fix(usage): Count Codex cached tokens once in token totals

plugin/statusline.sh

@@ -249,7 +249,8 @@ const local = localCost();
 const runUsage = getRunUsage();
 const orchestratorUsage = runUsage ? getOrchestratorUsage(runUsage) : undefined;
 const workerTokens = runUsage
-  ? runUsage.inputTokens + runUsage.outputTokens + runUsage.cachedTokens
+  ? nonNegativeNumber(runUsage.totalTokens) ??
+    runUsage.inputTokens + runUsage.outputTokens + runUsage.cachedTokens
   : undefined;
 const tokenField = () => {
   const total = workerTokens ?? (runUsage ? undefined : localTokens());

src/core/pricing.ts

@@ -179,3 +179,13 @@ export function computeSessionCost({
 export function cachedInInputFor(agent: string | undefined | null): boolean {
   return agent === "codex";
 }
+
+export function totalTokensFor(
+  agent: string | undefined | null,
+  usage?: SessionCostUsage,
+): number {
+  const inputTokens = usage?.inputTokens ?? 0;
+  const outputTokens = usage?.outputTokens ?? 0;
+  const cachedTokens = usage?.cachedTokens ?? 0;
+  return inputTokens + outputTokens + (cachedInInputFor(agent) ? 0 : cachedTokens);
+}

src/core/run-usage.ts

@@ -1,4 +1,4 @@
-import { cachedInInputFor, computeSessionCost } from "./pricing.js";
+import { cachedInInputFor, computeSessionCost, totalTokensFor } from "./pricing.js";
 import { isActiveStatus, type Session } from "./session.js";
 
 export interface RunAttribution {
@@ -17,6 +17,7 @@ export interface RunUsageSummary {
   inputTokens: number;
   outputTokens: number;
   cachedTokens: number;
+  totalTokens: number;
   costUsd: number;
   sessionCount: number;
   activeSessionCount: number;
@@ -28,6 +29,7 @@ export interface RunUsageSummary {
     inputTokens: number;
     outputTokens: number;
     cachedTokens: number;
+    totalTokens: number;
     sources: Array<{ nativeId: string; costUsd: number }>;
   };
   total: { costUsd: number };
@@ -56,6 +58,7 @@ export function aggregateRunUsage(
     inputTokens: 0,
     outputTokens: 0,
     cachedTokens: 0,
+    totalTokens: 0,
     costUsd: 0,
     sessionCount: 0,
     activeSessionCount: 0,
@@ -67,6 +70,7 @@ export function aggregateRunUsage(
       inputTokens: 0,
       outputTokens: 0,
       cachedTokens: 0,
+      totalTokens: 0,
       sources: [],
     },
     total: { costUsd: 0 },
@@ -79,6 +83,7 @@ export function aggregateRunUsage(
     target.inputTokens += usage?.inputTokens ?? 0;
     target.outputTokens += usage?.outputTokens ?? 0;
     target.cachedTokens += usage?.cachedTokens ?? 0;
+    target.totalTokens += totalTokensFor(session.agent, usage);
 
     const costUsd = computeSessionCost({
       model: session.model,

src/store/sessions.ts

@@ -1,7 +1,7 @@
 import type { DatabaseSync } from "node:sqlite";
 import { type Session, type SessionStatus, type AgentId, isActiveStatus } from "../core/session.js";
 import type { FailureInfo } from "../core/errors.js";
-import { cachedInInputFor, computeSessionCost } from "../core/pricing.js";
+import { cachedInInputFor, computeSessionCost, totalTokensFor } from "../core/pricing.js";
 import type { UsageQueryParams, UsageQueryResult, UsageMetricBucket, UsageTotals } from "../daemon/protocol.js";
 
 export interface SessionRow {
@@ -443,7 +443,7 @@ export class SessionStore {
       const inputTokens = row.usage_input_tokens ?? 0;
       const outputTokens = row.usage_output_tokens ?? 0;
       const cachedTokens = row.usage_cached_tokens ?? 0;
-      const totalTokens = inputTokens + outputTokens + cachedTokens;
+      const totalTokens = totalTokensFor(row.agent, { inputTokens, outputTokens, cachedTokens });
 
       const calculatedCost = computeSessionCost({
         model: row.model,
@@ -522,7 +522,7 @@ export class SessionStore {
       const inputTokens = row.input_tokens;
       const outputTokens = row.output_tokens;
       const cachedTokens = row.cached_tokens;
-      const totalTokens = inputTokens + outputTokens + cachedTokens;
+      const totalTokens = totalTokensFor("claude", { inputTokens, outputTokens, cachedTokens });
       const cost = row.cost;
 
       totals.sessionCount++;

