/** * @file transcript-sender.test.js * @description Unit tests for classifyTranscriptSender — the transcript viewer * must attribute each JSONL line to its TRUE sender, not blanket-label every * `type:"user"` line as the human. Cases mirror real Claude Code transcripts: * tool results, harness task-notifications, /loop (isMeta) re-injections, and a * subagent's orchestrator-assigned task. (Reported transcript mis-attribution.) * @author Nguyễn Ngọc Trí Vĩ */ const { describe, it, after } = require("node:test"); const assert = require("node:assert/strict"); const path = require("node:path"); const os = require("node:os"); const fs = require("node:fs"); // Point the db at a throwaway file before requiring the router (it pulls in db). const TEST_DB = path.join(os.tmpdir(), `transcript-sender-${Date.now()}-${process.pid}.db`); process.env.DASHBOARD_DB_PATH = TEST_DB; const { classifyTranscriptSender, patchCounts, askAnswers } = require("../routes/sessions"); const { db } = require("../db"); after(() => { if (db) db.close(); for (const s of ["", "-wal", "-shm"]) { try { fs.unlinkSync(TEST_DB + s); } catch { /* ignore */ } } }); // Shapes lifted from real ~/.claude transcripts. const realUser = (text) => ({ type: "user", message: { role: "user", content: text }, promptSource: "user_input", origin: "cli", }); const toolResult = () => ({ type: "user", message: { role: "user", content: [{ type: "tool_result", tool_use_id: "t1", content: "ok" }] }, toolUseResult: { stdout: "ok" }, }); const assistant = () => ({ type: "assistant", message: { role: "assistant", content: [] } }); describe("classifyTranscriptSender — main transcript", () => { it("real human message → user", () => { assert.equal(classifyTranscriptSender(realUser("spin up a team of agents"), false), "user"); }); it("tool result (toolUseResult / tool_result content) → tool", () => { assert.equal(classifyTranscriptSender(toolResult(), false), "tool"); }); it("harness task-notification → system", () => { const e = { type: "user", message: { role: "user", content: "\nabc\n" }, promptSource: "task_notification", origin: "system", }; assert.equal(classifyTranscriptSender(e, false), "system"); }); it("[SYSTEM NOTIFICATION …] background-task banner → system", () => { const e = { type: "user", message: { role: "user", content: "[SYSTEM NOTIFICATION - NOT USER INPUT]\nThis is an automated background-task event…\n…", }, }; assert.equal(classifyTranscriptSender(e, false), "system"); }); it("/loop re-injection (isMeta) → system", () => { const e = { type: "user", isMeta: true, message: { role: "user", content: "Sonnet cognition agent last one — stitch the brief" }, }; assert.equal(classifyTranscriptSender(e, false), "system"); }); it("assistant turn → assistant", () => { assert.equal(classifyTranscriptSender(assistant(), false), "assistant"); }); it("@agent mention typed by the human → user", () => { assert.equal(classifyTranscriptSender(realUser("@agent-ai-engineer hi"), false), "user"); }); it("local slash-command (type=system) → user", () => { assert.equal( classifyTranscriptSender( { type: "system", subtype: "local_command", content: "/color" }, false ), "user" ); }); }); describe("classifyTranscriptSender — subagent transcript", () => { it("orchestrator-assigned task (no promptSource/origin) → orchestrator", () => { const task = { type: "user", isSidechain: true, agentId: "a484", message: { role: "user", content: "Light research/synthesis task. Write a synthesis…" }, }; assert.equal(classifyTranscriptSender(task, true), "orchestrator"); }); it("tool result inside a subagent → tool", () => { assert.equal(classifyTranscriptSender(toolResult(), true), "tool"); }); it("human directly messaging the subagent (has provenance) → user", () => { const direct = { type: "user", isSidechain: true, agentId: "a484", message: { role: "user", content: "actually, focus on cognition" }, promptSource: "user_input", origin: "cli", }; assert.equal(classifyTranscriptSender(direct, true), "user"); }); it("assistant turn in a subagent → assistant", () => { assert.equal(classifyTranscriptSender(assistant(), true), "assistant"); }); }); describe("patchCounts — the +/- lines of a Write/Edit result", () => { it("counts a structuredPatch's added and removed lines, not its context", () => { const r = { filePath: "/w/a.ts", structuredPatch: [ { oldStart: 1, oldLines: 3, newStart: 1, newLines: 4, lines: [" a", "-b", "+c", "+d", " e"], }, { oldStart: 9, oldLines: 1, newStart: 10, newLines: 0, lines: ["-z"] }, ], }; assert.deepEqual(patchCounts(r), { added: 2, removed: 2 }); }); it("counts a created file's lines as added", () => { assert.deepEqual(patchCounts({ type: "create", content: "a\nb\nc\n", structuredPatch: [] }), { added: 3, removed: 0, }); }); it("answers null for a result that is no edit", () => { assert.equal(patchCounts({ stdout: "ok" }), null); assert.equal(patchCounts("text"), null); assert.equal(patchCounts(undefined), null); }); }); describe("askAnswers — what the user picked or typed for an AskUserQuestion", () => { it("returns each question's answer", () => { const r = { questions: [{ question: "Commit?", options: [{ label: "Yes" }, { label: "No" }] }], answers: { "Commit?": "Yes", "Notes?": "typed text" }, annotations: {}, }; assert.deepEqual(askAnswers(r), { "Commit?": "Yes", "Notes?": "typed text" }); }); it("drops a non-string answer and answers null for a result that is no question", () => { assert.deepEqual(askAnswers({ answers: { a: "x", b: 3 } }), { a: "x" }); assert.equal(askAnswers({ answers: {} }), null); assert.equal(askAnswers({ answers: ["x"] }), null); assert.equal(askAnswers({ stdout: "ok" }), null); assert.equal(askAnswers("text"), null); }); });