feat: add token counts and cost to audit log, report usage on all Cla…

.github/workflows/claude.yml

@@ -57,6 +57,7 @@ jobs:
           bun-version: "1.3.12"
 
       - name: Run Claude
+        id: claude
         if: steps.prompt.outputs.skip != 'true'
         uses: anthropics/claude-code-action@v1
         with:
@@ -78,6 +79,17 @@ jobs:
               }
             }
 
+      - name: Report token usage
+        if: steps.claude.outcome != 'skipped' && steps.claude.outputs.execution_file != ''
+        continue-on-error: true
+        env:
+          GITHUB_TOKEN: ${{ secrets.AGENT_TOKEN }}
+          EXEC_FILE: ${{ steps.claude.outputs.execution_file }}
+          ENTITY_TYPE: issue
+          ENTITY_NUMBER: ${{ github.event.issue.number }}
+          REPO: ${{ github.repository }}
+        run: bash scripts/report-token-usage.sh
+
   # ---------------------------------------------------------------------------
   # Comment on a plain issue → update the related PR
   # ---------------------------------------------------------------------------
@@ -117,6 +129,7 @@ jobs:
           bun-version: "1.3.12"
 
       - name: Run Claude
+        id: claude
         if: steps.prompt.outputs.skip != 'true'
         uses: anthropics/claude-code-action@v1
         with:
@@ -138,6 +151,17 @@ jobs:
               }
             }
 
+      - name: Report token usage
+        if: steps.claude.outcome != 'skipped' && steps.claude.outputs.execution_file != ''
+        continue-on-error: true
+        env:
+          GITHUB_TOKEN: ${{ secrets.AGENT_TOKEN }}
+          EXEC_FILE: ${{ steps.claude.outputs.execution_file }}
+          ENTITY_TYPE: issue
+          ENTITY_NUMBER: ${{ github.event.issue.number }}
+          REPO: ${{ github.repository }}
+        run: bash scripts/report-token-usage.sh
+
   # ---------------------------------------------------------------------------
   # PR opened/updated (not Claude's) → post a review
   # ---------------------------------------------------------------------------
@@ -264,6 +288,17 @@ jobs:
           GH_TOKEN: ${{ secrets.AGENT_TOKEN }}
           PR_NUMBER: ${{ github.event_name == 'pull_request' && github.event.pull_request.number || github.event.workflow_run.pull_requests[0].number }}
 
+      - name: Report token usage
+        if: steps.skip-check.outputs.skip != 'true' && steps.claude.outcome != 'skipped' && steps.claude.outputs.execution_file != ''
+        continue-on-error: true
+        env:
+          GITHUB_TOKEN: ${{ secrets.AGENT_TOKEN }}
+          EXEC_FILE: ${{ steps.claude.outputs.execution_file }}
+          ENTITY_TYPE: pr
+          ENTITY_NUMBER: ${{ github.event_name == 'pull_request' && github.event.pull_request.number || github.event.workflow_run.pull_requests[0].number }}
+          REPO: ${{ github.repository }}
+        run: bash scripts/report-token-usage.sh
+
   # ---------------------------------------------------------------------------
   # PR synchronize (new commits pushed) → clear status, check prior concerns
   # ---------------------------------------------------------------------------
@@ -314,6 +349,7 @@ jobs:
           bun-version: "1.3.12"
 
       - name: Run Claude
+        id: claude
         if: steps.prompt.outputs.skip != 'true'
         uses: anthropics/claude-code-action@v1
         with:
@@ -328,6 +364,17 @@ jobs:
               }
             }
 
+      - name: Report token usage
+        if: steps.claude.outcome != 'skipped' && steps.claude.outputs.execution_file != ''
+        continue-on-error: true
+        env:
+          GITHUB_TOKEN: ${{ secrets.AGENT_TOKEN }}
+          EXEC_FILE: ${{ steps.claude.outputs.execution_file }}
+          ENTITY_TYPE: pr
+          ENTITY_NUMBER: ${{ github.event.pull_request.number }}
+          REPO: ${{ github.repository }}
+        run: bash scripts/report-token-usage.sh
+
   # ---------------------------------------------------------------------------
   # Comment on Claude's own PR → address the feedback
   # Handles both timeline comments (issue_comment) and inline diff comments
@@ -380,6 +427,7 @@ jobs:
           bun-version: "1.3.12"
 
