveredLinks = new Map;
     let evidenceChars = 0;
     const sourceText = new Map;
     const rebalanceEvidence = () => {
@@ -121524,7 +121525,7 @@ ${prompt}`,
     const doRead = async (url) => {
       if (urls.has(url) || urls.size >= limits.reads)
         return;
-      if (!result.searches.some((search) => search.sources.some((source) => source.url === url)))
+      if (!discoveredLinks.has(url) && !result.searches.some((search) => search.sources.some((source) => source.url === url)))
         return;
       urls.add(url);
       const page = await operate("read", () => (config.reader ?? readRAGWebpage)({
@@ -121537,6 +121538,16 @@ ${prompt}`,
       if (page.status !== "ok")
         result.limitations.push(`Page read ${page.status}: ${url}`);
       if (page.text && page.status !== "error") {
+        for (const link of page.evidence.links) {
+          if (discoveredLinks.size >= 64)
+            break;
+          try {
+            const target = validatePublicWebUrl(new URL(link.url, page.finalUrl).href);
+            target.hash = "";
+            if (!discoveredLinks.has(target.href))
+              discoveredLinks.set(target.href, { url: target.href, label: link.label.slice(0, 200), discoveredOn: page.finalUrl });
+          } catch {}
+        }
         const prior = result.sources.find((source) => source.url === url);
         if (prior) {
           sourceText.set(prior.url, page.text.slice(0, limits.evidenceChars));
@@ -121557,18 +121568,30 @@ ${prompt}`,
       }
     };
     try {
-      for (const query of input.queries ?? [input.query])
-        await doSearch(query);
-      for (let round = 0;round < limits.rounds; round++) {
+      const needsInitialPlan = !input.queries && (input.queryMode === "task" || boundedQueries(input.query).length > 1) && limits.rounds > 0;
+      if (needsInitialPlan) {
+        const initial = await generate("plan", PlanSchema, "Turn the research task into concise search-engine queries. Preserve the requested company identity and target criteria, but never copy instructions, JSON, internal markers, or the whole brief into a query. Prioritize the most useful query first. Return at most remainingSearches queries. Return no URLs: source URLs must be discovered rather than invented.", {
+          query: input.query,
+          instructions: task.instructions,
+          remainingSearches: limits.searches
+        });
+        for (const query of initial.queries)
+          await doSearch(query);
+      } else {
+        for (const query of input.queries ?? [input.query])
+          await doSearch(query);
+      }
+      for (let round = needsInitialPlan ? 1 : 0;round < limits.rounds; round++) {
         if (!result.sources.length && queries.size >= limits.searches)
           break;
         let plan;
         try {
-          plan = await generate("plan", PlanSchema, "Select follow-up searches for missing evidence and primary-source URLs to read. For current people or roles, seek the organization\u2019s own leadership/about/team pages; testimonials, package authors, and job listings are not staff identity evidence. For policies, read the full policy or agreement to preserve section scope and exceptions. URLs must come from the supplied search results. done means available evidence addresses the requested schema, not merely that some results exist. Return no actions if further work is not useful.", {
+          plan = await generate("plan", PlanSchema, "Select follow-up searches for missing evidence and primary-source URLs to read. For current people or roles, seek the organization\u2019s own leadership/about/team pages; testimonials, package authors, and job listings are not staff identity evidence. For policies, read the full policy or agreement to preserve section scope and exceptions. URLs must come from the supplied search results or discoveredLinks. Discovered links are unread navigation hints, not evidence; read relevant product, pricing, checkout or policy details before concluding information is absent. done means available evidence addresses the requested schema, not merely that some results exist. Return no actions if further work is not useful.", {
             query: input.query,
             schema: task.schema,
             instructions: task.instructions,
             sources: result.sources,
+            discoveredLinks: [...discoveredLinks.values()].filter((link) => !urls.has(link.url)),
             searched: [...queries],
             read: [...urls],
             remainingSearches: limits.searches - queries.size,
@@ -121669,6 +121692,7 @@ var renderResearchResult = (result) => `<section aria-label="Research result"><p
 }).join("")}</li>`).join("")}</ul>${result.limitations.map((message) => `<p>${escape5(message)}</p>`).join("")}</section>`;
 var bodySchema = t.Object({
   query: t.String({ minLength: 1, maxLength: 8000 }),
+  queryMode: t.Optional(t.Union([t.Literal("search"), t.Literal("task")])),
   task: t.Optional(t.String({ maxLength: 128 })),
   freshness: t.Optional(t.String({ maxLength: 64 }))
 }, { additionalProperties: false });
@@ -122388,6 +122412,7 @@ var researchManifest = defineManifest()({
       },
       input: Type.Object({
         query: Type.String({ minLength: 1, maxLength: 8000 }),
+        queryMode: Type.Optional(Type.Union([Type.Literal("search"), Type.Literal("task")])),
         task: Type.Optional(Type.String({ maxLength: 128 }))
       }),
       handler: async (input, runtime) => JSON.stringify(await runtime.run(input))
@@ -122486,5 +122511,5 @@ export {
   selectResearchMonitorEvents
 };
 
-//# debugId=BC8292806DC74FDD64756E2164756E21
+//# debugId=98A81F60666A9C0F64756E2164756E21
 //# sourceMappingURL=index.js.map