tests/pricing.test.ts

@@ -1,5 +1,11 @@
 import { describe, expect, it } from "vitest";
-import { cachedInInputFor, computeSessionCost, MODEL_PRICES, resolveModelPrice } from "../src/core/pricing.js";
+import {
+  cachedInInputFor,
+  computeSessionCost,
+  MODEL_PRICES,
+  resolveModelPrice,
+  totalTokensFor,
+} from "../src/core/pricing.js";
 
 describe("computeSessionCost", () => {
   it("uses a valid reported cost, including zero, without recalculating", () => {
@@ -242,3 +248,12 @@ describe("computeSessionCost", () => {
     expect(resolveModelPrice("CLAUDE-SONNET-4-6")).toEqual(MODEL_PRICES["claude-sonnet-4-6"]);
   });
 });
+
+describe("totalTokensFor", () => {
+  it("counts cached tokens once for Codex and separately for Claude", () => {
+    const usage = { inputTokens: 1_000, outputTokens: 100, cachedTokens: 800 };
+
+    expect(totalTokensFor("codex", usage)).toBe(1_100);
+    expect(totalTokensFor("claude", usage)).toBe(1_900);
+  });
+});

tests/statusline.test.ts

@@ -265,6 +265,49 @@ describe("Claude statusline", () => {
     expect(stripAnsi(small.output)).toContain("980 tok");
   });
 
+  it("uses a valid run totalTokens value for the tok field", async () => {
+    const result = await render({
+      payload: payload(0),
+      runId: "run-cached-tokens",
+      usage: {
+        runId: "run-cached-tokens",
+        inputTokens: 1_000,
+        outputTokens: 100,
+        cachedTokens: 800,
+        totalTokens: 1_100,
+        costUsd: 0,
+        sessionCount: 1,
+        activeSessionCount: 1,
+        costComplete: true,
+        sessionsWithoutCost: 0,
+      },
+    });
+
+    expect(stripAnsi(result.output)).toContain("1.1k tok");
+    expect(stripAnsi(result.output)).not.toContain("1.9k tok");
+  });
+
+  it("falls back to the token sum when totalTokens is invalid", async () => {
+    const result = await render({
+      payload: payload(0),
+      runId: "run-invalid-token-total",
+      usage: {
+        runId: "run-invalid-token-total",
+        inputTokens: 1_000,
+        outputTokens: 100,
+        cachedTokens: 800,
+        totalTokens: -1,
+        costUsd: 0,
+        sessionCount: 1,
+        activeSessionCount: 1,
+        costComplete: true,
+        sessionsWithoutCost: 0,
+      },
+    });
+
+    expect(stripAnsi(result.output)).toContain("1.9k tok");
+  });
+
   it("marks a partial aggregate even when the local cost is zero", async () => {
     const result = await render({
       payload: payload(0),

tests/usage-cli.test.ts

@@ -18,6 +18,7 @@ const runSummary = {
   inputTokens: 100,
   outputTokens: 20,
   cachedTokens: 5,
+  totalTokens: 125,
   costUsd: 0.25,
   sessionCount: 1,
   activeSessionCount: 0,
@@ -29,6 +30,7 @@ const runSummary = {
     inputTokens: 0,
     outputTokens: 0,
     cachedTokens: 0,
+    totalTokens: 0,
     sources: [],
   },
   total: { costUsd: 0.25 },

tests/usage-daemon.test.ts

@@ -93,6 +93,7 @@ describe("usage daemon methods", () => {
       inputTokens: 1_300,
       outputTokens: 850,
       cachedTokens: 320,
+      totalTokens: 2_170,
       costUsd: 0.5,
       sessionCount: 2,
       activeSessionCount: 1,
@@ -104,6 +105,7 @@ describe("usage daemon methods", () => {
         inputTokens: 0,
         outputTokens: 0,
         cachedTokens: 0,
+        totalTokens: 0,
         sources: [],
       },
       total: { costUsd: 0.5 },

tests/usage-query.test.ts

@@ -125,6 +125,40 @@ describe("SessionStore.queryUsage", () => {
     expect(worker).toMatchObject({ sessionCount: 2, costUsd: 1.25 });
   });
 
