DeepTutor: Lifelong Personalized Tutoring. https://deeptutor.info/.
Python ai-agentsai-tutorclawbotcli-tooldeepresearchinteractive-learninglearningmulti-agent-systemsrag
import { File as NodeFile } from "node:buffer";
import { describe, expect, it } from "vitest";
import { parseCodexSession } from "@/lib/chat-import/codex";
import type { SessionRef } from "@/lib/chat-import/types";
function sessionRef(records: unknown[]): SessionRef {
const jsonl = records.map((record) => JSON.stringify(record)).join("\n");
const file = new NodeFile([jsonl], "rollout-test.jsonl", {
lastModified: Date.parse("2026-06-25T12:00:00Z"),
});
return {
externalId: "session-1",
provisionalTitle: "",
cwd: "/workspace/project",
date: "2026-06-25",
lastModified: file.lastModified,
sizeBytes: file.size,
handle: {
getFile: async () => file as unknown as File,
} as FileSystemFileHandle,
};
}
describe("parseCodexSession", () => {
it("parses current response_item message records", async () => {
const ref = sessionRef([
{
timestamp: "2026-06-25T10:00:00Z",
type: "session_meta",
payload: { id: "session-1", cwd: "/workspace/project", thread_source: "user" },
},
{
timestamp: "2026-06-25T10:00:01Z",
type: "response_item",
payload: {
type: "message",
role: "developer",
content: [{ type: "input_text", text: "hidden instructions" }],
},
},
{
timestamp: "2026-06-25T10:00:02Z",
type: "response_item",
payload: {
type: "message",
role: "user",
content: [{ type: "input_text", text: "Explain Fourier transforms" }],
},
},
{
timestamp: "2026-06-25T10:00:03Z",
type: "response_item",
payload: {
type: "message",
role: "assistant",
content: [
{ type: "output_text", text: "A Fourier transform" },
{ type: "output_text", text: "decomposes a signal." },
],
},
},
]);
const parsed = await parseCodexSession(ref);
expect(parsed).not.toBeNull();
expect(parsed?.title).toBe("Explain Fourier transforms");
expect(parsed?.messages).toEqual([
{
role: "user",
content: "Explain Fourier transforms",
created_at: Date.parse("2026-06-25T10:00:02Z") / 1000,
},
{
role: "assistant",
content: "A Fourier transform\n\ndecomposes a signal.",
created_at: Date.parse("2026-06-25T10:00:03Z") / 1000,
},
]);
});
it("prefers legacy event messages when both storage layers are present", async () => {
const ref = sessionRef([
{
timestamp: "2026-06-25T10:00:00Z",
type: "session_meta",
payload: { cwd: "/workspace/project", thread_source: "user" },
},
{
timestamp: "2026-06-25T10:00:01Z",
type: "response_item",
payload: {
type: "message",
role: "user",
content: [{ type: "input_text", text: "duplicate user" }],
},
},
{
timestamp: "2026-06-25T10:00:01Z",
type: "event_msg",
payload: { type: "user_message", message: "legacy user" },
},
{
timestamp: "2026-06-25T10:00:02Z",
type: "response_item",
payload: {
type: "message",
role: "assistant",
content: [{ type: "output_text", text: "duplicate assistant" }],
},
},
{
timestamp: "2026-06-25T10:00:02Z",
type: "event_msg",
payload: { type: "agent_message", message: "legacy assistant" },
},
]);
const parsed = await parseCodexSession(ref);
expect(parsed?.messages.map((message) => message.content)).toEqual([
"legacy user",
"legacy assistant",
]);
});
// Codex delivers its own context to the model as `role: "user"` items. The
// session this came from had exactly one user turn — a 4KB plugin catalogue
// — so the import gave the learner a message they never wrote and a title
// made of it (#1354).
it("drops the harness's own injected turns, whatever they are tagged", async () => {
const injected = (text: string) => ({
timestamp: "2026-06-25T10:00:01Z",
type: "response_item",
payload: {
type: "message",
role: "user",
content: [{ type: "input_text", text }],
},
});
const ref = sessionRef([
{
timestamp: "2026-06-25T10:00:00Z",
type: "session_meta",
payload: { id: "session-1", cwd: "/workspace/project", thread_source: "user" },
},
injected("<recommended_plugins>\nAirtable, Alpaca, …\n</recommended_plugins>\n<environment_context>\n <shell>zsh</shell>\n</environment_context>"),
injected('<codex_internal_context source="goal">\nContinue the thread goal.\n</codex_internal_context>'),
injected("<turn_aborted>\nThe user interrupted the previous turn.\n</turn_aborted>"),
{
timestamp: "2026-06-25T10:00:02Z",
type: "response_item",
payload: {
type: "message",
role: "user",
content: [{ type: "input_text", text: "Explain Fourier transforms" }],
},
},
]);
const parsed = await parseCodexSession(ref);
expect(parsed?.messages.map((message) => message.content)).toEqual([
"Explain Fourier transforms",
]);
expect(parsed?.title).toBe("Explain Fourier transforms");
});
// The most common machine prefix of all, and the one that must survive: an
// attachment turn is `<image …></image>` followed by the actual question.
it("keeps a turn whose markup is only a prefix to what the person typed", async () => {
const ref = sessionRef([
{
timestamp: "2026-06-25T10:00:00Z",
type: "session_meta",
payload: { id: "session-1", cwd: "/workspace/project", thread_source: "user" },
},
{
timestamp: "2026-06-25T10:00:01Z",
type: "response_item",
payload: {
type: "message",
role: "user",
content: [
{
type: "input_text",
text: "<image name=[Image #1]></image>[Image #1] Can you read this chart?",
},
],
},
},
]);
const parsed = await parseCodexSession(ref);
expect(parsed?.messages).toHaveLength(1);
expect(parsed?.messages[0].content).toContain("Can you read this chart?");
});
// A person pasting markup with no prose is the false positive this trades
// away, so keep the trade narrow: anything that does not close cleanly, or
// that has a single character outside the elements, stays.
it("keeps markup that does not resolve into whole top-level elements", async () => {
const ref = sessionRef([
{
timestamp: "2026-06-25T10:00:00Z",
type: "session_meta",
payload: { id: "session-1", cwd: "/workspace/project", thread_source: "user" },
},
{
timestamp: "2026-06-25T10:00:01Z",
type: "response_item",
payload: {
type: "message",
role: "user",
content: [{ type: "input_text", text: "<config>\n <port>8080</port>\nwhy does this fail?" }],
},
},
]);
const parsed = await parseCodexSession(ref);
expect(parsed?.messages[0].content).toContain("why does this fail?");
});
});