       - name: Run Claude
+        id: claude
         if: steps.prompt.outputs.skip != 'true'
         uses: anthropics/claude-code-action@v1
         with:
@@ -400,3 +448,14 @@ jobs:
                 ]
               }
             }
+
+      - name: Report token usage
+        if: steps.claude.outcome != 'skipped' && steps.claude.outputs.execution_file != ''
+        continue-on-error: true
+        env:
+          GITHUB_TOKEN: ${{ secrets.AGENT_TOKEN }}
+          EXEC_FILE: ${{ steps.claude.outputs.execution_file }}
+          ENTITY_TYPE: issue
+          ENTITY_NUMBER: ${{ github.event.issue.number || github.event.pull_request.number }}
+          REPO: ${{ github.repository }}
+        run: bash scripts/report-token-usage.sh

packages/agent/src/__tests__/agent.test.ts

@@ -24,7 +24,7 @@ describe("runAgent", () => {
       })()
     );
 
-    const result = await runAgent("find the Q4 report", "You are Squirrel.");
+    const { result } = await runAgent("find the Q4 report", "You are Squirrel.");
     expect(result).toBe("Here is the document you requested.");
   });
 
@@ -35,7 +35,7 @@ describe("runAgent", () => {
       })()
     );
 
-    const result = await runAgent("find something", "system");
+    const { result } = await runAgent("find something", "system");
     expect(result).toBe("I was unable to generate a response.");
   });
 
@@ -47,10 +47,51 @@ describe("runAgent", () => {
       })()
     );
 
-    const result = await runAgent("prompt", "system");
+    const { result } = await runAgent("prompt", "system");
     expect(result).toBe("second");
   });
 
+  it("captures token usage and cost from the result message", async () => {
+    vi.mocked(query).mockReturnValue(
+      (async function* () {
+        yield {
+          result: "answer",
+          total_cost_usd: 0.0042,
+          // real SDKResultMessage.usage uses the API's snake_case keys
+          usage: {
+            input_tokens: 800,
+            output_tokens: 300,
+            cache_read_input_tokens: 150,
+            cache_creation_input_tokens: 50,
+          },
+        };
+      })()
+    );
+
+    const agentResult = await runAgent("prompt", "system");
+    expect(agentResult.result).toBe("answer");
+    expect(agentResult.costUsd).toBeCloseTo(0.0042);
+    expect(agentResult.inputTokens).toBe(800);
+    expect(agentResult.outputTokens).toBe(300);
+    expect(agentResult.cacheReadTokens).toBe(150);
+    expect(agentResult.cacheCreationTokens).toBe(50);
+  });
+
+  it("defaults token counts and cost to zero when usage is absent", async () => {
+    vi.mocked(query).mockReturnValue(
+      (async function* () {
+        yield { result: "ok" };
+      })()
+    );
+
+    const agentResult = await runAgent("prompt", "system");
+    expect(agentResult.costUsd).toBe(0);
+    expect(agentResult.inputTokens).toBe(0);
+    expect(agentResult.outputTokens).toBe(0);
+    expect(agentResult.cacheReadTokens).toBe(0);
+    expect(agentResult.cacheCreationTokens).toBe(0);
+  });
+
   it("passes systemPrompt and model to query options", async () => {
     vi.mocked(query).mockReturnValue(
       (async function* () {
@@ -126,7 +167,7 @@ describe("runAgent", () => {
 
     const resultPromise = runAgent("prompt", "system");
     await vi.runAllTimersAsync();
-    const result = await resultPromise;
+    const { result } = await resultPromise;
     expect(result).toBe("recovered");
     expect(vi.mocked(query)).toHaveBeenCalledTimes(2);
     vi.useRealTimers();

packages/agent/src/__tests__/index.test.ts

@@ -45,8 +45,12 @@ vi.mock("../slack.js", () => ({
   postEphemeral: vi.fn().mockResolvedValue(undefined),
 }));
 
