fix(status): account cached prompt tokens

src/agents/agents.test.ts

@@ -36,12 +36,17 @@ describe("pi agent helpers", () => {
   it("parses final assistant message and preserves usage meta", () => {
     const stdout = [
       '{"type":"message_start","message":{"role":"assistant"}}',
-      '{"type":"message_end","message":{"role":"assistant","content":[{"type":"text","text":"hello world"}],"usage":{"input":10,"output":5},"model":"pi-1","provider":"inflection","stopReason":"end"}}',
+      '{"type":"message_end","message":{"role":"assistant","content":[{"type":"text","text":"hello world"}],"usage":{"input":10,"output":5,"cacheRead":100,"cacheWrite":20,"totalTokens":135},"model":"pi-1","provider":"inflection","stopReason":"end"}}',
     ].join("\n");
     const parsed = piSpec.parseOutput(stdout);
     expect(parsed.texts?.[0]).toBe("hello world");
     expect(parsed.meta?.provider).toBe("inflection");
     expect((parsed.meta?.usage as { output?: number })?.output).toBe(5);
+    expect((parsed.meta?.usage as { cacheRead?: number })?.cacheRead).toBe(100);
+    expect((parsed.meta?.usage as { cacheWrite?: number })?.cacheWrite).toBe(
+      20,
+    );
+    expect((parsed.meta?.usage as { total?: number })?.total).toBe(135);
   });
 
   it("piSpec carries tool names when present", () => {

src/agents/pi.ts

@@ -6,11 +6,12 @@ import type {
   AgentSpec,
   AgentToolResult,
 } from "./types.js";
+import { normalizeUsage, type UsageLike } from "./usage.js";
 
 type PiAssistantMessage = {
   role?: string;
   content?: Array<{ type?: string; text?: string }>;
-  usage?: { input?: number; output?: number };
+  usage?: UsageLike;
   model?: string;
   provider?: string;
   stopReason?: string;
@@ -153,7 +154,7 @@ function parsePiJson(raw: string): AgentParseResult {
         model: lastAssistant.model,
         provider: lastAssistant.provider,
         stopReason: lastAssistant.stopReason,
-        usage: lastAssistant.usage,
+        usage: normalizeUsage(lastAssistant.usage),
       }
     : undefined;
 

src/agents/usage.ts

@@ -0,0 +1,69 @@
+export type UsageLike = {
+  input?: number;
+  output?: number;
+  cacheRead?: number;
+  cacheWrite?: number;
+  total?: number;
+  // Some agents/logs emit alternate naming.
+  totalTokens?: number;
+  total_tokens?: number;
+  cache_read?: number;
+  cache_write?: number;
+};
+
+const asFiniteNumber = (value: unknown): number | undefined => {
+  if (typeof value !== "number") return undefined;
+  if (!Number.isFinite(value)) return undefined;
+  return value;
+};
+
+export function normalizeUsage(raw?: UsageLike | null):
+  | {
+      input?: number;
+      output?: number;
+      cacheRead?: number;
+      cacheWrite?: number;
+      total?: number;
+    }
+  | undefined {
+  if (!raw) return undefined;
+
+  const input = asFiniteNumber(raw.input);
+  const output = asFiniteNumber(raw.output);
+  const cacheRead = asFiniteNumber(raw.cacheRead ?? raw.cache_read);
+  const cacheWrite = asFiniteNumber(raw.cacheWrite ?? raw.cache_write);
+  const total = asFiniteNumber(
+    raw.total ?? raw.totalTokens ?? raw.total_tokens,
+  );
+
+  if (
+    input === undefined &&
+    output === undefined &&
+    cacheRead === undefined &&
+    cacheWrite === undefined &&
+    total === undefined
+  ) {
+    return undefined;
+  }
+
+  return {
+    input,
+    output,
+    cacheRead,
+    cacheWrite,
+    total,
+  };
+}
+
+export function derivePromptTokens(usage?: {
+  input?: number;
+  cacheRead?: number;
+  cacheWrite?: number;
+}): number | undefined {
+  if (!usage) return undefined;
+  const input = usage.input ?? 0;
+  const cacheRead = usage.cacheRead ?? 0;
+  const cacheWrite = usage.cacheWrite ?? 0;
+  const sum = input + cacheRead + cacheWrite;
+  return sum > 0 ? sum : undefined;
+}

src/auto-reply/reply.ts

@@ -822,12 +822,15 @@ export async function getReplyFromConfig(
           if (entry) {
             const input = usage.input ?? 0;
             const output = usage.output ?? 0;
-            const total = usage.total ?? input + output;
+            const promptTokens =
+              input + (usage.cacheRead ?? 0) + (usage.cacheWrite ?? 0);
             sessionEntry = {
               ...entry,
-              inputTokens: (entry.inputTokens ?? 0) + input,
-              outputTokens: (entry.outputTokens ?? 0) + output,
-              totalTokens: (entry.totalTokens ?? 0) + total,
+              inputTokens: input,
+              outputTokens: output,
+              // Track the effective prompt/context size (cached + uncached input).
+              totalTokens:
+                promptTokens > 0 ? promptTokens : (usage.total ?? input),
               model,
               contextTokens: contextTokens ?? entry.contextTokens,
               updatedAt: Date.now(),

src/auto-reply/status.test.ts

@@ -1,5 +1,7 @@
+import fs from "node:fs";
+import os from "node:os";
+import path from "node:path";
 import { afterEach, describe, expect, it, vi } from "vitest";
-
 import { buildStatusMessage } from "./status.js";
 
