fix(acp): prevent OOM crash loop via event compaction and debounced persist (#4032)

.github/workflows/ci.yml

@@ -182,7 +182,7 @@ jobs:
             kill "$AEGIS_PID" || true
           fi
       - name: Check bundle size
-        run: "THRESHOLD_KB=2195\nSERVER_SIZE=$(find dist/ -name \"*.js\" ! -path \"\
+        run: "THRESHOLD_KB=2210\nSERVER_SIZE=$(find dist/ -name \"*.js\" ! -path \"\
           */__tests__/*\" ! -path \"*/dashboard/*\" -exec du -ck {} + | tail -1 | awk '{print $1}')\nSERVER_SIZE_KB=$((SERVER_SIZE))\n\
           echo \"## Bundle Size Report\" >> \"$GITHUB_STEP_SUMMARY\"\necho \"\" >> \"\
           $GITHUB_STEP_SUMMARY\"\necho \"| Scope | Size (KB) | Threshold (KB) | Status\
@@ -330,7 +330,7 @@ jobs:
           }
       - name: Check bundle size
         if: runner.os != 'Windows'
-        run: "THRESHOLD_KB=2195\nSERVER_SIZE=$(find dist/ -name \"*.js\" ! -path \"\
+        run: "THRESHOLD_KB=2210\nSERVER_SIZE=$(find dist/ -name \"*.js\" ! -path \"\
           */__tests__/*\" ! -path \"*/dashboard/*\" -exec du -ck {} + | tail -1 | awk '{print $1}')\nSERVER_SIZE_KB=$((SERVER_SIZE))\n\
           echo \"## Bundle Size Report\" >> \"$GITHUB_STEP_SUMMARY\"\necho \"\" >> \"\
           $GITHUB_STEP_SUMMARY\"\necho \"| Scope | Size (KB) | Threshold (KB) | Status\

scripts/check-bundle-size.sh

@@ -2,7 +2,7 @@
 # Bundle size gate — mirrors CI threshold from .github/workflows/ci.yml
 set -euo pipefail
 
-THRESHOLD_KB=2195
+THRESHOLD_KB=2210
 
 if [ ! -d "dist" ]; then
   echo "❌ dist/ not found — run 'npm run build' first"

src/__tests__/acp-local-storage.test.ts

@@ -210,7 +210,7 @@ describe('file ACP local-dev storage profile', () => {
 
   it('persists sessions, event replay, and queued actions across profile restart', async () => {
     const storageFile = path.join(scratchDir, 'local-acp-storage.json');
-    const firstProfile = createFileAcpLocalStorageProfile({ filePath: storageFile });
+    const firstProfile = createFileAcpLocalStorageProfile({ filePath: storageFile, persistDebounceMs: 0 });
     await firstProfile.start();
 
     await firstProfile.sessionStore.create(makeSessionRecord({ status: 'running' }));
@@ -227,7 +227,7 @@ describe('file ACP local-dev storage profile', () => {
     );
     await firstProfile.stop();
 
-    const secondProfile = createFileAcpLocalStorageProfile({ filePath: storageFile });
+    const secondProfile = createFileAcpLocalStorageProfile({ filePath: storageFile, persistDebounceMs: 0 });
     await secondProfile.start();
 
     const session = await secondProfile.sessionStore.get('session-1', scope);

src/__tests__/fix-3366-acp-persist-cascade.test.ts

@@ -48,7 +48,7 @@ describe('Issue #3366: ACP local storage persist() failure cascade', () => {
   });
 
   it('recovers from a failed persist — subsequent writes succeed', async () => {
-    const profile = createFileAcpLocalStorageProfile({ filePath });
+    const profile = createFileAcpLocalStorageProfile({ filePath, persistDebounceMs: 0 });
     await profile.start();
     expect(profile.getPersistError()).toBeNull();
 