+const mockAgentResult = (result: string, costUsd = 0) => ({
+  result, inputTokens: 100, outputTokens: 50, cacheReadTokens: 10, cacheCreationTokens: 5, costUsd,
+});
+
 vi.mock("../agent.js", () => ({
-  runAgent: vi.fn().mockResolvedValue("Agent response"),
+  runAgent: vi.fn().mockResolvedValue(mockAgentResult("Agent response")),
 }));
 
 vi.mock("../prompt.js", () => ({
@@ -103,7 +107,7 @@ describe("processEvent", () => {
   it("fetches thread history, runs agent, and posts response", async () => {
     const history = [{ user: "U1", text: "earlier message" }];
     vi.mocked(fetchThreadHistory).mockResolvedValueOnce(history);
-    vi.mocked(runAgent).mockResolvedValueOnce("Here is your answer.");
+    vi.mocked(runAgent).mockResolvedValueOnce(mockAgentResult("Here is your answer."));
 
     await processEvent(baseEvent);
 
@@ -120,7 +124,7 @@ describe("processEvent", () => {
   });
 
   it("does not post or audit when agent returns the no-reply sentinel", async () => {
-    vi.mocked(runAgent).mockResolvedValueOnce("__NO_REPLY__");
+    vi.mocked(runAgent).mockResolvedValueOnce(mockAgentResult("__NO_REPLY__"));
     await processEvent({ ...baseEvent, requires_discretion: true });
     expect(postMessage).not.toHaveBeenCalled();
     expect(mockAuditLog).not.toHaveBeenCalled();
@@ -136,7 +140,7 @@ describe("processEvent", () => {
   });
 
   it("passes the agent response to postMessage", async () => {
-    vi.mocked(runAgent).mockResolvedValueOnce("Custom agent answer");
+    vi.mocked(runAgent).mockResolvedValueOnce(mockAgentResult("Custom agent answer"));
     await processEvent(baseEvent);
     expect(postMessage).toHaveBeenCalledWith("C_CHAN", "1.0", "Custom agent answer");
   });
@@ -176,9 +180,9 @@ describe("processEvent", () => {
     await expect(processEvent(baseEvent)).rejects.toThrow("Agent failed");
   });
 
-  it("writes audit record with channel, user, model, and duration after posting response", async () => {
+  it("writes audit record with channel, user, model, duration, and token usage after posting response", async () => {
     vi.mocked(mockConfigStore.get).mockResolvedValueOnce("claude-opus-4-6");
-    vi.mocked(runAgent).mockResolvedValueOnce("The answer.");
+    vi.mocked(runAgent).mockResolvedValueOnce(mockAgentResult("The answer.", 0.0042));
 
     await processEvent(baseEvent);
 
@@ -189,6 +193,11 @@ describe("processEvent", () => {
         user_id: "U1",
         response: "The answer.",
         model: "claude-opus-4-6",
+        input_tokens: 100,
+        output_tokens: 50,
+        cache_read_tokens: 10,
+        cache_creation_tokens: 5,
+        cost_usd: "0.004200",
       })
     );
     expect(typeof mockAuditLog.mock.calls[0][0].duration_ms).toBe("number");

packages/agent/src/agent.ts

@@ -3,13 +3,22 @@ import * as path from "path";
 import { randomUUID } from "crypto";
 import type { ImageContent } from "./slack.js";
 
+export interface AgentResult {
+  result: string;
+  inputTokens: number;
+  outputTokens: number;
+  cacheReadTokens: number;
+  cacheCreationTokens: number;
+  costUsd: number;
+}
+
 export async function runAgent(
   prompt: string,
   systemPrompt: string,
   model?: string,
   maxTokens?: number,
   images?: ImageContent[],
-): Promise<string> {
+): Promise<AgentResult> {
   const maxRetries = parseInt(process.env.MAX_MCP_RETRIES ?? "2", 10);
   let lastError: Error | undefined;
 
