codemem 0.36.0 → 0.37.0

package/dist/index.js

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-import { DEDUP_KEY_BACKFILL_JOB, DEFAULT_COORDINATOR_DB_PATH, DedupKeyBackfillRunner, MUTATING_TOOL_NAMES, MemoryStore, ObserverClient, REF_BACKFILL_JOB, RawEventSweeper, RefBackfillRunner, SCOPE_BACKFILL_JOB, SESSION_CONTEXT_BACKFILL_JOB, SUMMARY_DEDUP_BACKFILL_JOB, ScopeBackfillRunner, SessionContextBackfillRunner, SummaryDedupBackfillRunner, SyncRetentionRunner, VERSION, VectorModelMigrationRunner, aiBackfillStructuredContent, applyBootstrapSnapshot, backfillMemoryDedupKeys, backfillNarrativeFromBody, backfillTagsText, backfillVectors, buildAuthHeaders, buildBaseUrl, buildRawEventEnvelopeFromCodexHook, buildRawEventEnvelopeFromHook, compareMemoryRoleReports, connect, coordinatorCreateGroupAction, coordinatorCreateInviteAction, coordinatorCreateScopeAction, coordinatorDisableDeviceAction, coordinatorEnrollDeviceAction, coordinatorGrantScopeMembershipAction, coordinatorImportInviteAction, coordinatorListBootstrapGrantsAction, coordinatorListDevicesAction, coordinatorListGroupsAction, coordinatorListJoinRequestsAction, coordinatorListScopeMembershipsAction, coordinatorListScopesAction, coordinatorRemoveDeviceAction, coordinatorRenameDeviceAction, coordinatorReviewJoinRequestAction, coordinatorRevokeBootstrapGrantAction, coordinatorRevokeScopeMembershipAction, coordinatorUpdateScopeAction, createBetterSqliteCoordinatorApp, deactivateLowSignalMemories, deactivateLowSignalObservations, dedupNearDuplicateMemories, ensureDeviceIdentity, ensureSchemaBootstrapped, exportMemories, extractApplyPatchPaths, fetchAllSnapshotPages, fingerprintPublicKey, flushRawEvents, formatHostPort, getExtractionBenchmarkProfile, getInjectionEvalScenarioPack, getInjectionEvalScenarioPrompts, getMaintenanceJob, getMemoryRoleReport, getRawEventRelinkPlan, getRawEventRelinkReport, getRawEventStatus, getSemanticIndexDiagnostics, getSessionExtractionEval, getSessionExtractionEvalScenario, getWorkspaceCodememConfigPath, hasPendingDedupKeyBackfill, hasPendingRefBackfill, hasPendingScopeBackfill, hasPendingSessionContextBackfill, hasPendingSummaryDedupBackfill, hasUnsyncedSharedMemoryChanges, importMemories, initDatabase, isEmbeddingDisabled, listMaintenanceJobs, listPerPeerScopeSyncState, loadObserverConfig, loadPublicKey, loadSqliteVec, mdnsEnabled, planReplicationOpsAgePrune, pruneReplicationOpsUntilCaughtUp, rawEventsGate, readCodememConfigFile, readCodememConfigFileAtPath, readCoordinatorSyncConfig, readImportPayload, replayBatchExtraction, replayBatchExtractionWithTierRouting, requestJson, resolveCodememConfigPath, resolveDbPath, resolveHookProject, resolveProject, retryRawEventFailures, runSyncDaemon, runSyncPass, scanSecretsRetroactive, schema, setPeerProjectFilter, stripJsonComments, stripPrivateObj, stripTrailingCommas, syncPassPreflight, updatePeerAddresses, vacuumDatabase, writeCodememConfigFile } from "@codemem/core";
+import { DEDUP_KEY_BACKFILL_JOB, DEFAULT_COORDINATOR_DB_PATH, DedupKeyBackfillRunner, MUTATING_TOOL_NAMES, MemoryStore, ObserverClient, REF_BACKFILL_JOB, RawEventSweeper, RefBackfillRunner, SCOPE_BACKFILL_JOB, SESSION_CONTEXT_BACKFILL_JOB, SUMMARY_DEDUP_BACKFILL_JOB, ScopeBackfillRunner, SessionContextBackfillRunner, SummaryDedupBackfillRunner, SyncRetentionRunner, VERSION, VectorModelMigrationRunner, aiBackfillStructuredContent, applyBootstrapSnapshot, backfillMemoryDedupKeys, backfillNarrativeFromBody, backfillTagsText, backfillVectors, buildAuthHeaders, buildBaseUrl, buildRawEventEnvelopeFromCodexHook, buildRawEventEnvelopeFromHook, compareMemoryRoleReports, connect, coordinatorCreateGroupAction, coordinatorCreateInviteAction, coordinatorCreateScopeAction, coordinatorDisableDeviceAction, coordinatorEnrollDeviceAction, coordinatorGrantScopeMembershipAction, coordinatorImportInviteAction, coordinatorListBootstrapGrantsAction, coordinatorListDevicesAction, coordinatorListGroupsAction, coordinatorListJoinRequestsAction, coordinatorListScopeMembershipsAction, coordinatorListScopesAction, coordinatorRemoveDeviceAction, coordinatorRenameDeviceAction, coordinatorReviewJoinRequestAction, coordinatorRevokeBootstrapGrantAction, coordinatorRevokeScopeMembershipAction, coordinatorUpdateScopeAction, createBetterSqliteCoordinatorApp, deactivateLowSignalMemories, deactivateLowSignalObservations, dedupNearDuplicateMemories, ensureDeviceIdentity, ensureSchemaBootstrapped, exportMemories, extractApplyPatchPaths, fetchAllSnapshotPages, fingerprintPublicKey, flushRawEvents, formatHostPort, getExtractionBenchmarkProfile, getInjectionEvalScenarioPack, getInjectionEvalScenarioPrompts, getMaintenanceJob, getMemoryRoleReport, getRawEventRelinkPlan, getRawEventRelinkReport, getRawEventStatus, getSemanticIndexDiagnostics, getSessionExtractionEval, getSessionExtractionEvalScenario, getWorkspaceCodememConfigPath, hasPendingDedupKeyBackfill, hasPendingRefBackfill, hasPendingScopeBackfill, hasPendingSessionContextBackfill, hasPendingSummaryDedupBackfill, hasUnsyncedSharedMemoryChanges, importMemories, initDatabase, isEmbeddingDisabled, listMaintenanceJobs, listPerPeerScopeSyncState, listRetentionScopeIds, loadObserverConfig, loadPublicKey, loadSqliteVec, mdnsEnabled, planReplicationOpsAgePrune, pruneReplicationOpsUntilCaughtUp, rawEventsGate, readCodememConfigFile, readCodememConfigFileAtPath, readCoordinatorSyncConfig, readImportPayload, replayBatchExtraction, replayBatchExtractionWithTierRouting, requestJson, resolveCodememConfigPath, resolveDbPath, resolveHookProject, resolveProject, retryRawEventFailures, runSyncDaemon, runSyncPass, scanSecretsRetroactive, schema, setPeerProjectFilter, stripJsonComments, stripPrivateObj, stripTrailingCommas, syncPassPreflight, updatePeerAddresses, vacuumDatabase, writeCodememConfigFile } from "@codemem/core";
 import { Command, Option } from "commander";
 import omelette from "omelette";
 import { appendFileSync, copyFileSync, existsSync, mkdirSync, readFileSync, readdirSync, renameSync, rmSync, rmdirSync, statSync, unlinkSync, writeFileSync } from "node:fs";
@@ -2864,18 +2864,33 @@ function buildCoordinatorCommand() {
 	coordServeCmd.addOption(new Option("--host <host>", "bind host").hideHelp());
 	coordServeCmd.addOption(new Option("--port <port>", "bind port").hideHelp());
 	coordServeCmd.action(async (opts) => {
-		const host = String(opts.coordinatorHost ?? opts.host ?? "127.0.0.1").trim() || "127.0.0.1";