 afterEach(() => {
@@ -60,4 +62,55 @@ describe("buildStatusMessage", () => {
     expect(text).toContain("Context:");
     expect(text).toContain("Web: not linked");
   });
+
+  it("prefers cached prompt tokens from the session log", () => {
+    const dir = fs.mkdtempSync(path.join(os.tmpdir(), "clawdis-status-"));
+    const storePath = path.join(dir, "sessions.json");
+    const sessionId = "sess-1";
+    const logPath = path.join(dir, `${sessionId}.jsonl`);
+
+    fs.writeFileSync(
+      logPath,
+      [
+        JSON.stringify({
+          type: "message",
+          message: {
+            role: "assistant",
+            model: "claude-opus-4-5",
+            usage: {
+              input: 1,
+              output: 2,
+              cacheRead: 1000,
+              cacheWrite: 0,
+              totalTokens: 1003,
+            },
+          },
+        }),
+      ].join("\n"),
+      "utf-8",
+    );
+
+    const text = buildStatusMessage({
+      reply: {
+        mode: "command",
+        command: ["echo", "{{Body}}"],
+        agent: { kind: "pi", model: "claude-opus-4-5", contextTokens: 32_000 },
+        session: { scope: "per-sender" },
+      },
+      sessionEntry: {
+        sessionId,
+        updatedAt: 0,
+        totalTokens: 3, // would be wrong if cached prompt tokens exist
+        contextTokens: 32_000,
+      },
+      sessionKey: "main",
+      sessionScope: "per-sender",
+      storePath,
+      webLinked: true,
+    });
+
+    expect(text).toContain("Context: 1.0k/32k");
+
+    fs.rmSync(dir, { recursive: true, force: true });
+  });
 });

src/auto-reply/status.ts

@@ -5,6 +5,11 @@ import path from "node:path";
 
 import { lookupContextTokens } from "../agents/context.js";
 import { DEFAULT_CONTEXT_TOKENS, DEFAULT_MODEL } from "../agents/defaults.js";
+import {
+  derivePromptTokens,
+  normalizeUsage,
+  type UsageLike,
+} from "../agents/usage.js";
 import type { ClawdisConfig } from "../config/config.js";
 import type { SessionEntry, SessionScope } from "../config/sessions.js";
 import type { ThinkLevel, VerboseLevel } from "./thinking.js";
@@ -117,6 +122,7 @@ const readUsageFromSessionLog = (
   | {
       input: number;
       output: number;
+      promptTokens: number;
       total: number;
       model?: string;
     }
@@ -144,33 +150,38 @@ const readUsageFromSessionLog = (
     const lines = fs.readFileSync(logPath, "utf-8").split(/\n+/);
     let input = 0;
     let output = 0;
+    let promptTokens = 0;
     let model: string | undefined;
+    let lastUsage: ReturnType<typeof normalizeUsage> | undefined;
 
     for (const line of lines) {
       if (!line.trim()) continue;
       try {
         const parsed = JSON.parse(line) as {
           message?: {
-            usage?: { input?: number; output?: number; total?: number };
+            usage?: UsageLike;
             model?: string;
           };
-          usage?: { input?: number; output?: number; total?: number };
+          usage?: UsageLike;
           model?: string;
         };
-        const usage = parsed.message?.usage ?? parsed.usage;
-        if (usage) {
-          input += usage.input ?? 0;
-          output += usage.output ?? 0;
-        }
+        const usageRaw = parsed.message?.usage ?? parsed.usage;
+        const usage = normalizeUsage(usageRaw);
+        if (usage) lastUsage = usage;
         model = parsed.message?.model ?? parsed.model ?? model;
       } catch {
         // ignore bad lines
       }
     }
 
-    const total = input + output;
-    if (total === 0) return undefined;
-    return { input, output, total, model };
+    if (!lastUsage) return undefined;
+    input = lastUsage.input ?? 0;
+    output = lastUsage.output ?? 0;
+    promptTokens =
+      derivePromptTokens(lastUsage) ?? lastUsage.total ?? input + output;
+    const total = lastUsage.total ?? promptTokens + output;
+    if (promptTokens === 0 && total === 0) return undefined;
+    return { input, output, promptTokens, total, model };
   } catch {
     return undefined;
   }
@@ -190,15 +201,17 @@ export function buildStatusMessage(args: StatusArgs): string {
     entry?.totalTokens ??
     (entry?.inputTokens ?? 0) + (entry?.outputTokens ?? 0);
 
-  // Fallback: derive usage from the session transcript if the store lacks it
-  if (!totalTokens || totalTokens === 0) {
-    const logUsage = readUsageFromSessionLog(entry?.sessionId, args.storePath);
-    if (logUsage) {
-      totalTokens = logUsage.total;
-      if (!model) model = logUsage.model ?? model;
-      if (!contextTokens && logUsage.model) {
-        contextTokens = lookupContextTokens(logUsage.model) ?? contextTokens;
-      }
+  // Prefer prompt-size tokens from the session transcript when it looks larger
+  // (cached prompt tokens are often missing from agent meta/store).
+  const logUsage = readUsageFromSessionLog(entry?.sessionId, args.storePath);
+  if (logUsage) {
+    const candidate = logUsage.promptTokens || logUsage.total;
+    if (!totalTokens || totalTokens === 0 || candidate > totalTokens) {
+      totalTokens = candidate;
+    }
+    if (!model) model = logUsage.model ?? model;
+    if (!contextTokens && logUsage.model) {
+      contextTokens = lookupContextTokens(logUsage.model) ?? contextTokens;
     }
   }
   const agentProbe = probeAgentCommand(args.reply?.command);