initial cost tracking

Signed-off-by: Eason Goodale <easong@openai.com>

codex-cli/tests/context-percent.test.ts

@@ -0,0 +1,28 @@
+import { describe, expect, it } from "vitest";
+
+import type { ResponseItem } from "openai/resources/responses/responses.mjs";
+
+import { calculateContextPercentRemaining } from "../src/components/chat/terminal-chat-utils.js";
+
+function makeUserMessage(id: string, text: string): ResponseItem {
+  return {
+    id,
+    type: "message",
+    role: "user",
+    content: [{ type: "input_text", text }],
+  } as ResponseItem;
+}
+
+describe("calculateContextPercentRemaining", () => {
+  it("includes extra context characters in calculation", () => {
+    const msgText = "a".repeat(40); // 40 chars → 10 tokens
+    const items = [makeUserMessage("1", msgText)];
+
+    const model = "gpt-4-16k";
+
+    const base = calculateContextPercentRemaining(items, model);
+    const withExtra = calculateContextPercentRemaining(items, model, 8); // +8 chars → +2 tokens
+
+    expect(withExtra).toBeLessThan(base);
+  });
+});

codex-cli/tests/estimate-cost.test.ts

@@ -0,0 +1,50 @@
+import { describe, expect, test } from "vitest";
+
+import { estimateCostUSD } from "../src/utils/estimate-cost.js";
+import { SessionCostTracker } from "../src/utils/session-cost.js";
+import type { ResponseItem } from "openai/resources/responses/responses.mjs";
+
+// Helper to craft a minimal ResponseItem for tests
+function makeMessage(
+  id: string,
+  role: "user" | "assistant",
+  text: string,
+): ResponseItem {
+  return {
+    id,
+    type: "message",
+    role,
+    content: [{ type: role === "user" ? "input_text" : "output_text", text }],
+  } as ResponseItem;
+}
+
+describe("estimateCostUSD", () => {
+  test("returns a proportional, positive estimate for known models", () => {
+    const items: Array<ResponseItem> = [
+      makeMessage("1", "user", "hello world"),
+      makeMessage("2", "assistant", "hi there"),
+    ];
+
+    const cost = estimateCostUSD(items, "gpt-3.5-turbo");
+    expect(cost).not.toBeNull();
+    expect(cost!).toBeGreaterThan(0);
+
+    // Adding another token should increase the estimate
+    const cost2 = estimateCostUSD(
+      items.concat([makeMessage("3", "user", "extra")]),
+      "gpt-3.5-turbo",
+    );
+    expect(cost2!).toBeGreaterThan(cost!);
+  });
+});
+
+describe("SessionCostTracker", () => {
+  test("accumulates items and reports tokens & cost", () => {
+    const tracker = new SessionCostTracker("gpt-3.5-turbo");
+    tracker.addItems([makeMessage("1", "user", "foo")]);
+    tracker.addItems([makeMessage("2", "assistant", "bar baz")]);
+
+    expect(tracker.getTokensUsed()).toBeGreaterThan(0);
+    expect(tracker.getCostUSD()!).toBeGreaterThan(0);
+  });
+});

codex-cli/tests/session-cost.test.ts

@@ -0,0 +1,79 @@
+import { afterEach, describe, expect, it, vi } from "vitest";
+
+import type { ResponseItem } from "openai/resources/responses/responses.mjs";
+
+import {
+  ensureSessionTracker,
+  getSessionTracker,
+  printAndResetSessionSummary,
+} from "../src/utils/session-cost.js";
+
+function makeMessage(id: string, role: "user" | "assistant", text: string): ResponseItem {
+  return {
+    id,
+    type: "message",
+    role,
+    content: [{ type: role === "user" ? "input_text" : "output_text", text }],
+  } as ResponseItem;
+}
+
+describe("printAndResetSessionSummary", () => {
+  afterEach(() => {
+    vi.restoreAllMocks();
+  });
+
+  it("/clear resets tracker so successive conversations start fresh", () => {
+    const spy = vi.spyOn(console, "log").mockImplementation(() => {});
+
+    const perSessionTokens: Array<number> = [];
+
+    for (let i = 1; i <= 3; i++) {
+      const tracker = ensureSessionTracker("gpt-3.5-turbo");
+      tracker.addTokens(i * 10); // 10, 20, 30
+      perSessionTokens.push(tracker.getTokensUsed());
+
+      // Simulate user typing /clear which prints & resets
+      printAndResetSessionSummary();
+
+      expect(getSessionTracker()).toBeNull();
+    }
+
+    expect(perSessionTokens).toEqual([10, 20, 30]);
+
+    spy.mockRestore();
+  });
+
+  it("prints a summary and resets the global tracker", () => {
+    const spy = vi.spyOn(console, "log").mockImplementation(() => {});
+
+    const tracker = ensureSessionTracker("gpt-3.5-turbo");
+    tracker.addItems([
+      makeMessage("1", "user", "hello"),
+      makeMessage("2", "assistant", "hi"),
+    ]);
+
+    printAndResetSessionSummary();
+
+    expect(spy).toHaveBeenCalled();
+    expect(getSessionTracker()).toBeNull();
+  });
+
+  it("prefers exact token counts added via addTokens() over heuristic", () => {
+    const tracker = ensureSessionTracker("gpt-3.5-turbo");
+
+    // Add a long message (heuristic would count >1 token)
+    tracker.addItems([
+      makeMessage("x", "user", "a".repeat(400)), // ~100 tokens
+    ]);
+
+    const heuristicTokens = tracker.getTokensUsed();
+    expect(heuristicTokens).toBeGreaterThan(50);
+
+    // Now inject an exact low token count and ensure it overrides
+    tracker.addTokens(10);
+    expect(tracker.getTokensUsed()).toBe(heuristicTokens + (10 - heuristicTokens));
+
+    const cost = tracker.getCostUSD();
+    expect(cost).not.toBeNull();
+  });
+});