-		const port = Number.parseInt(String(opts.coordinatorPort ?? opts.port ?? "7347"), 10);
-		const dbPath = resolveDbOpt(opts) ?? DEFAULT_COORDINATOR_DB_PATH;
-		const app = createBetterSqliteCoordinatorApp({ dbPath });
-		p.intro("codemem coordinator serve");
-		p.log.success(`Coordinator listening at http://${host}:${port}`);
-		p.log.info(`DB: ${dbPath}`);
-		serve({
-			fetch: app.fetch,
-			hostname: host,
-			port
-		});
+		try {
+			const host = String(opts.coordinatorHost ?? opts.host ?? "127.0.0.1").trim() || "127.0.0.1";
+			const port = Number.parseInt(String(opts.coordinatorPort ?? opts.port ?? "7347"), 10);
+			if (!Number.isFinite(port) || port < 0 || port > 65535) {
+				p.log.error(`Invalid port: ${opts.coordinatorPort ?? opts.port}`);
+				process.exitCode = 1;
+				return;
+			}
+			const dbPath = resolveDbOpt(opts) ?? DEFAULT_COORDINATOR_DB_PATH;
+			const app = createBetterSqliteCoordinatorApp({ dbPath });
+			p.intro("codemem coordinator serve");
+			serve({
+				fetch: app.fetch,
+				hostname: host,
+				port
+			}, (info) => {
+				p.log.success(`Coordinator listening at http://${host}:${info.port}`);
+				p.log.info(`DB: ${dbPath}`);
+			}).on("error", (error) => {
+				const message = error.code === "EADDRINUSE" ? `Port ${port} is already in use` : error instanceof Error ? error.message : String(error);
+				p.log.error(message);
+				process.exitCode = 1;
+			});
+		} catch (error) {
+			p.log.error(error instanceof Error ? error.message : String(error));
+			process.exitCode = 1;
+		}
 	});
 	cmd.addCommand(coordServeCmd);
 	const listBootstrapGrantsCmd = new Command("list-bootstrap-grants").configureHelp(helpStyle).description("List bootstrap grants for a coordinator group").argument("<group>", "group id").option("--remote-url <url>", "remote coordinator URL override").option("--admin-secret <secret>", "remote coordinator admin secret override");
@@ -3137,22 +3152,32 @@ var dbCommand = new Command("db").configureHelp(helpStyle).description("Database
 var initCmd = new Command("init").configureHelp(helpStyle).description("Create or verify the SQLite database and schema");
 addDbOption(initCmd);
 initCmd.action((opts) => {
-	const result = initDatabase(resolveDbOpt(opts));
-	p.intro("codemem db init");
-	p.log.success(`Database ready: ${result.path}`);
-	p.outro(`Size: ${result.sizeBytes.toLocaleString()} bytes`);
+	try {
+		const result = initDatabase(resolveDbOpt(opts));
+		p.intro("codemem db init");
+		p.log.success(`Database ready: ${result.path}`);
+		p.outro(`Size: ${result.sizeBytes.toLocaleString()} bytes`);
+	} catch (error) {
+		p.log.error(error instanceof Error ? error.message : String(error));
+		process.exitCode = 1;
+	}
 });
 dbCommand.addCommand(initCmd);
 var vacuumCmd = new Command("vacuum").configureHelp(helpStyle).description("Run VACUUM on the SQLite database");
 addDbOption(vacuumCmd);
 vacuumCmd.action((opts) => {
-	const result = vacuumDatabase(resolveDbOpt(opts));
-	p.intro("codemem db vacuum");
-	p.log.success(`Vacuumed: ${result.path}`);
-	p.outro(`Size: ${result.sizeBytes.toLocaleString()} bytes`);
+	try {
+		const result = vacuumDatabase(resolveDbOpt(opts));
+		p.intro("codemem db vacuum");
+		p.log.success(`Vacuumed: ${result.path}`);
+		p.outro(`Size: ${result.sizeBytes.toLocaleString()} bytes`);
+	} catch (error) {
+		p.log.error(error instanceof Error ? error.message : String(error));
+		process.exitCode = 1;
+	}
 });
 dbCommand.addCommand(vacuumCmd);
-var pruneReplCmd = new Command("prune-replication-ops").configureHelp(helpStyle).description("Prune replication op history with approximate oldest-first retention, dry-run, and progress reporting").option("--dry-run", "show current size/targets without deleting").option("--max-age-days <days>", "retention age threshold in days", "30").option("--max-size-mb <mb>", "target replication log budget in MB", "512").option("--batch-ops <n>", "max ops deleted per batch", "5000").option("--batch-runtime-ms <ms>", "max runtime per batch in ms", "2000").option("--vacuum", "run VACUUM explicitly after prune completes");
+var pruneReplCmd = new Command("prune-replication-ops").configureHelp(helpStyle).description("Prune replication op history with approximate oldest-first retention, dry-run, and progress reporting").option("--dry-run", "show current size/targets without deleting").option("--max-age-days <days>", "retention age threshold in days", "30").option("--max-size-mb <mb>", "target per-scope replication log budget in MB", "512").option("--batch-ops <n>", "max ops deleted per batch", "5000").option("--batch-runtime-ms <ms>", "max runtime per batch in ms", "2000").option("--vacuum", "run VACUUM explicitly after prune completes");
 addDbOption(pruneReplCmd);
 pruneReplCmd.action((opts) => {
 	const dbPath = resolveDbPath(resolveDbOpt(opts));
@@ -3166,29 +3191,41 @@ pruneReplCmd.action((opts) => {
 		const beforeBytes = estimateReplicationOpsBytes(db);
 		p.intro("codemem db prune-replication-ops");
 		p.log.info(`Replication ops size: ${formatBytes(beforeBytes)}`);
-		p.log.info(`Policy: approximately prune oldest-first toward <= ${maxSizeMb} MB while removing history older than ${maxAgeDays} day(s), subject to per-pass batch/runtime limits`);
+		p.log.info(`Policy: across all replication scopes, approximately prune oldest-first toward <= ${maxSizeMb} MB PER SCOPE while removing history older than ${maxAgeDays} day(s), subject to per-pass batch/runtime limits (the size budget is applied per scope, matching the background retention runner; total ops across scopes may exceed it)`);
+		const scopeIds = listRetentionScopeIds(db);
 		const agePlan = planReplicationOpsAgePrune(db, maxAgeDays, batchOps);
-		if (agePlan.candidate_ops > 0) p.log.info(`Age pass plan: ${agePlan.candidate_ops.toLocaleString()} ops, ~${formatBytes(agePlan.estimated_candidate_bytes)} in ~${agePlan.estimated_batches.toLocaleString()} batch(es), cutoff ${agePlan.cutoff_cursor}`);
+		if (agePlan.candidate_ops > 0) p.log.info(`Age pass plan (all scopes): ${agePlan.candidate_ops.toLocaleString()} ops, ~${formatBytes(agePlan.estimated_candidate_bytes)} in ~${agePlan.estimated_batches.toLocaleString()} batch(es), cutoff ${agePlan.cutoff_cursor}`);
 		else p.log.info("Age pass plan: no ops older than cutoff");
 		if (opts.dryRun) {
 			p.outro("Dry run only; no changes made");
 			return;
 		}