@@ -75,7 +75,7 @@ describe('Issue #3366: ACP local storage persist() failure cascade', () => {
   });
 
   it('does NOT cascade rejection across multiple sequential failures', async () => {
-    const profile = createFileAcpLocalStorageProfile({ filePath });
+    const profile = createFileAcpLocalStorageProfile({ filePath, persistDebounceMs: 0 });
     await profile.start();
 
     // Make dir read-only
@@ -106,7 +106,7 @@ describe('Issue #3366: ACP local storage persist() failure cascade', () => {
   });
 
   it('stop() does not throw even when writeChain has a rejection', async () => {
-    const profile = createFileAcpLocalStorageProfile({ filePath });
+    const profile = createFileAcpLocalStorageProfile({ filePath, persistDebounceMs: 0 });
     await profile.start();
 
     // Make dir read-only so persist fails

src/__tests__/server-core-coverage.test.ts

@@ -17,6 +17,7 @@ const originalEnv: Record<string, string | undefined> = {
   AEGIS_HOST: process.env.AEGIS_HOST,
   AEGIS_AUTH_TOKEN: process.env.AEGIS_AUTH_TOKEN,
   AEGIS_ALLOWED_WORK_DIRS: process.env.AEGIS_ALLOWED_WORK_DIRS,
+  AEGIS_PERSIST_DEBOUNCE_MS: process.env.AEGIS_PERSIST_DEBOUNCE_MS,
 };
 
 const authToken = 'server-core-token';
@@ -100,6 +101,7 @@ describe('server core coverage integration', () => {
     process.env.AEGIS_HOST = '127.0.0.1';
     process.env.AEGIS_AUTH_TOKEN = authToken;
     process.env.AEGIS_ALLOWED_WORK_DIRS = sandboxRoot;
+    process.env.AEGIS_PERSIST_DEBOUNCE_MS = '0';
 
     vi.spyOn(globalThis, 'setInterval').mockImplementation((() => 0) as any);
     vi.spyOn(globalThis, 'clearInterval').mockImplementation((() => undefined) as any);

src/__tests__/server-phase3.test.ts

@@ -25,6 +25,7 @@ const originalEnv: Record<string, string | undefined> = {
   AEGIS_HOST: process.env.AEGIS_HOST,
   AEGIS_AUTH_TOKEN: process.env.AEGIS_AUTH_TOKEN,
   AEGIS_ALLOWED_WORK_DIRS: process.env.AEGIS_ALLOWED_WORK_DIRS,
+  AEGIS_PERSIST_DEBOUNCE_MS: process.env.AEGIS_PERSIST_DEBOUNCE_MS,
 };
 
 const authToken = 'phase3-test-token';
@@ -111,6 +112,7 @@ describe('server.ts Phase 3 — internal functions', () => {
     process.env.AEGIS_HOST = '127.0.0.1';
     process.env.AEGIS_AUTH_TOKEN = authToken;
     process.env.AEGIS_ALLOWED_WORK_DIRS = sandboxRoot;
+    process.env.AEGIS_PERSIST_DEBOUNCE_MS = '0';
 
     // Capture interval callbacks instead of discarding them
     vi.spyOn(globalThis, 'setInterval').mockImplementation(((cb: (...args: unknown[]) => void, ms?: number) => {

src/server.ts

@@ -806,8 +806,13 @@ async function main(): Promise<void> {
   sessions = new SessionManager(config, sessionStore);
 
   // Issue #2607 / ACP-064: Initialize ACP local storage profile and backend
+  // Issue #4032: Allow test override of persist debounce via env var.
+  const persistDebounceMs = process.env.AEGIS_PERSIST_DEBOUNCE_MS
+    ? parseInt(process.env.AEGIS_PERSIST_DEBOUNCE_MS, 10)
+    : undefined;
   acpLocalProfile = createFileAcpLocalStorageProfile({
     filePath: path.join(config.stateDir, 'acp-local-storage.json'),
+    ...(persistDebounceMs !== undefined ? { persistDebounceMs } : {}),
   });
   await acpLocalProfile.start();
   // eslint-disable-next-line @typescript-eslint/no-explicit-any