From ddf693272e48b2e12cde03a66679761cd168042f Mon Sep 17 00:00:00 2001
From: oktofeesh1 <287075021+oktofeesh1@users.noreply.github.com>
Date: Thu, 2 Jul 2026 18:19:59 -0700
Subject: [PATCH] feat(review): add issue RAG query builder

---
 src/review/issue-rag-wire.ts     | 37 +++++++++++++++++++++
 src/review/rag.ts                |  2 +-
 test/unit/issue-rag-wire.test.ts | 57 ++++++++++++++++++++++++++++++++
 3 files changed, 95 insertions(+), 1 deletion(-)
 create mode 100644 src/review/issue-rag-wire.ts
 create mode 100644 test/unit/issue-rag-wire.test.ts

diff --git a/src/review/issue-rag-wire.ts b/src/review/issue-rag-wire.ts
new file mode 100644
index 0000000000..394ac01528
--- /dev/null
+++ b/src/review/issue-rag-wire.ts
@@ -0,0 +1,37 @@
+// Issue-centric RAG query composition (#2320). The miner analyze phase has no PR diff yet, so it feeds retrieval
+// from the issue's title/body/labels while reusing the existing RAG engine unchanged.
+
+import { MIN_QUERY_CHARS } from "./rag";
+
+const MAX_ISSUE_BODY_CHARS = 4000;
+const MAX_ISSUE_LABELS = 20;
+
+export type IssueRagQueryInput = {
+  title: string;
+  body?: string | undefined;
+  labels?: string[] | undefined;
+};
+
+function cleanLabels(labels: string[] | undefined): string[] {
+  if (!labels) return [];
+  return labels
+    .map((label) => label.trim())
+    .filter(Boolean)
+    .slice(0, MAX_ISSUE_LABELS);
+}
+
+export function buildIssueRagQuery(input: IssueRagQueryInput): { queryText: string } {
+  const sections: string[] = [];
+  const title = input.title.trim();
+  if (title) sections.push(title);
+
+  const body = (input.body ?? "").trim().slice(0, MAX_ISSUE_BODY_CHARS);
+  if (body) sections.push(body);
+
+  const labels = cleanLabels(input.labels);
+  if (labels.length > 0) sections.push(`Labels: ${labels.join(", ")}`);
+
+  const queryText = sections.join("\n\n").trim();
+  if (queryText.length < MIN_QUERY_CHARS) return { queryText: "" };
+  return { queryText };
+}
diff --git a/src/review/rag.ts b/src/review/rag.ts
index f7ce37f524..eba90a1d19 100644
--- a/src/review/rag.ts
+++ b/src/review/rag.ts
@@ -355,7 +355,7 @@ export async function deleteChunksForPaths(infra: RagInfra, project: string, rep
 // ── Retrieval (fail-safe: "" when anything is missing/broken) ────────────────────────────────────
 /** Skip retrieval for a trivially-short query (e.g. a one-word scope string): not worth an embed +
  *  a vector query, and the matches would be noise. (#cloud-opt) */
-const MIN_QUERY_CHARS = 40;
+export const MIN_QUERY_CHARS = 40;
 /** Hard cap on neighbours per query — bounds vector-index cost even if a caller passes a large topK. (#cloud-opt) */
 const RAG_MAX_TOPK = 20;
 const EMPTY_RAG_RETRIEVAL_METRICS: RagRetrievalMetrics = {
diff --git a/test/unit/issue-rag-wire.test.ts b/test/unit/issue-rag-wire.test.ts
new file mode 100644
index 0000000000..a9df5b9bcd
--- /dev/null
+++ b/test/unit/issue-rag-wire.test.ts
@@ -0,0 +1,57 @@
+import { describe, expect, it } from "vitest";
+import { buildIssueRagQuery } from "../../src/review/issue-rag-wire";
+
+describe("buildIssueRagQuery (#2320)", () => {
+  it("uses a sufficiently descriptive title when the issue has an empty body", () => {
+    expect(
+      buildIssueRagQuery({
+        title: "Add observability context for self-hosted review planning failures",
+        body: "",
+      }),
+    ).toEqual({
+      queryText: "Add observability context for self-hosted review planning failures",
+    });
+  });
+
+  it("prepends the issue title, includes the body, and appends labels as a hint line", () => {
+    const { queryText } = buildIssueRagQuery({
+      title: "Improve SQLite backup readiness checks",
+      body: "Operators need restore guidance tied to the existing self-host backup flow.",
+      labels: ["gittensor:feature", "selfhost", "  "],
+    });
+
+    expect(queryText).toContain("Improve SQLite backup readiness checks");
+    expect(queryText).toContain("Operators need restore guidance");
+    expect(queryText).toContain("Labels: gittensor:feature, selfhost");
+    expect(queryText.indexOf("Improve SQLite")).toBeLessThan(queryText.indexOf("Operators need"));
+    expect(queryText.indexOf("Operators need")).toBeLessThan(queryText.indexOf("Labels:"));
+  });
+
+  it("bounds long issue bodies without dropping the label hint", () => {
+    const { queryText } = buildIssueRagQuery({
+      title: "Investigate flaky queue dispatch telemetry",
+      body: `${"a".repeat(4100)}SHOULD_NOT_APPEAR`,
+      labels: ["queue"],
+    });
+
+    expect(queryText).toContain("Investigate flaky queue dispatch telemetry");
+    expect(queryText).toContain("Labels: queue");
+    expect(queryText).not.toContain("SHOULD_NOT_APPEAR");
+  });
+
+  it("returns an empty query for a one-line issue below the retrieval floor", () => {
+    expect(buildIssueRagQuery({ title: "Tiny" })).toEqual({ queryText: "" });
+  });
+
+  it("uses a descriptive body when the title is blank and omits blank labels", () => {
+    const { queryText } = buildIssueRagQuery({
+      title: "   ",
+      body: "Document how the miner should build issue context before a pull request exists.",
+      labels: [" ", ""],
+    });
+
+    expect(queryText).toBe(
+      "Document how the miner should build issue context before a pull request exists.",
+    );
+  });
+});