-		const loopResult = pruneReplicationOpsUntilCaughtUp(db, {
-			maxAgeDays,
-			maxSizeBytes: maxSizeMb * 1024 * 1024,
-			maxDeleteOps: batchOps,
-			maxRuntimeMs: batchRuntimeMs,
-			onPass: (pass) => {
-				if (pass.deleted === 0 && !pass.stoppedByBudget) return;
-				const suffix = pass.stoppedByBudget ? " (budget-limited)" : "";
-				p.log.step(`Pass ${pass.passNumber}: deleted ${pass.deleted.toLocaleString()} ops, remaining size ~${formatBytes(pass.afterBytes)}${suffix}`);
-			}
-		});
-		p.log.info(`Deleted ops: ${loopResult.totalDeleted.toLocaleString()}`);
-		p.log.info(`Estimated replication ops size after prune: ${formatBytes(loopResult.afterBytes)} (approximate)`);
-		if (loopResult.lastFloor) p.log.info(`Retained floor: ${loopResult.lastFloor}`);
-		if (loopResult.stoppedByBudget) p.log.warn("Prune stopped because the current batch/runtime budget was exhausted before retention caught up. Re-run with higher --batch-ops or --batch-runtime-ms for faster catch-up.");
+		let totalDeleted = 0;
+		let anyStoppedByBudget = false;
+		let lastFloor = null;
+		for (const scopeId of scopeIds) {
+			const loopResult = pruneReplicationOpsUntilCaughtUp(db, {
+				maxAgeDays,
+				maxSizeBytes: maxSizeMb * 1024 * 1024,
+				maxDeleteOps: batchOps,
+				maxRuntimeMs: batchRuntimeMs,
+				scopeId,
+				onPass: (pass) => {
+					if (pass.deleted === 0 && !pass.stoppedByBudget) return;
+					const suffix = pass.stoppedByBudget ? " (budget-limited)" : "";
+					p.log.step(`scope ${scopeId} pass ${pass.passNumber}: deleted ${pass.deleted.toLocaleString()} ops${suffix}`);
+				}
+			});
+			totalDeleted += loopResult.totalDeleted;
+			if (loopResult.totalDeleted > 0) p.log.info(`scope ${scopeId}: deleted ${loopResult.totalDeleted.toLocaleString()} ops`);
+			if (loopResult.lastFloor) lastFloor = loopResult.lastFloor;
+			if (loopResult.stoppedByBudget) anyStoppedByBudget = true;
+		}
+		p.log.info(`Deleted ops (all scopes): ${totalDeleted.toLocaleString()}`);
+		const afterBytes = estimateReplicationOpsBytes(db);
+		p.log.info(`Estimated replication ops size after prune: ${formatBytes(afterBytes)} (approximate)`);
+		if (lastFloor) p.log.info(`Retained floor: ${lastFloor}`);
+		if (anyStoppedByBudget) p.log.warn("Prune stopped because the current batch/runtime budget was exhausted before retention caught up. Re-run with higher --batch-ops or --batch-runtime-ms for faster catch-up.");
 		if (opts.vacuum) {
 			p.log.step("Running VACUUM as requested...");
 			db.close();
@@ -3203,31 +3240,88 @@ pruneReplCmd.action((opts) => {
 	}
 });
 dbCommand.addCommand(pruneReplCmd);
+var pruneRawEventsCmd = new Command("prune-raw-events").configureHelp(helpStyle).description("Delete raw_events older than a cutoff (age-based), with dry-run and VACUUM support").option("--dry-run", "show current size/targets without deleting").option("--max-age-days <days>", "retention age threshold in days", "90").option("--vacuum", "run VACUUM explicitly after prune completes");
+addDbOption(pruneRawEventsCmd);
+pruneRawEventsCmd.action((opts) => {
+	const rawAge = opts.maxAgeDays.trim();
+	const maxAgeDays = Number.parseInt(rawAge, 10);
+	if (!/^\d+$/.test(rawAge) || !Number.isInteger(maxAgeDays) || maxAgeDays < 1) {
+		p.log.error(`--max-age-days must be a positive integer (got "${opts.maxAgeDays}"). Refusing to run a destructive prune on invalid input.`);
+		process.exitCode = 1;
+		return;
+	}
+	const dbPath = resolveDbPath(resolveDbOpt(opts));
+	let store = null;
+	try {
+		store = new MemoryStore(dbPath);
+		const maxAgeMs = maxAgeDays * 864e5;
+		const status = store.rawEventsRetentionStatus(maxAgeMs);
+		p.intro("codemem db prune-raw-events");
+		p.log.warn("raw_events is the re-extraction source. Age-based purge is NOT extraction-aware: pruning a window shorter than your extraction lag can delete un-extracted events. The 90-day default mitigates this.");
+		p.log.info(`raw_events: ${status.total_rows.toLocaleString()} rows, ~${formatBytes(status.estimated_bytes)} on disk`);
+		p.log.info(`Older than ${maxAgeDays} day(s): ${status.candidate_rows.toLocaleString()} row(s) to delete`);
+		if (opts.dryRun) {
+			p.outro("Dry run only; no changes made");
+			return;
+		}
+		const deleted = store.purgeRawEvents(maxAgeMs);
+		p.log.info(`Deleted raw_events: ${deleted.toLocaleString()}`);
+		if (opts.vacuum) {
+			p.log.step("Running VACUUM as requested...");
+			store.close();
+			store = null;
+			const vacuumed = vacuumDatabase(dbPath);
+			p.outro(`Done. VACUUM complete. File size is now ${formatBytes(vacuumed.sizeBytes)}.`);
+			return;
+		}
+		p.outro("Done. SQLite file size may not shrink until you run `codemem db vacuum` explicitly (or re-run this command with --vacuum).");
+	} catch (error) {
+		p.log.error(error instanceof Error ? error.message : String(error));
+		process.exitCode = 1;
+	} finally {
+		store?.close();
+	}
+});
+dbCommand.addCommand(pruneRawEventsCmd);
 var rawEventsStatusCmd = new Command("raw-events-status").configureHelp(helpStyle).description("Show pending raw-event backlog by source stream").option("-n, --limit <n>", "max rows to show", "25");
 addDbOption(rawEventsStatusCmd);
 addJsonOption(rawEventsStatusCmd);
 rawEventsStatusCmd.action((opts) => {
-	const result = getRawEventStatus(resolveDbOpt(opts), Number.parseInt(opts.limit, 10) || 25);
-	if (opts.json) {
-		console.log(JSON.stringify(result, null, 2));
-		return;
-	}
-	p.intro("codemem db raw-events-status");
-	p.log.info(`Totals: ${result.totals.pending.toLocaleString()} pending across ${result.totals.sessions.toLocaleString()} session(s)`);
-	if (result.items.length === 0) {
-		p.outro("No pending raw events");
-		return;
+	try {
+		const result = getRawEventStatus(resolveDbOpt(opts), Number.parseInt(opts.limit, 10) || 25);
+		if (opts.json) {
+			console.log(JSON.stringify(result, null, 2));
+			return;
+		}
+		p.intro("codemem db raw-events-status");
+		p.log.info(`Totals: ${result.totals.pending.toLocaleString()} pending across ${result.totals.sessions.toLocaleString()} session(s)`);
+		if (result.items.length === 0) {
+			p.outro("No pending raw events");
+			return;
+		}
+		for (const item of result.items) p.log.message(`${item.source}:${item.stream_id} pending=${Math.max(0, item.last_received_event_seq - item.last_flushed_event_seq)} received=${item.last_received_event_seq} flushed=${item.last_flushed_event_seq} project=${item.project ?? ""}`);
