fix(codex): persist automatic compaction history (#110587)

steipete · web-flow · commit f14947d256d8 · 2026-07-18T10:37:25.000+01:00
* fix(codex): persist automatic compaction counts

* test(codex): stress compaction and restart recovery

* test(codex): bind compaction waves to fresh counts

* chore: keep Codex release note in PR
diff --git a/docs/help/testing-live.md b/docs/help/testing-live.md
@@ -334,6 +334,17 @@ Docker notes:
   history. Override the bounded counts with
   `OPENCLAW_LIVE_CODEX_HARNESS_RESUME_STRESS_HISTORY_TURNS` (1-20) and
   `OPENCLAW_LIVE_CODEX_HARNESS_RESUME_STRESS_RESTARTS` (1-10).
+- Optional fan-out stress: set `OPENCLAW_LIVE_CODEX_HARNESS_SUBAGENT_PROBE=1`
+  and `OPENCLAW_LIVE_CODEX_HARNESS_SUBAGENT_COUNT` (1-12). The harness starts
+  every child concurrently, waits for every terminal run, and verifies each
+  unique child reply and native thread identity.
+- Optional compaction stress: `OPENCLAW_LIVE_CODEX_HARNESS_COMPACTION_STRESS=1`
+  generates bounded native tool output, requires automatic compaction events,
+  verifies the persisted compaction count and hidden-marker recall, restarts
+  the Gateway and physical Codex app-server, then repeats the output and
+  compaction wave. Tune the bounded work with
+  `OPENCLAW_LIVE_CODEX_HARNESS_COMPACTION_STRESS_TURNS` (1-8) and
+  `OPENCLAW_LIVE_CODEX_HARNESS_LARGE_OUTPUT_BYTES` (100000-1000000).
 - Optional loop-relay opt-out probe:
   `OPENCLAW_LIVE_CODEX_HARNESS_DISABLE_LOOP_RELAY=1`
 - The requested thinking preference may map to the nearest effort advertised
@@ -367,6 +378,17 @@ Restart and history stress:
 
 ```bash
 OPENCLAW_LIVE_CODEX_HARNESS_RESUME_STRESS=1 \
+pnpm test:docker:live-codex-harness
+```
+
+Fan-out, large-output, compaction, and restart stress:
+
+```bash
+OPENCLAW_LIVE_CODEX_HARNESS_AUTH=api-key \
+  OPENCLAW_LIVE_CODEX_HARNESS_SUBAGENT_PROBE=1 \
+  OPENCLAW_LIVE_CODEX_HARNESS_SUBAGENT_COUNT=8 \
+  OPENCLAW_LIVE_CODEX_HARNESS_RESUME_STRESS=1 \
+  OPENCLAW_LIVE_CODEX_HARNESS_COMPACTION_STRESS=1 \
   pnpm test:docker:live-codex-harness
 ```
 
diff --git a/extensions/codex/src/app-server/event-projector.test.ts b/extensions/codex/src/app-server/event-projector.test.ts
@@ -2998,7 +2998,10 @@ describe("CodexAppServerEventProjector", () => {
     expect(JSON.stringify(result.messagesSnapshot[2])).toContain("Codex plan");
     expect(JSON.stringify(result.messagesSnapshot[2])).toContain("next");
     expect(JSON.stringify(result.messagesSnapshot[2])).toContain("[in_progress] patch");
-    expect(requireRecord(result.itemLifecycle, "item lifecycle").compactionCount).toBe(1);
+    expect(result.compactionCount).toBe(1);
+    expect(requireRecord(result.itemLifecycle, "item lifecycle")).not.toHaveProperty(
+      "compactionCount",
+    );
     expect(onContextCompacted).toHaveBeenCalledOnce();
   });
 
