fix(compact): read usage without the debug body sampler

Author: DevMello

src/server/responses/compact.ts

@@ -83,10 +83,10 @@ import { redactSecretString } from "../../lib/redact";
 import { readBoundedResponseBody } from "../../lib/bounded-body";
 import { supportedLadderFor } from "../effort-policy";
 import {
+  applyResponseLogMetadata,
   beginRequestAttempt,
   catalogModelSupportsServiceTier,
   finishRequestAttempt,
-  inspectResponseLogJson,
   noteAttemptSend,
   readConfiguredCodexServiceTier,
   requestLogSpeedLabel,
@@ -483,9 +483,16 @@ export async function handleResponsesCompact(
     // 200 upstream response must not soft-avoid a healthy account or rotate a thread.
     recordCompactPoolOutcome(outcomeCtx, upstream.status, { retryAfter, resetAt });
     // Lift usage and response metadata from the buffered upstream JSON into the
-    // request log; the routed branch gets the same through handleResponses. The
-    // synthetic buffer errors are not upstream bodies and stay uninspected.
-    if (buffered.ok) inspectResponseLogJson(logCtx, await buffered.clone().text());
+    // request log; the routed branch gets the same through handleResponses. Only
+    // the parsed fields are read: a compact body is replacement history derived
+    // from the conversation, so it must never reach the usage debug body sampler.
+    if (buffered.ok) {
+      try {
+        applyResponseLogMetadata(logCtx, JSON.parse(await buffered.clone().text()));
+      } catch {
+        /* body may not be JSON; usage stays unreported */
+      }
+    }
     return buffered;
   }
 

tests/responses-compaction-routing.test.ts

@@ -180,15 +180,26 @@ describe("native compact usage reporting", () => {
     } as unknown as OcxConfig;
     globalThis.fetch = (async () => jsonResponse(completedPayload("native summary"))) as typeof fetch;
     const logCtx: RequestLogContext = { model: "", provider: "" };
-    const response = await handleResponsesCompact(
-      compactionRequest(baseCompactionBody({ model: "openai-apikey/gpt-5.5" })),
-      config,
-      logCtx,
-    );
+    const previousUsageDebug = process.env.OPENCODEX_USAGE_DEBUG;
+    process.env.OPENCODEX_USAGE_DEBUG = "1";
+    let response: Response;
+    try {
+      response = await handleResponsesCompact(
+        compactionRequest(baseCompactionBody({ model: "openai-apikey/gpt-5.5" })),
+        config,
+        logCtx,
+      );
+    } finally {
+      if (previousUsageDebug === undefined) delete process.env.OPENCODEX_USAGE_DEBUG;
+      else process.env.OPENCODEX_USAGE_DEBUG = previousUsageDebug;
+    }
     expect(response.status).toBe(200);
-    const body = await response.json() as { usage?: Record<string, unknown> };
-    expect(body.usage).toMatchObject({ input_tokens: 10, output_tokens: 5, total_tokens: 15 });
+    expect(await response.json()).toEqual(completedPayload("native summary"));
     expect(logCtx.usage).toMatchObject({ inputTokens: 10, outputTokens: 5, totalTokens: 15 });
+    // The compact body is replacement history; even with usage debug on it must
+    // never be sampled into the debug log.
+    expect(logCtx.usageDebugBodyKind).toBeUndefined();
+    expect(logCtx.usageDebugBodySample).toBeUndefined();
   });
 });