+		p.outro("done");
+	} catch (error) {
+		const message = error instanceof Error ? error.message : "Failed to read raw-event status";
+		if (opts.json) emitJsonError("raw_events_status_failed", message);
+		else {
+			p.log.error(message);
+			process.exitCode = 1;
+		}
 	}
-	for (const item of result.items) p.log.message(`${item.source}:${item.stream_id} pending=${Math.max(0, item.last_received_event_seq - item.last_flushed_event_seq)} received=${item.last_received_event_seq} flushed=${item.last_flushed_event_seq} project=${item.project ?? ""}`);
-	p.outro("done");
 });
 dbCommand.addCommand(rawEventsStatusCmd);
 var rawEventsRetryCmd = new Command("raw-events-retry").configureHelp(helpStyle).description("Requeue failed raw-event flush batches").option("-n, --limit <n>", "max failed batches to requeue", "25");
 addDbOption(rawEventsRetryCmd);
 rawEventsRetryCmd.action((opts) => {
-	const result = retryRawEventFailures(resolveDbOpt(opts), Number.parseInt(opts.limit, 10) || 25);
-	p.intro("codemem db raw-events-retry");
-	p.outro(`Requeued ${result.retried.toLocaleString()} failed batch(es)`);
+	try {
+		const result = retryRawEventFailures(resolveDbOpt(opts), Number.parseInt(opts.limit, 10) || 25);
+		p.intro("codemem db raw-events-retry");
+		p.outro(`Requeued ${result.retried.toLocaleString()} failed batch(es)`);
+	} catch (error) {
+		p.log.error(error instanceof Error ? error.message : String(error));
+		process.exitCode = 1;
+	}
 });
 dbCommand.addCommand(rawEventsRetryCmd);
 var rawEventsGateCmd = new Command("raw-events-gate").configureHelp(helpStyle).description("Validate raw-event reliability thresholds (non-zero exit on failure)").option("--min-flush-success-rate <rate>", "minimum flush success rate", "0.95").option("--max-dropped-event-rate <rate>", "maximum dropped event rate", "0.05").option("--min-session-boundary-accuracy <rate>", "minimum session boundary accuracy", "0.9").option("--window-hours <hours>", "lookback window in hours", "24");
@@ -3287,6 +3381,9 @@ renameProjectCmd.action((oldName, newName, opts) => {
 		p.log.info([`Sessions: ${counts.sessions}`, `Raw event sessions: ${counts.raw_event_sessions}`].join("\n"));
 		if (dryRun) p.outro("Pass --apply to execute");
 		else p.outro("done");
+	} catch (error) {
+		p.log.error(error instanceof Error ? error.message : String(error));
+		process.exitCode = 1;
 	} finally {
 		store.close();
 	}
@@ -3332,6 +3429,9 @@ normalizeProjectsCmd.action((opts) => {
 		}
 		if (dryRun) p.outro("Pass --apply to execute");
 		else p.outro("done");
+	} catch (error) {
+		p.log.error(error instanceof Error ? error.message : String(error));
+		process.exitCode = 1;
 	} finally {
 		store.close();
 	}
@@ -4429,58 +4529,68 @@ function createMemoryRoleReportCommand() {
 	addDbOption(cmd);
 	addJsonOption(cmd);
 	cmd.action((opts) => {
-		const project = opts.allProjects === true ? null : opts.project?.trim() || process.env.CODEMEM_PROJECT?.trim() || resolveProject(process.cwd(), null);
-		const invalidScenario = (opts.scenario ?? []).find((id) => getInjectionEvalScenarioPack(id) == null);
-		if (invalidScenario) throw new Error(`Unknown eval scenario pack: ${invalidScenario}`);
-		const probes = [...opts.probe ?? [], ...getInjectionEvalScenarioPrompts(opts.scenario ?? [])];
-		const result = getMemoryRoleReport(resolveDbOpt(opts), {
-			project,
-			allProjects: opts.allProjects === true,
-			includeInactive: opts.inactive === true,
-			probes
-		});
-		if (opts.json) {
-			console.log(JSON.stringify(result, null, 2));
-			return;
-		}
-		p.intro("codemem memory role-report");
-		p.log.info([
-			`Memories: ${result.totals.memories}`,
-			`Active: ${result.totals.active}`,
-			`Sessions: ${result.totals.sessions}`
-		].join("\n"));
-		p.log.info("Counts by role:");
-		for (const [role, count] of Object.entries(result.counts_by_role)) p.log.message(`  ${role.padEnd(10)} ${String(count)}`);
-		p.log.info("Counts by mapping:");
-		p.log.message(`  mapped      ${result.counts_by_mapping.mapped}`);
-		p.log.message(`  unmapped    ${result.counts_by_mapping.unmapped}`);
-		p.log.info("Summary lineages:");
-		p.log.message(`  session_summary         ${result.summary_lineages.session_summary}`);
-		p.log.message(`  legacy_metadata_summary ${result.summary_lineages.legacy_metadata_summary}`);
-		p.log.message(`  summary_mapped          ${result.summary_mapping.mapped}`);
-		p.log.message(`  summary_unmapped        ${result.summary_mapping.unmapped}`);
-		p.log.info("Project quality:");
-		for (const [bucket, count] of Object.entries(result.project_quality)) p.log.message(`  ${bucket.padEnd(12)} ${String(count)}`);
-		p.log.info("Session classes:");
-		for (const [bucket, count] of Object.entries(result.session_class_buckets)) p.log.message(`  ${bucket.padEnd(20)} ${String(count)}`);
-		p.log.info("Summary dispositions:");
-		for (const [bucket, count] of Object.entries(result.summary_disposition_buckets)) p.log.message(`  ${bucket.padEnd(20)} ${String(count)}`);
-		if (result.probe_results.length > 0) {
-			p.log.info("Probe results:");
-			for (const probe of result.probe_results) {
-				p.log.message(`  query: ${probe.query}`);
-				if (probe.scenario_id) p.log.message(`    scenario: ${probe.scenario_id} (${probe.scenario_category ?? "unknown"})${probe.scenario_title ? ` — ${probe.scenario_title}` : ""}`);
-				p.log.message(`    mode: ${probe.mode}`);
-				p.log.message(`    top roles: durable=${probe.top_role_counts.durable} recap=${probe.top_role_counts.recap} ephemeral=${probe.top_role_counts.ephemeral} general=${probe.top_role_counts.general}`);
-				p.log.message(`    top mapping: mapped=${probe.top_mapping_counts.mapped} unmapped=${probe.top_mapping_counts.unmapped}`);
-				p.log.message(`    burden: recap_share=${probe.top_burden.recap_share.toFixed(2)} unmapped_share=${probe.top_burden.unmapped_share.toFixed(2)} recap_unmapped_share=${probe.top_burden.recap_unmapped_share.toFixed(2)}`);
-				if (probe.simulated_demoted_unmapped_recap) p.log.message(`    simulated demote-unmapped-recap burden: recap_share=${probe.simulated_demoted_unmapped_recap.top_burden.recap_share.toFixed(2)} unmapped_share=${probe.simulated_demoted_unmapped_recap.top_burden.unmapped_share.toFixed(2)} recap_unmapped_share=${probe.simulated_demoted_unmapped_recap.top_burden.recap_unmapped_share.toFixed(2)}`);