diff --git a/extensions/codex/src/app-server/event-projector.ts b/extensions/codex/src/app-server/event-projector.ts
@@ -434,6 +434,9 @@ export class CodexAppServerEventProjector {
       successfulCronAdds: toolTelemetry.successfulCronAdds,
       cloudCodeAssistFormatError: false,
       attemptUsage: projectedUsage,
+      ...(this.completedCompactionCount > 0
+        ? { compactionCount: this.completedCompactionCount }
+        : {}),
       replayMetadata: {
         hadPotentialSideEffects,
         replaySafe: !hadPotentialSideEffects,
@@ -442,9 +445,6 @@ export class CodexAppServerEventProjector {
         startedCount: this.activeItemIds.size + this.completedItemIds.size,
         completedCount: this.completedItemIds.size,
         activeCount: this.activeItemIds.size,
-        ...(this.completedCompactionCount > 0
-          ? { compactionCount: this.completedCompactionCount }
-          : {}),
       },
       yieldDetected: options?.yieldDetected || false,
       didSendDeterministicApprovalPrompt:
diff --git a/extensions/codex/src/app-server/run-attempt-start.ts b/extensions/codex/src/app-server/run-attempt-start.ts
@@ -159,7 +159,12 @@ export async function startCodexAttemptRuntime(resources: CodexAttemptResources)
     state.codexSandboxPolicy = startupResult.sandboxPolicy;
     void emitCodexAppServerEvent(params, {
       stream: "codex_app_server.lifecycle",
-      data: { phase: "thread_ready", threadId: state.thread.threadId },
+      data: {
+        phase: "thread_ready",
+        threadId: state.thread.threadId,
+        action: state.thread.lifecycle.action,
+        clientId: state.client.getInstanceId(),
+      },
     });
   } catch (error) {
     activateNativePreToolUseFailureFallback();
diff --git a/scripts/test-live-codex-harness-docker.sh b/scripts/test-live-codex-harness-docker.sh
@@ -367,11 +367,17 @@ echo "==> Chat image probe: ${OPENCLAW_LIVE_CODEX_HARNESS_CHAT_IMAGE_PROBE:-0}"
 echo "==> Image probe: ${OPENCLAW_LIVE_CODEX_HARNESS_IMAGE_PROBE:-1}"
 echo "==> MCP probe: ${OPENCLAW_LIVE_CODEX_HARNESS_MCP_PROBE:-1}"
 echo "==> Subagent probe: ${OPENCLAW_LIVE_CODEX_HARNESS_SUBAGENT_PROBE:-1}"
+echo "==> Subagent count: ${OPENCLAW_LIVE_CODEX_HARNESS_SUBAGENT_COUNT:-1}"
 echo "==> Subagent-only fast path: ${OPENCLAW_LIVE_CODEX_HARNESS_SUBAGENT_ONLY:-auto}"
 echo "==> Guardian probe: ${OPENCLAW_LIVE_CODEX_HARNESS_GUARDIAN_PROBE:-1}"
+echo "==> Code-mode-only probe: ${OPENCLAW_LIVE_CODEX_HARNESS_CODE_MODE_ONLY:-0}"
+echo "==> Loop relay disabled: ${OPENCLAW_LIVE_CODEX_HARNESS_DISABLE_LOOP_RELAY:-0}"
 echo "==> Resume stress: ${OPENCLAW_LIVE_CODEX_HARNESS_RESUME_STRESS:-0}"
 echo "==> Resume stress history turns: ${OPENCLAW_LIVE_CODEX_HARNESS_RESUME_STRESS_HISTORY_TURNS:-4}"
 echo "==> Resume stress restarts: ${OPENCLAW_LIVE_CODEX_HARNESS_RESUME_STRESS_RESTARTS:-3}"
+echo "==> Compaction stress: ${OPENCLAW_LIVE_CODEX_HARNESS_COMPACTION_STRESS:-0}"
