diff --git a/apps/server/src/project/AgentSessionScanner.test.ts b/apps/server/src/project/AgentSessionScanner.test.ts
index f6d95d8cc6ec..ce75e456158a 100644
--- a/apps/server/src/project/AgentSessionScanner.test.ts
+++ b/apps/server/src/project/AgentSessionScanner.test.ts
@@ -2024,6 +2024,144 @@ it.layer(NodeServices.layer)("AgentSessionScanner", (it) => {
}),
);
+ it.effect("ignores sessions without visible messages without spending the import budget", () =>
+ Effect.gen(function* () {
+ const path = yield* Path.Path;
+ const nowMs = Date.parse("2026-08-24T12:00:00.000Z");
+ yield* TestClock.setTime(nowMs);
+ const claudeHomePath = yield* makeTempDir("t3code-claude-home-");
+ const codexHomePath = yield* makeTempDir("t3code-codex-home-");
+ const workspace = yield* makeTempDir("t3code-workspace-");
+ const codexRollout = (name: string) =>
+ path.join(codexHomePath, "sessions", "2026", "08", "24", `rollout-${name}.jsonl`);
+ // Newer empty sessions would exhaust the 100-transcript import budget before any real one.
+ const emptySessions = Array.from({ length: 100 }, (_, index) =>
+ index % 2 === 0
+ ? {
+ filePath: codexRollout(`empty-${index}`),
+ records: [
+ {
+ type: "session_meta",
+ payload: { id: `codex-empty-${index}`, cwd: workspace, source: "vscode" },
+ },
+ {
+ type: "event_msg",
+ payload: {
+ type: "thread_settings_applied",
+ thread_settings: { cwd: workspace },
+ },
+ },
+ ],
+ }
+ : {
+ filePath: path.join(
+ claudeHomePath,
+ "projects",
+ "-slug",
+ `claude-empty-${index}.jsonl`,
+ ),
+ records: [
+ { type: "system", subtype: "informational", content: "Status", cwd: workspace },
+ {
+ type: "user",
+ isMeta: true,
+ message: {
+ role: "user",
+ content: "Caveat",
+ },
+ cwd: workspace,
+ },
+ ],
+ },
+ );
+ const transcripts = [
+ ...emptySessions,
+ {
+ // Without session_meta there is no resumable session ID.
+ filePath: codexRollout("unusable"),
+ records: [
+ { type: "turn_context", payload: { cwd: workspace } },
+ {
+ type: "event_msg",
+ payload: { type: "user_message", message: "Import this prompt" },
+ },
+ ],
+ },
+ {
+ filePath: codexRollout("real"),
+ records: [
+ { type: "session_meta", payload: { id: "real-session", cwd: workspace } },
+ {
+ type: "event_msg",
+ payload: { type: "user_message", message: "Import this prompt" },
+ },
+ ],
+ },
+ ];
+ for (const [index, { filePath, records }] of transcripts.entries()) {
+ yield* writeTranscript({
+ filePath,
+ contents: records.map((record) => encodeTranscriptRecord(record)).join("\n"),
+ mtimeMs: nowMs - index * 60_000,
+ });
+ }
+
+ const outcomes = yield* runRecentThreadOutcomes({
+ claudeHomePath,
+ codexHomePath,
+ workspaceRoot: workspace,
+ });
+
+ expect(
+ outcomes.map((outcome) =>
+ outcome._tag === "Importable" ? outcome.thread.providerSessionId : outcome._tag,
+ ),
+ ).toEqual(["Skipped", "real-session"]);
+ }),
+ );
+
+ it.effect("skips a rewound Codex rollout whose inherited history is unavailable", () =>
+ Effect.gen(function* () {
+ const path = yield* Path.Path;
+ const nowMs = Date.parse("2026-08-24T12:00:00.000Z");
+ yield* TestClock.setTime(nowMs);
+ const claudeHomePath = yield* makeTempDir("t3code-claude-home-");
+ const codexHomePath = yield* makeTempDir("t3code-codex-home-");
+ const workspace = yield* makeTempDir("t3code-workspace-");
+ const threadId = "019a0000-0000-7000-8000-000000000001";
+ const rolloutId = "019a0000-0000-7000-8000-000000000002";
+ // The ancestor rollout is missing, and this rollout added no messages of its own.