-				if (probe.simulated_demoted_unmapped_recap_and_ephemeral) p.log.message(`    simulated demote-unmapped-recap+ephemeral burden: recap_share=${probe.simulated_demoted_unmapped_recap_and_ephemeral.top_burden.recap_share.toFixed(2)} unmapped_share=${probe.simulated_demoted_unmapped_recap_and_ephemeral.top_burden.unmapped_share.toFixed(2)} recap_unmapped_share=${probe.simulated_demoted_unmapped_recap_and_ephemeral.top_burden.recap_unmapped_share.toFixed(2)}`);
-				if (probe.scenario_score) p.log.message(`    scenario score: mode_match=${probe.scenario_score.mode_match ? "yes" : "no"} top1_primary=${probe.scenario_score.primary_in_top1 ? "yes" : "no"} top3_primary=${probe.scenario_score.primary_in_top3_count} top1_anti=${probe.scenario_score.anti_signal_in_top1 ? "yes" : "no"} primary=${probe.scenario_score.primary_match_count} anti=${probe.scenario_score.anti_signal_count} recap=${probe.scenario_score.recap_count} unmapped_recap=${probe.scenario_score.unmapped_recap_count} chatter=${probe.scenario_score.administrative_chatter_count} net=${probe.scenario_score.score}`);
-				for (const item of probe.items.slice(0, 5)) p.log.message(`    [${item.id}] (${item.kind}/${item.role}/${item.mapping}/${item.session_class}/${item.summary_disposition}) ${item.title} — ${item.role_reason}`);
+		try {
+			const project = opts.allProjects === true ? null : opts.project?.trim() || process.env.CODEMEM_PROJECT?.trim() || resolveProject(process.cwd(), null);
+			const invalidScenario = (opts.scenario ?? []).find((id) => getInjectionEvalScenarioPack(id) == null);
+			if (invalidScenario) throw new Error(`Unknown eval scenario pack: ${invalidScenario}`);
+			const probes = [...opts.probe ?? [], ...getInjectionEvalScenarioPrompts(opts.scenario ?? [])];
+			const result = getMemoryRoleReport(resolveDbOpt(opts), {
+				project,
+				allProjects: opts.allProjects === true,
+				includeInactive: opts.inactive === true,
+				probes
+			});
+			if (opts.json) {
+				console.log(JSON.stringify(result, null, 2));
+				return;
+			}
+			p.intro("codemem memory role-report");
+			p.log.info([
+				`Memories: ${result.totals.memories}`,
+				`Active: ${result.totals.active}`,
+				`Sessions: ${result.totals.sessions}`
+			].join("\n"));
+			p.log.info("Counts by role:");
+			for (const [role, count] of Object.entries(result.counts_by_role)) p.log.message(`  ${role.padEnd(10)} ${String(count)}`);
+			p.log.info("Counts by mapping:");
+			p.log.message(`  mapped      ${result.counts_by_mapping.mapped}`);
+			p.log.message(`  unmapped    ${result.counts_by_mapping.unmapped}`);
+			p.log.info("Summary lineages:");
+			p.log.message(`  session_summary         ${result.summary_lineages.session_summary}`);
+			p.log.message(`  legacy_metadata_summary ${result.summary_lineages.legacy_metadata_summary}`);
+			p.log.message(`  summary_mapped          ${result.summary_mapping.mapped}`);
+			p.log.message(`  summary_unmapped        ${result.summary_mapping.unmapped}`);
+			p.log.info("Project quality:");
+			for (const [bucket, count] of Object.entries(result.project_quality)) p.log.message(`  ${bucket.padEnd(12)} ${String(count)}`);
+			p.log.info("Session classes:");
+			for (const [bucket, count] of Object.entries(result.session_class_buckets)) p.log.message(`  ${bucket.padEnd(20)} ${String(count)}`);
+			p.log.info("Summary dispositions:");
+			for (const [bucket, count] of Object.entries(result.summary_disposition_buckets)) p.log.message(`  ${bucket.padEnd(20)} ${String(count)}`);
+			if (result.probe_results.length > 0) {
+				p.log.info("Probe results:");
+				for (const probe of result.probe_results) {
+					p.log.message(`  query: ${probe.query}`);
+					if (probe.scenario_id) p.log.message(`    scenario: ${probe.scenario_id} (${probe.scenario_category ?? "unknown"})${probe.scenario_title ? ` — ${probe.scenario_title}` : ""}`);
+					p.log.message(`    mode: ${probe.mode}`);
+					p.log.message(`    top roles: durable=${probe.top_role_counts.durable} recap=${probe.top_role_counts.recap} ephemeral=${probe.top_role_counts.ephemeral} general=${probe.top_role_counts.general}`);
+					p.log.message(`    top mapping: mapped=${probe.top_mapping_counts.mapped} unmapped=${probe.top_mapping_counts.unmapped}`);
+					p.log.message(`    burden: recap_share=${probe.top_burden.recap_share.toFixed(2)} unmapped_share=${probe.top_burden.unmapped_share.toFixed(2)} recap_unmapped_share=${probe.top_burden.recap_unmapped_share.toFixed(2)}`);
+					if (probe.simulated_demoted_unmapped_recap) p.log.message(`    simulated demote-unmapped-recap burden: recap_share=${probe.simulated_demoted_unmapped_recap.top_burden.recap_share.toFixed(2)} unmapped_share=${probe.simulated_demoted_unmapped_recap.top_burden.unmapped_share.toFixed(2)} recap_unmapped_share=${probe.simulated_demoted_unmapped_recap.top_burden.recap_unmapped_share.toFixed(2)}`);
+					if (probe.simulated_demoted_unmapped_recap_and_ephemeral) p.log.message(`    simulated demote-unmapped-recap+ephemeral burden: recap_share=${probe.simulated_demoted_unmapped_recap_and_ephemeral.top_burden.recap_share.toFixed(2)} unmapped_share=${probe.simulated_demoted_unmapped_recap_and_ephemeral.top_burden.unmapped_share.toFixed(2)} recap_unmapped_share=${probe.simulated_demoted_unmapped_recap_and_ephemeral.top_burden.recap_unmapped_share.toFixed(2)}`);
+					if (probe.scenario_score) p.log.message(`    scenario score: mode_match=${probe.scenario_score.mode_match ? "yes" : "no"} top1_primary=${probe.scenario_score.primary_in_top1 ? "yes" : "no"} top3_primary=${probe.scenario_score.primary_in_top3_count} top1_anti=${probe.scenario_score.anti_signal_in_top1 ? "yes" : "no"} primary=${probe.scenario_score.primary_match_count} anti=${probe.scenario_score.anti_signal_count} recap=${probe.scenario_score.recap_count} unmapped_recap=${probe.scenario_score.unmapped_recap_count} chatter=${probe.scenario_score.administrative_chatter_count} net=${probe.scenario_score.score}`);
+					for (const item of probe.items.slice(0, 5)) p.log.message(`    [${item.id}] (${item.kind}/${item.role}/${item.mapping}/${item.session_class}/${item.summary_disposition}) ${item.title} — ${item.role_reason}`);
+				}
+			}
+			p.outro("done");
+		} catch (error) {
+			const message = error instanceof Error ? error.message : "Role report failed";
+			if (opts.json) emitJsonError("role_report_failed", message);
+			else {
+				p.log.error(message);
+				process.exitCode = 1;
