From d6c202ae7eae1491072ec3ff7bdeaf19252a8ebc Mon Sep 17 00:00:00 2001
From: genisis0x <manietdavv@gmail.com>
Date: Mon, 29 Jun 2026 12:25:46 +0530
Subject: [PATCH] fix(memory-ingest): render Codex rollout message bodies

Codex rollout JSONL now carries each turn as a `response_item` record
with the message on `payload.{type:"message", role, content[]}`, not on
the legacy `payload.message`. parseTranscriptJsonl only read
`payload.message`, so every Codex session imported with an empty body and
message_count 0.

Handle the current `payload.type === "message"` shape alongside the
legacy wrapper: read `payload.role` and flatten `payload.content`. Adds a
product-path regression that ingests a response_item session and asserts
the staged transcript renders both the user and assistant turns (fails on
the old parser with an empty body).

Closes #2105
---
 bin/gstack-memory-ingest.ts       | 11 +++++-
 test/gstack-memory-ingest.test.ts | 64 +++++++++++++++++++++++++++++++
 2 files changed, 74 insertions(+), 1 deletion(-)

diff --git a/bin/gstack-memory-ingest.ts b/bin/gstack-memory-ingest.ts
index 653d4069a..77258c2b0 100644
--- a/bin/gstack-memory-ingest.ts
+++ b/bin/gstack-memory-ingest.ts
@@ -617,7 +617,7 @@ function parseTranscriptJsonl(path: string): ParsedSession | null {
       const tool = rec?.name || rec?.tool || rec?.tool_call?.name || "tool";
       bodyParts.push(`### Tool call: ${tool}`);
     } else if (isCodex && rec?.payload?.message) {
-      // Codex shape: each record has payload.message
+      // Legacy Codex shape: each record has payload.message
       const msg = rec.payload.message;
       const role = msg.role || "user";
       const content = extractContentText(msg);
@@ -625,6 +625,15 @@ function parseTranscriptJsonl(path: string): ParsedSession | null {
         bodyParts.push(`## ${role.charAt(0).toUpperCase() + role.slice(1)}\n\n${content}`);
         messageCount++;
       }
+    } else if (isCodex && rec?.payload?.type === "message") {
+      // Current Codex rollout shape: a `response_item` record carries
+      // payload.{role, content[]} directly, with no payload.message wrapper.
+      const role = rec.payload.role || "user";
+      const content = extractContentText(rec.payload);
+      if (content) {
+        bodyParts.push(`## ${role.charAt(0).toUpperCase() + role.slice(1)}\n\n${content}`);
+        messageCount++;
+      }
     }
   }
 
diff --git a/test/gstack-memory-ingest.test.ts b/test/gstack-memory-ingest.test.ts
index fef9070c4..9121c26a2 100644
--- a/test/gstack-memory-ingest.test.ts
+++ b/test/gstack-memory-ingest.test.ts
@@ -542,6 +542,70 @@ esac
     rmSync(home, { recursive: true, force: true });
   });
 
+  // Regression for #2105: current Codex rollout records carry the message on
+  // `payload.{type:"message", role, content[]}` (a `response_item`), not on the
+  // legacy `payload.message`. The old parser only read `payload.message`, so
+  // every Codex session staged with an empty body (message_count 0). This proves
+  // the rendered transcript carries the user + assistant turns.
+  it("renders Codex response_item message bodies (payload.type === message)", () => {
+    const home = makeTestHome();
+    const gstackHome = join(home, ".gstack");
+    mkdirSync(gstackHome, { recursive: true });
+
+    // Shim copies the staging dir so we can read the exact staged page bytes.
+    const binDir = join(home, "fake-bin");
+    mkdirSync(binDir, { recursive: true });
+    const stagingCopy = join(home, "staging-copy");
+    const script = `#!/usr/bin/env bash
+case "\${1:-}" in
+  --help|-h) echo "Usage: gbrain <command>"; echo "Commands:"; echo "  import <dir>   Import"; exit 0 ;;
+  import)
+    DIR="\${2:-}"
+    cp -R "\$DIR" "${stagingCopy}" 2>/dev/null || true
+    if [[ " \$* " == *" --json "* ]]; then
+      echo '{"status":"success","duration_s":0.1,"imported":1,"skipped":0,"errors":0,"chunks":1,"total_files":1}'
+    fi
+    exit 0 ;;
+  *) echo "unknown"; exit 2 ;;
+esac
+`;
+    const binPath = join(binDir, "gbrain");
+    writeFileSync(binPath, script, "utf-8");
+    chmodSync(binPath, 0o755);
+
+    const today = new Date();
+    const ymd = `${today.getFullYear()}-${String(today.getMonth() + 1).padStart(2, "0")}-${String(today.getDate()).padStart(2, "0")}`;
+    const session =
+      `{"type":"session_meta","payload":{"id":"codex-body-1","cwd":"/tmp/codex-body"},"timestamp":"${today.toISOString()}"}\n` +
+      `{"type":"response_item","payload":{"type":"message","role":"user","content":[{"type":"input_text","text":"hello from codex rollout"}]},"timestamp":"2026-05-01T00:00:01Z"}\n` +
+      `{"type":"response_item","payload":{"type":"message","role":"assistant","content":[{"type":"output_text","text":"codex assistant reply body"}]},"timestamp":"2026-05-01T00:00:02Z"}\n`;
+    writeCodexSession(home, ymd, session);
+
+    const r = runScript(["--bulk", "--include-unattributed", "--quiet"], {
+      HOME: home,
+      GSTACK_HOME: gstackHome,
+      PATH: `${binDir}:${process.env.PATH || ""}`,
+    });
+
+    expect(r.exitCode).toBe(0);
+    expect(existsSync(stagingCopy)).toBe(true);
+    const findMd = spawnSync("find", [stagingCopy, "-name", "*.md", "-type", "f"], {
+      encoding: "utf-8",
+    });
+    const mdPaths = (findMd.stdout || "").trim().split("\n").filter(Boolean);
+    const codexMd = mdPaths.find((p) => p.includes("/transcripts/codex/"));
+    expect(codexMd).toBeDefined();
+    const body = readFileSync(codexMd as string, "utf-8");
+
+    // Old parser left the body empty; both turns must now render.
+    expect(body).toContain("## User");
+    expect(body).toContain("hello from codex rollout");
+    expect(body).toContain("## Assistant");
+    expect(body).toContain("codex assistant reply body");
+
+    rmSync(home, { recursive: true, force: true });
+  });
+
   it("injects title/type/tags into the staged page's YAML frontmatter", () => {
     const home = makeTestHome();
     const gstackHome = join(home, ".gstack");