@@ -82,7 +91,7 @@ async function runAgentOnce(
   model?: string,
   maxTokens?: number,
   images?: ImageContent[],
-): Promise<string> {
+): Promise<AgentResult> {
   const urlFetcherPath = path.resolve(
     __dirname,
     "../../mcp-url-fetcher/dist/index.js"
@@ -105,6 +114,11 @@ async function runAgentOnce(
     : prompt;
 
   let result = "";
+  let inputTokens = 0;
+  let outputTokens = 0;
+  let cacheReadTokens = 0;
+  let cacheCreationTokens = 0;
+  let costUsd = 0;
 
   for await (const message of query({
     prompt: sdkPrompt,
@@ -174,8 +188,25 @@ async function runAgentOnce(
   })) {
     if ("result" in message) {
       result = message.result;
+      const msg = message as Record<string, unknown>;
+      costUsd = typeof msg.total_cost_usd === "number" ? msg.total_cost_usd : 0;
+      // SDKResultMessage.usage is the API Usage shape — snake_case keys
+      const usage = msg.usage as Record<string, number> | undefined;
+      if (usage) {
+        inputTokens = usage.input_tokens ?? 0;
+        outputTokens = usage.output_tokens ?? 0;
+        cacheReadTokens = usage.cache_read_input_tokens ?? 0;
+        cacheCreationTokens = usage.cache_creation_input_tokens ?? 0;
+      }
     }
   }
 
-  return result || "I was unable to generate a response.";
+  return {
+    result: result || "I was unable to generate a response.",
+    inputTokens,
+    outputTokens,
+    cacheReadTokens,
+    cacheCreationTokens,
+    costUsd,
+  };
 }

packages/agent/src/index.ts

@@ -76,23 +76,28 @@ export async function processEvent(event: SlackEvent): Promise<void> {
     : [];
 
   const start = Date.now();
-  const response = await runAgent(prompt, systemPrompt, model, maxTokens, images);
+  const agentResult = await runAgent(prompt, systemPrompt, model, maxTokens, images);
   const duration_ms = Date.now() - start;
 
-  if (response.trim() === NO_REPLY_SENTINEL) {
+  if (agentResult.result.trim() === NO_REPLY_SENTINEL) {
     console.log(`[discretion] agent elected not to reply in channel ${event.channel}`);
     return;
   }
 
-  await postMessage(event.channel, event.thread_ts, response);
+  await postMessage(event.channel, event.thread_ts, agentResult.result);
   await auditLogger.log({
     channel: event.channel,
     thread_ts: event.thread_ts,
     user_id: event.user,
     prompt,
-    response,
+    response: agentResult.result,
     model: model ?? null,
     duration_ms,
+    input_tokens: agentResult.inputTokens,
+    output_tokens: agentResult.outputTokens,
+    cache_read_tokens: agentResult.cacheReadTokens,
+    cache_creation_tokens: agentResult.cacheCreationTokens,
+    cost_usd: agentResult.costUsd.toFixed(6),
   });
 }
 

packages/db/drizzle/0001_audit_token_cost.sql

@@ -0,0 +1,9 @@
+ALTER TABLE "audit_log" ADD COLUMN "input_tokens" integer NOT NULL DEFAULT 0;
+--> statement-breakpoint
+ALTER TABLE "audit_log" ADD COLUMN "output_tokens" integer NOT NULL DEFAULT 0;
+--> statement-breakpoint
+ALTER TABLE "audit_log" ADD COLUMN "cache_read_tokens" integer NOT NULL DEFAULT 0;
+--> statement-breakpoint
+ALTER TABLE "audit_log" ADD COLUMN "cache_creation_tokens" integer NOT NULL DEFAULT 0;
+--> statement-breakpoint
+ALTER TABLE "audit_log" ADD COLUMN "cost_usd" text NOT NULL DEFAULT '0';

packages/db/drizzle/meta/_journal.json

@@ -8,6 +8,13 @@
       "when": 1742688000000,
       "tag": "0000_initial",
       "breakpoints": true
+    },
+    {
+      "idx": 1,
+      "version": "7",
+      "when": 1751760000000,
+      "tag": "0001_audit_token_cost",
+      "breakpoints": true
     }
   ]
 }