 			}
+			return;
 		}
-		p.outro("done");
 	});
 	return cmd;
 }
@@ -4488,49 +4598,59 @@ function createMemoryRoleCompareCommand() {
 	const cmd = new Command("role-compare").configureHelp(helpStyle).description("Compare inferred memory-role and probe metrics across two DB snapshots").argument("<baseline_db>", "baseline sqlite database path").argument("<candidate_db>", "candidate sqlite database path").option("--project <project>", "project identifier (defaults to git repo root)").option("--all-projects", "analyze across all projects").option("--probe <query>", "run a retrieval probe query against both snapshots", (value, prev) => [...prev, value], []).option("--scenario <id>", "run a named injection-first eval scenario pack (can be repeated)", (value, prev) => [...prev, value], []).option("--inactive", "include inactive memories");
 	addJsonOption(cmd);
 	cmd.action((baselineDb, candidateDb, opts) => {
-		const project = opts.allProjects === true ? null : opts.project?.trim() || process.env.CODEMEM_PROJECT?.trim() || resolveProject(process.cwd(), null);
-		const invalidScenario = (opts.scenario ?? []).find((id) => getInjectionEvalScenarioPack(id) == null);
-		if (invalidScenario) throw new Error(`Unknown eval scenario pack: ${invalidScenario}`);
-		const probes = [...opts.probe ?? [], ...getInjectionEvalScenarioPrompts(opts.scenario ?? [])];
-		const result = compareMemoryRoleReports(baselineDb, candidateDb, {
-			project,
-			allProjects: opts.allProjects === true,
-			includeInactive: opts.inactive === true,
-			probes
-		});
-		if (opts.json) {
-			console.log(JSON.stringify(result, null, 2));
-			return;
-		}
-		p.intro("codemem memory role-compare");
-		p.log.info([
-			`Baseline sessions: ${result.baseline.totals.sessions}`,
-			`Candidate sessions: ${result.candidate.totals.sessions}`,
-			`Delta sessions: ${result.delta.totals.sessions}`,
-			`Mapped delta: ${result.delta.counts_by_mapping.mapped}`,
-			`Unmapped delta: ${result.delta.counts_by_mapping.unmapped}`,
-			`Summary mapped delta: ${result.delta.summary_mapping.mapped}`,
-			`Summary unmapped delta: ${result.delta.summary_mapping.unmapped}`
-		].join("\n"));
-		p.log.info("Role deltas:");
-		for (const [role, count] of Object.entries(result.delta.counts_by_role)) p.log.message(`  ${role.padEnd(10)} ${String(count)}`);
-		p.log.info("Session class deltas:");
-		for (const [bucket, count] of Object.entries(result.delta.session_class_buckets)) p.log.message(`  ${bucket.padEnd(20)} ${String(count)}`);
-		p.log.info("Summary disposition deltas:");
-		for (const [bucket, count] of Object.entries(result.delta.summary_disposition_buckets)) p.log.message(`  ${bucket.padEnd(20)} ${String(count)}`);
-		if (result.probe_comparisons.length > 0) {
-			p.log.info("Probe comparisons:");
-			for (const probe of result.probe_comparisons) {
-				p.log.message(`  query: ${probe.query}`);
-				p.log.message(`    modes: baseline=${probe.baseline_mode ?? "-"} candidate=${probe.candidate_mode ?? "-"}`);
-				p.log.message(`    overlap: shared_top_keys=${probe.shared_item_keys.length} baseline_top=${probe.baseline_item_ids.slice(0, 5).join(",") || "-"} candidate_top=${probe.candidate_item_ids.slice(0, 5).join(",") || "-"}`);
-				if (probe.delta_top_burden) p.log.message(`    burden delta: recap_share=${probe.delta_top_burden.recap_share.toFixed(2)} unmapped_share=${probe.delta_top_burden.unmapped_share.toFixed(2)} recap_unmapped_share=${probe.delta_top_burden.recap_unmapped_share.toFixed(2)}`);
-				if (probe.delta_top_mapping_counts) p.log.message(`    mapping delta: mapped=${probe.delta_top_mapping_counts.mapped} unmapped=${probe.delta_top_mapping_counts.unmapped}`);
-				if (probe.baseline_scenario_score || probe.candidate_scenario_score) p.log.message(`    scenario scores: baseline=${probe.baseline_scenario_score?.score ?? "-"} candidate=${probe.candidate_scenario_score?.score ?? "-"}`);
-				if (probe.delta_scenario_score) p.log.message(`    scenario delta: mode_match=${probe.delta_scenario_score.mode_match ?? "-"} top1_primary=${probe.delta_scenario_score.primary_in_top1 ?? "-"} top3_primary=${probe.delta_scenario_score.primary_in_top3_count ?? "-"} top1_anti=${probe.delta_scenario_score.anti_signal_in_top1 ?? "-"} primary=${probe.delta_scenario_score.primary_match_count ?? "-"} anti=${probe.delta_scenario_score.anti_signal_count ?? "-"} recap=${probe.delta_scenario_score.recap_count ?? "-"} unmapped_recap=${probe.delta_scenario_score.unmapped_recap_count ?? "-"} chatter=${probe.delta_scenario_score.administrative_chatter_count ?? "-"} net=${probe.delta_scenario_score.score ?? "-"}`);
+		try {
+			const project = opts.allProjects === true ? null : opts.project?.trim() || process.env.CODEMEM_PROJECT?.trim() || resolveProject(process.cwd(), null);
+			const invalidScenario = (opts.scenario ?? []).find((id) => getInjectionEvalScenarioPack(id) == null);
+			if (invalidScenario) throw new Error(`Unknown eval scenario pack: ${invalidScenario}`);
+			const probes = [...opts.probe ?? [], ...getInjectionEvalScenarioPrompts(opts.scenario ?? [])];
+			const result = compareMemoryRoleReports(baselineDb, candidateDb, {
+				project,
+				allProjects: opts.allProjects === true,
+				includeInactive: opts.inactive === true,
+				probes
+			});
+			if (opts.json) {
+				console.log(JSON.stringify(result, null, 2));
+				return;
+			}
+			p.intro("codemem memory role-compare");
+			p.log.info([
+				`Baseline sessions: ${result.baseline.totals.sessions}`,
+				`Candidate sessions: ${result.candidate.totals.sessions}`,
+				`Delta sessions: ${result.delta.totals.sessions}`,
+				`Mapped delta: ${result.delta.counts_by_mapping.mapped}`,
+				`Unmapped delta: ${result.delta.counts_by_mapping.unmapped}`,
+				`Summary mapped delta: ${result.delta.summary_mapping.mapped}`,
+				`Summary unmapped delta: ${result.delta.summary_mapping.unmapped}`
+			].join("\n"));
+			p.log.info("Role deltas:");
+			for (const [role, count] of Object.entries(result.delta.counts_by_role)) p.log.message(`  ${role.padEnd(10)} ${String(count)}`);
+			p.log.info("Session class deltas:");
+			for (const [bucket, count] of Object.entries(result.delta.session_class_buckets)) p.log.message(`  ${bucket.padEnd(20)} ${String(count)}`);
+			p.log.info("Summary disposition deltas:");
+			for (const [bucket, count] of Object.entries(result.delta.summary_disposition_buckets)) p.log.message(`  ${bucket.padEnd(20)} ${String(count)}`);