+ yield* writeTranscript({
+ filePath: path.join(
+ codexHomePath,
+ "sessions",
+ "2026",
+ "08",
+ "24",
+ `rollout-2026-08-24T11-00-00-${threadId}_${rolloutId}.jsonl`,
+ ),
+ contents: encodeTranscriptRecord({
+ type: "session_meta",
+ payload: {
+ id: threadId,
+ cwd: workspace,
+ history_mode: "paginated",
+ history_base: { thread_id: threadId, end_ordinal_exclusive: 3, end_byte_offset: 512 },
+ },
+ }),
+ mtimeMs: nowMs,
+ });
+
+ const outcomes = yield* runRecentThreadOutcomes({
+ claudeHomePath,
+ codexHomePath,
+ workspaceRoot: workspace,
+ });
+
+ expect(outcomes).toEqual([{ _tag: "Skipped" }]);
+ }),
+ );
+
it.effect("skips excessive records without blocking an older valid transcript", () =>
Effect.gen(function* () {
const path = yield* Path.Path;
diff --git a/apps/server/src/project/AgentSessionScanner.ts b/apps/server/src/project/AgentSessionScanner.ts
index b12f347d7aa3..7a89f9c687de 100644
--- a/apps/server/src/project/AgentSessionScanner.ts
+++ b/apps/server/src/project/AgentSessionScanner.ts
@@ -447,7 +447,8 @@ export function parseAgentSessionTranscript(
input.source === "claudeAgent"
? records.reduce(foldClaudeFields, emptyClaudeFields)
: emptyClaudeFields;
- return parseAgentSessionRecords(input, records, claudeFields, 0);
+ const thread = parseAgentSessionRecords(input, records, claudeFields, 0);
+ return thread === "empty" ? null : thread;
}
function parseAgentSessionRecords(
@@ -455,7 +456,7 @@ function parseAgentSessionRecords(
records: ReadonlyArray,
claudeFields: ClaudeFields,
droppedMessageCount: number,
-): AgentSessionThread | null {
+): AgentSessionThread | "empty" | null {
const fallbackTimestamp = DateTime.formatIso(DateTime.makeUnsafe(input.lastActiveAtMs));
// Claude filenames are session IDs. Codex rollout filenames include extra
// timestamp text, so only transcript metadata can provide a resumable ID.
@@ -648,6 +649,7 @@ function parseAgentSessionRecords(
});
}
+ if (messages.length === 0 && droppedMessageCount === 0) return "empty";
const visibleMessages = messages.map(
({ codexResponseUser: _codexResponseUser, ...message }) => message,
);
@@ -2139,6 +2141,15 @@ export const make = Effect.gen(function* () {
snapshot.claudeFields,
snapshot.droppedMessageCount,
);
+ if (parsedThread === "empty") {
+ // null means an unreadable history base, which may hold the conversation.
+ if (candidate.source === "codex" && inheritedRecords === null) {
+ return Option.some({ _tag: "Skipped" });
+ }
+ // Never recorded as imported, so charging them would starve real sessions every run.
+ transcriptsRemaining += 1;
+ return Option.none();
+ }
if (parsedThread === null) {
return Option.some({ _tag: "Skipped" });
}
diff --git a/docs/user/welcome-page.md b/docs/user/welcome-page.md
index 63fa86b52b98..a91434969415 100644
--- a/docs/user/welcome-page.md
+++ b/docs/user/welcome-page.md
@@ -100,9 +100,11 @@ a time and skips files larger than 4 GiB. It also skips a conversation when the
message text and metadata it keeps need more than 32 MiB of memory. Tool output
does not count toward that limit. It skips a conversation with more than 100,000
records every time, so opening the page again does not import it. It ignores
-malformed records and skips unreadable or unparseable conversations.
+malformed records and conversations with no messages, such as a session opened
+in an editor but never used. It skips unreadable or unparseable conversations.
-Each import attempt reads up to 100 conversation files and 4 GiB per project. To
+Each import attempt reads up to 100 conversations and 4 GiB per project.
+Conversations with no messages count toward the 4 GiB but not the 100. To
continue a large batch, open the Welcome page again and pick the project in
**Review all**. Completed conversations are not imported again. You can continue
without the remaining history.