+  it("uses harness-aware totals in every usage bucket", () => {
+    store.create(
+      makeSession("codex", {
+        agent: "codex",
+        model: "gpt-5.6-luna",
+        repository: "/dev/codedeck",
+        origin: null,
+        usage: { inputTokens: 1_000, outputTokens: 100, cachedTokens: 800 },
+      }),
+    );
+    store.create(
+      makeSession("claude", {
+        agent: "claude",
+        model: "claude-sonnet-4-6",
+        repository: "/dev/codedeck",
+        origin: "open",
+        usage: { inputTokens: 1_000, outputTokens: 100, cachedTokens: 800 },
+      }),
+    );
+
+    const result = store.queryUsage({ period: "all" });
+
+    expect(result.totals.totalTokens).toBe(3_000);
+    expect(result.byDay[0]?.totalTokens).toBe(3_000);
+    expect(result.byRepository.find((bucket) => bucket.key === "codedeck")?.totalTokens).toBe(3_000);
+    expect(result.byModel.find((bucket) => bucket.key === "gpt-5.6-luna")?.totalTokens).toBe(1_100);
+    expect(result.byModel.find((bucket) => bucket.key === "claude-sonnet-4-6")?.totalTokens).toBe(1_900);
+    expect(result.byAgent.find((bucket) => bucket.key === "codex")?.totalTokens).toBe(1_100);
+    expect(result.byAgent.find((bucket) => bucket.key === "claude")?.totalTokens).toBe(1_900);
+    expect(result.byRun.find((bucket) => bucket.key === "run-1")?.totalTokens).toBe(3_000);
+    expect(result.byOrigin.find((bucket) => bucket.key === "worker")?.totalTokens).toBe(1_100);
+    expect(result.byOrigin.find((bucket) => bucket.key === "orchestrator")?.totalTokens).toBe(1_900);
+  });
+
   it("merges in-range legacy usage into analytics without listing it as a session", () => {
     const endedAt = new Date(2026, 8, 7, 12).toISOString();
     const outsideRange = new Date(2026, 7, 31, 12).toISOString();

tests/usage.test.ts

@@ -31,6 +31,7 @@ describe("aggregateRunUsage", () => {
       inputTokens: 1_300,
       outputTokens: 850,
       cachedTokens: 320,
+      totalTokens: 2_150,
       costUsd: 0.5,
       sessionCount: 2,
       activeSessionCount: 0,
@@ -42,6 +43,7 @@ describe("aggregateRunUsage", () => {
         inputTokens: 0,
         outputTokens: 0,
         cachedTokens: 0,
+        totalTokens: 0,
         sources: [],
       },
       total: { costUsd: 0.5 },
@@ -65,6 +67,7 @@ describe("aggregateRunUsage", () => {
       inputTokens: 300,
       outputTokens: 150,
       cachedTokens: 30,
+      totalTokens: 450,
       costUsd: 0.42,
       sessionCount: 2,
       activeSessionCount: 0,
@@ -76,6 +79,7 @@ describe("aggregateRunUsage", () => {
         inputTokens: 0,
         outputTokens: 0,
         cachedTokens: 0,
+        totalTokens: 0,
         sources: [],
       },
       total: { costUsd: 0.42 },
@@ -142,6 +146,7 @@ describe("aggregateRunUsage", () => {
       inputTokens: 0,
       outputTokens: 0,
       cachedTokens: 0,
+      totalTokens: 0,
       costUsd: 0,
       sessionCount: 0,
       activeSessionCount: 0,
@@ -153,6 +158,7 @@ describe("aggregateRunUsage", () => {
         inputTokens: 0,
         outputTokens: 0,
         cachedTokens: 0,
+        totalTokens: 0,
         sources: [],
       },
       total: { costUsd: 0 },
@@ -184,6 +190,7 @@ describe("aggregateRunUsage", () => {
       inputTokens: 0,
       outputTokens: 0,
       cachedTokens: 0,
+      totalTokens: 0,
       costUsd: 0.5,
       sessionCount: 2,
       activeSessionCount: 0,
@@ -195,6 +202,7 @@ describe("aggregateRunUsage", () => {
         inputTokens: 10,
         outputTokens: 20,
         cachedTokens: 3,
+        totalTokens: 33,
         sources: [
           { nativeId: "X", costUsd: 3 },
           { nativeId: "Y", costUsd: 0.8 },
@@ -250,4 +258,21 @@ describe("aggregateRunUsage", () => {
     expect(summary.total.costUsd).toBe(1);
     expect(summary.costComplete).toBe(true);
   });
+
+  it("counts cached tokens once for Codex workers and separately for Claude orchestrators", () => {
+    const summary = aggregateRunUsage("run-tokens", [
+      makeSession("codex", {
+        agent: "codex",
+        usage: { inputTokens: 1_000, outputTokens: 100, cachedTokens: 800, cost: 0 },
+      }),
+      makeSession("claude", {
+        agent: "claude",
+        origin: "open",
+        usage: { inputTokens: 1_000, outputTokens: 100, cachedTokens: 800, cost: 0 },
+      }),
+    ]);
+
+    expect(summary.totalTokens).toBe(1_100);
+    expect(summary.orchestrator.totalTokens).toBe(1_900);
+  });
 });