+			if (result.probe_comparisons.length > 0) {
+				p.log.info("Probe comparisons:");
+				for (const probe of result.probe_comparisons) {
+					p.log.message(`  query: ${probe.query}`);
+					p.log.message(`    modes: baseline=${probe.baseline_mode ?? "-"} candidate=${probe.candidate_mode ?? "-"}`);
+					p.log.message(`    overlap: shared_top_keys=${probe.shared_item_keys.length} baseline_top=${probe.baseline_item_ids.slice(0, 5).join(",") || "-"} candidate_top=${probe.candidate_item_ids.slice(0, 5).join(",") || "-"}`);
+					if (probe.delta_top_burden) p.log.message(`    burden delta: recap_share=${probe.delta_top_burden.recap_share.toFixed(2)} unmapped_share=${probe.delta_top_burden.unmapped_share.toFixed(2)} recap_unmapped_share=${probe.delta_top_burden.recap_unmapped_share.toFixed(2)}`);
+					if (probe.delta_top_mapping_counts) p.log.message(`    mapping delta: mapped=${probe.delta_top_mapping_counts.mapped} unmapped=${probe.delta_top_mapping_counts.unmapped}`);
+					if (probe.baseline_scenario_score || probe.candidate_scenario_score) p.log.message(`    scenario scores: baseline=${probe.baseline_scenario_score?.score ?? "-"} candidate=${probe.candidate_scenario_score?.score ?? "-"}`);
+					if (probe.delta_scenario_score) p.log.message(`    scenario delta: mode_match=${probe.delta_scenario_score.mode_match ?? "-"} top1_primary=${probe.delta_scenario_score.primary_in_top1 ?? "-"} top3_primary=${probe.delta_scenario_score.primary_in_top3_count ?? "-"} top1_anti=${probe.delta_scenario_score.anti_signal_in_top1 ?? "-"} primary=${probe.delta_scenario_score.primary_match_count ?? "-"} anti=${probe.delta_scenario_score.anti_signal_count ?? "-"} recap=${probe.delta_scenario_score.recap_count ?? "-"} unmapped_recap=${probe.delta_scenario_score.unmapped_recap_count ?? "-"} chatter=${probe.delta_scenario_score.administrative_chatter_count ?? "-"} net=${probe.delta_scenario_score.score ?? "-"}`);
+				}
+			}
+			p.outro("done");
+		} catch (error) {
+			const message = error instanceof Error ? error.message : "Role compare failed";
+			if (opts.json) emitJsonError("role_compare_failed", message);
+			else {
+				p.log.error(message);
+				process.exitCode = 1;
 			}
+			return;
 		}
-		p.outro("done");
 	});
 	return cmd;
 }
@@ -4539,55 +4659,65 @@ function createMemoryExtractionReportCommand() {
 	addDbOption(cmd);
 	addJsonOption(cmd);
 	cmd.action((opts) => {
-		const sessionIdInput = opts.sessionId?.trim() ?? "";
-		const batchIdInput = opts.batchId?.trim() ?? "";
-		const hasSessionId = sessionIdInput.length > 0;
-		const hasBatchId = batchIdInput.length > 0;
-		if (hasSessionId === hasBatchId) throw new Error("Provide exactly one of --session-id or --batch-id");
-		const sessionId = hasSessionId ? parseStrictPositiveId(sessionIdInput) : null;
-		if (hasSessionId && sessionId === null) throw new Error(`Invalid session ID: ${sessionIdInput || opts.sessionId}`);
-		const batchId = hasBatchId ? parseStrictPositiveId(batchIdInput) : null;
-		if (hasBatchId && batchId === null) throw new Error(`Invalid batch ID: ${batchIdInput || opts.batchId}`);
-		const scenarioId = opts.scenario?.trim() ?? "";
-		const scenario = getSessionExtractionEvalScenario(scenarioId);
-		if (!scenario) throw new Error(`Unknown extraction eval scenario: ${scenarioId || opts.scenario}`);
-		const result = batchId != null ? getSessionExtractionEval(resolveDbOpt(opts), {
-			batchId,
-			scenarioId: scenario.id,
-			includeInactive: opts.inactive === true
-		}) : getSessionExtractionEval(resolveDbOpt(opts), {
-			sessionId,
-			scenarioId: scenario.id,
-			includeInactive: opts.inactive === true
-		});
-		if (opts.json) {
-			console.log(JSON.stringify(result, null, 2));
+		try {
+			const sessionIdInput = opts.sessionId?.trim() ?? "";
+			const batchIdInput = opts.batchId?.trim() ?? "";
+			const hasSessionId = sessionIdInput.length > 0;
+			const hasBatchId = batchIdInput.length > 0;
+			if (hasSessionId === hasBatchId) throw new Error("Provide exactly one of --session-id or --batch-id");
+			const sessionId = hasSessionId ? parseStrictPositiveId(sessionIdInput) : null;
+			if (hasSessionId && sessionId === null) throw new Error(`Invalid session ID: ${sessionIdInput || opts.sessionId}`);
+			const batchId = hasBatchId ? parseStrictPositiveId(batchIdInput) : null;
+			if (hasBatchId && batchId === null) throw new Error(`Invalid batch ID: ${batchIdInput || opts.batchId}`);
+			const scenarioId = opts.scenario?.trim() ?? "";
+			const scenario = getSessionExtractionEvalScenario(scenarioId);
+			if (!scenario) throw new Error(`Unknown extraction eval scenario: ${scenarioId || opts.scenario}`);
+			const result = batchId != null ? getSessionExtractionEval(resolveDbOpt(opts), {
+				batchId,
+				scenarioId: scenario.id,
+				includeInactive: opts.inactive === true
+			}) : getSessionExtractionEval(resolveDbOpt(opts), {
+				sessionId,
+				scenarioId: scenario.id,
+				includeInactive: opts.inactive === true
+			});
+			if (opts.json) {
+				console.log(JSON.stringify(result, null, 2));
+				return;
+			}
+			p.intro("codemem memory extraction-report");
+			p.log.info([
+				`Scenario: ${result.scenario.id} — ${result.scenario.title}`,
+				`Target: ${result.target.type}${result.target.batchId != null ? ` #${result.target.batchId}` : ""}`,
+				`Session: ${result.session.id} (${result.session.project ?? "no-project"})`,
+				`Session class: ${result.session.sessionClass}`,
+				`Summary disposition: ${result.session.summaryDisposition}`
+			].join("\n"));
+			p.log.info([
+				`Pass: ${result.pass ? "yes" : "no"}`,
+				`Summary count: ${result.counts.summaries}`,
+				`Observation count: ${result.counts.observations}`,
+				`Summary thread coverage: ${result.coverage.summaryThreadCoverage}`,
+				`Observation thread coverage: ${result.coverage.observationThreadCoverage}`,
+				`Total thread coverage: ${result.coverage.totalThreadCoverage}`,
+				`Duplicate observation threads: ${result.coverage.duplicateObservationThreads}`
+			].join("\n"));
+			if (result.failureReasons.length > 0) {
+				p.log.warn("Failure reasons:");
+				for (const reason of result.failureReasons) p.log.message(`  - ${reason}`);
+			}
+			p.log.info("Thread coverage:");
+			for (const thread of result.threads) p.log.message(`  ${thread.id.padEnd(22)} summary=${thread.summaryMatch ? "yes" : "no"} observations=${thread.observationMatch ? "yes" : "no"}`);