+echo "==> Compaction stress turns: ${OPENCLAW_LIVE_CODEX_HARNESS_COMPACTION_STRESS_TURNS:-4}"
+echo "==> Large output bytes: ${OPENCLAW_LIVE_CODEX_HARNESS_LARGE_OUTPUT_BYTES:-300000}"
 echo "==> Auth mode: $CODEX_HARNESS_AUTH_MODE"
 echo "==> Profile file: $PROFILE_STATUS"
 echo "==> CI-safe Codex config: ${OPENCLAW_LIVE_CODEX_HARNESS_USE_CI_SAFE_CODEX_CONFIG:-1}"
@@ -401,9 +407,14 @@ DOCKER_RUN_ARGS+=(--rm -t \
   -e OPENCLAW_LIVE_CODEX_HARNESS_AUTH="$CODEX_HARNESS_AUTH_MODE" \
   -e OPENCLAW_LIVE_CODEX_HARNESS=1 \
   -e OPENCLAW_LIVE_CODEX_HARNESS_CHAT_IMAGE_PROBE="${OPENCLAW_LIVE_CODEX_HARNESS_CHAT_IMAGE_PROBE:-0}" \
+  -e OPENCLAW_LIVE_CODEX_HARNESS_CODE_MODE_ONLY="${OPENCLAW_LIVE_CODEX_HARNESS_CODE_MODE_ONLY:-0}" \
+  -e OPENCLAW_LIVE_CODEX_HARNESS_COMPACTION_STRESS="${OPENCLAW_LIVE_CODEX_HARNESS_COMPACTION_STRESS:-0}" \
+  -e OPENCLAW_LIVE_CODEX_HARNESS_COMPACTION_STRESS_TURNS="${OPENCLAW_LIVE_CODEX_HARNESS_COMPACTION_STRESS_TURNS:-4}" \
   -e OPENCLAW_LIVE_CODEX_HARNESS_DEBUG="${OPENCLAW_LIVE_CODEX_HARNESS_DEBUG:-}" \
+  -e OPENCLAW_LIVE_CODEX_HARNESS_DISABLE_LOOP_RELAY="${OPENCLAW_LIVE_CODEX_HARNESS_DISABLE_LOOP_RELAY:-0}" \
   -e OPENCLAW_LIVE_CODEX_HARNESS_GUARDIAN_PROBE="${OPENCLAW_LIVE_CODEX_HARNESS_GUARDIAN_PROBE:-1}" \
   -e OPENCLAW_LIVE_CODEX_HARNESS_IMAGE_PROBE="${OPENCLAW_LIVE_CODEX_HARNESS_IMAGE_PROBE:-1}" \
+  -e OPENCLAW_LIVE_CODEX_HARNESS_LARGE_OUTPUT_BYTES="${OPENCLAW_LIVE_CODEX_HARNESS_LARGE_OUTPUT_BYTES:-300000}" \
   -e OPENCLAW_LIVE_CODEX_HARNESS_MCP_PROBE="${OPENCLAW_LIVE_CODEX_HARNESS_MCP_PROBE:-1}" \
   -e OPENCLAW_LIVE_CODEX_HARNESS_MODEL="${OPENCLAW_LIVE_CODEX_HARNESS_MODEL:-openai/gpt-5.6-luna}" \
   -e OPENCLAW_LIVE_CODEX_HARNESS_TARGETS="${OPENCLAW_LIVE_CODEX_HARNESS_TARGETS:-}" \
@@ -416,6 +427,7 @@ DOCKER_RUN_ARGS+=(--rm -t \
   -e OPENCLAW_LIVE_CODEX_HARNESS_RESUME_STRESS_RESTARTS="${OPENCLAW_LIVE_CODEX_HARNESS_RESUME_STRESS_RESTARTS:-3}" \
   -e OPENCLAW_LIVE_CODEX_HARNESS_SETUP_TIMEOUT_SECONDS="$CODEX_HARNESS_SETUP_TIMEOUT_SECONDS" \
   -e OPENCLAW_LIVE_CODEX_HARNESS_SUBAGENT_ONLY="${OPENCLAW_LIVE_CODEX_HARNESS_SUBAGENT_ONLY:-}" \
+  -e OPENCLAW_LIVE_CODEX_HARNESS_SUBAGENT_COUNT="${OPENCLAW_LIVE_CODEX_HARNESS_SUBAGENT_COUNT:-1}" \
   -e OPENCLAW_LIVE_CODEX_HARNESS_SUBAGENT_PROBE="${OPENCLAW_LIVE_CODEX_HARNESS_SUBAGENT_PROBE:-1}" \
   -e OPENCLAW_LIVE_CODEX_HARNESS_USE_CI_SAFE_CODEX_CONFIG="${OPENCLAW_LIVE_CODEX_HARNESS_USE_CI_SAFE_CODEX_CONFIG:-1}" \
   -e OPENCLAW_LIVE_CODEX_CLI_PACKAGE_SPEC="$CODEX_CLI_PACKAGE_SPEC" \
diff --git a/src/gateway/gateway-codex-harness.live-helpers.test.ts b/src/gateway/gateway-codex-harness.live-helpers.test.ts
@@ -10,6 +10,7 @@ import {
   isExpectedYieldedAgentTimeout,
   isRetryableCodexHarnessLiveError,
   isStrictExpectedCodexModelsCommandText,
+  shouldUseCodexHarnessSubagentOnlyFastPath,
 } from "./gateway-codex-harness.live-helpers.js";
 
 const includesExpectedCodexModelsCommandText = (text: string) =>
@@ -30,6 +31,30 @@ function expectStrictCodexModelsCommandText(text: string): void {
 }
 
 describe("gateway codex harness live helpers", () => {
+  it("keeps combined stress probes out of the subagent-only fast path", () => {
+    const base = {
+      chatImageProbe: false,
+      codeModeOnly: false,
+      compactionStress: false,
+      explicitOptOut: false,
+      guardianProbe: false,
+      imageProbe: false,
+      mcpProbe: false,
+      resumeStress: false,
+      subagentProbe: true,
+    };
+
+    expect(shouldUseCodexHarnessSubagentOnlyFastPath(base)).toBe(true);
+    expect(shouldUseCodexHarnessSubagentOnlyFastPath({ ...base, resumeStress: true })).toBe(false);
+    expect(shouldUseCodexHarnessSubagentOnlyFastPath({ ...base, compactionStress: true })).toBe(
+      false,
+    );
+    expect(shouldUseCodexHarnessSubagentOnlyFastPath({ ...base, codeModeOnly: true })).toBe(false);
+    expect(shouldUseCodexHarnessSubagentOnlyFastPath({ ...base, explicitOptOut: true })).toBe(
+      false,
+    );
+  });
+
   it("classifies sessions.list timeouts as retryable live Codex errors", () => {
     const error = new Error("gateway request timeout for sessions.list");
 
diff --git a/src/gateway/gateway-codex-harness.live-helpers.ts b/src/gateway/gateway-codex-harness.live-helpers.ts
@@ -81,6 +81,30 @@ export const EXPECTED_CODEX_MODELS_COMMAND_TEXT = [
   "Current OpenClaw session status reports the active model as:",
 ] as const;
 
+export function shouldUseCodexHarnessSubagentOnlyFastPath(params: {
+  chatImageProbe: boolean;
+  codeModeOnly: boolean;
+  compactionStress: boolean;
+  explicitOptOut: boolean;
+  guardianProbe: boolean;
+  imageProbe: boolean;
+  mcpProbe: boolean;
+  resumeStress: boolean;
+  subagentProbe: boolean;
+}): boolean {
+  return (
+    params.subagentProbe &&
+    !params.chatImageProbe &&
+    !params.codeModeOnly &&
+    !params.compactionStress &&
+    !params.guardianProbe &&
+    !params.imageProbe &&
+    !params.mcpProbe &&
+    !params.resumeStress &&
+    !params.explicitOptOut
+  );
+}
+
 const HEALTHY_CODEX_MODELS_COMMAND_TEXT = [
   "Codex models:",
   "Available Codex models",
diff --git a/src/gateway/gateway-codex-harness.live.test.ts b/src/gateway/gateway-codex-harness.live.test.ts
diff --git a/test/scripts/test-live-codex-harness-docker.test.ts b/test/scripts/test-live-codex-harness-docker.test.ts