+			p.outro("done");
+		} catch (error) {
+			const message = error instanceof Error ? error.message : "Extraction report failed";
+			if (opts.json) emitJsonError("extraction_report_failed", message);
+			else {
+				p.log.error(message);
+				process.exitCode = 1;
+			}
 			return;
 		}
-		p.intro("codemem memory extraction-report");
-		p.log.info([
-			`Scenario: ${result.scenario.id} — ${result.scenario.title}`,
-			`Target: ${result.target.type}${result.target.batchId != null ? ` #${result.target.batchId}` : ""}`,
-			`Session: ${result.session.id} (${result.session.project ?? "no-project"})`,
-			`Session class: ${result.session.sessionClass}`,
-			`Summary disposition: ${result.session.summaryDisposition}`
-		].join("\n"));
-		p.log.info([
-			`Pass: ${result.pass ? "yes" : "no"}`,
-			`Summary count: ${result.counts.summaries}`,
-			`Observation count: ${result.counts.observations}`,
-			`Summary thread coverage: ${result.coverage.summaryThreadCoverage}`,
-			`Observation thread coverage: ${result.coverage.observationThreadCoverage}`,
-			`Total thread coverage: ${result.coverage.totalThreadCoverage}`,
-			`Duplicate observation threads: ${result.coverage.duplicateObservationThreads}`
-		].join("\n"));
-		if (result.failureReasons.length > 0) {
-			p.log.warn("Failure reasons:");
-			for (const reason of result.failureReasons) p.log.message(`  - ${reason}`);
-		}
-		p.log.info("Thread coverage:");
-		for (const thread of result.threads) p.log.message(`  ${thread.id.padEnd(22)} summary=${thread.summaryMatch ? "yes" : "no"} observations=${thread.observationMatch ? "yes" : "no"}`);
-		p.outro("done");
 	});
 	return cmd;
 }
@@ -4596,73 +4726,83 @@ function createMemoryExtractionReplayCommand() {
 	addDbOption(cmd);
 	addJsonOption(cmd);
 	cmd.action(async (opts) => {
-		const batchIdInput = opts.batchId?.trim() ?? "";
-		const batchId = parseStrictPositiveId(batchIdInput);
-		if (batchId === null) throw new Error(`Invalid batch ID: ${batchIdInput || opts.batchId}`);
-		const scenarioId = opts.scenario?.trim() ?? "";
-		const scenario = getSessionExtractionEvalScenario(scenarioId);
-		if (!scenario) throw new Error(`Unknown extraction eval scenario: ${scenarioId || opts.scenario}`);
-		const transcriptBudgetInput = opts.transcriptBudget?.trim() ?? "";
-		const transcriptBudget = transcriptBudgetInput.length > 0 ? parseStrictPositiveId(transcriptBudgetInput) : null;
-		if (transcriptBudgetInput.length > 0 && transcriptBudget === null) throw new Error(`Invalid transcript budget: ${transcriptBudgetInput || opts.transcriptBudget}`);
-		const observerTemperatureInput = opts.observerTemperature?.trim() ?? "";
-		let observerTemperature;
-		if (observerTemperatureInput.length > 0) {
-			const parsed = Number(observerTemperatureInput);
-			if (!Number.isFinite(parsed)) throw new Error(`Invalid observer temperature: ${observerTemperatureInput || opts.observerTemperature}`);
-			observerTemperature = parsed;
-		}
-		const maxOutputTokensInput = opts.maxOutputTokens?.trim() ?? "";
-		const maxOutputTokens = maxOutputTokensInput.length > 0 ? parseStrictPositiveId(maxOutputTokensInput) : null;
-		if (maxOutputTokensInput.length > 0 && maxOutputTokens === null) throw new Error(`Invalid max output tokens: ${maxOutputTokensInput || opts.maxOutputTokens}`);
-		const observerConfig = loadObserverConfig();
-		const observerConfigWithOverrides = {
-			...observerConfig,
-			observerTemperature: observerTemperature ?? observerConfig.observerTemperature,
-			observerOpenAIUseResponses: opts.openaiResponses === true,
-			observerReasoningEffort: opts.reasoningEffort?.trim() || null,
-			observerReasoningSummary: opts.reasoningSummary?.trim() || null,
-			observerMaxOutputTokens: maxOutputTokens ?? observerConfig.observerMaxTokens
-		};
-		const observer = new ObserverClient(observerConfigWithOverrides);
-		const result = opts.observerTierRouting === true ? await replayBatchExtractionWithTierRouting(resolveDbOpt(opts), observerConfigWithOverrides, {
-			batchId,
-			scenarioId: scenario.id,
-			transcriptBudget: transcriptBudget ?? void 0
-		}) : await replayBatchExtraction(resolveDbOpt(opts), observer, {
-			batchId,
-			scenarioId: scenario.id,
-			transcriptBudget: transcriptBudget ?? void 0
-		});
-		if (opts.json) {
-			console.log(JSON.stringify(result, null, 2));
+		try {
+			const batchIdInput = opts.batchId?.trim() ?? "";
+			const batchId = parseStrictPositiveId(batchIdInput);
+			if (batchId === null) throw new Error(`Invalid batch ID: ${batchIdInput || opts.batchId}`);
+			const scenarioId = opts.scenario?.trim() ?? "";
+			const scenario = getSessionExtractionEvalScenario(scenarioId);
+			if (!scenario) throw new Error(`Unknown extraction eval scenario: ${scenarioId || opts.scenario}`);
+			const transcriptBudgetInput = opts.transcriptBudget?.trim() ?? "";
+			const transcriptBudget = transcriptBudgetInput.length > 0 ? parseStrictPositiveId(transcriptBudgetInput) : null;
+			if (transcriptBudgetInput.length > 0 && transcriptBudget === null) throw new Error(`Invalid transcript budget: ${transcriptBudgetInput || opts.transcriptBudget}`);
+			const observerTemperatureInput = opts.observerTemperature?.trim() ?? "";
+			let observerTemperature;
+			if (observerTemperatureInput.length > 0) {
+				const parsed = Number(observerTemperatureInput);
+				if (!Number.isFinite(parsed)) throw new Error(`Invalid observer temperature: ${observerTemperatureInput || opts.observerTemperature}`);
+				observerTemperature = parsed;
+			}
+			const maxOutputTokensInput = opts.maxOutputTokens?.trim() ?? "";
+			const maxOutputTokens = maxOutputTokensInput.length > 0 ? parseStrictPositiveId(maxOutputTokensInput) : null;
+			if (maxOutputTokensInput.length > 0 && maxOutputTokens === null) throw new Error(`Invalid max output tokens: ${maxOutputTokensInput || opts.maxOutputTokens}`);
+			const observerConfig = loadObserverConfig();
+			const observerConfigWithOverrides = {
+				...observerConfig,
+				observerTemperature: observerTemperature ?? observerConfig.observerTemperature,
+				observerOpenAIUseResponses: opts.openaiResponses === true,
+				observerReasoningEffort: opts.reasoningEffort?.trim() || null,
+				observerReasoningSummary: opts.reasoningSummary?.trim() || null,
+				observerMaxOutputTokens: maxOutputTokens ?? observerConfig.observerMaxTokens
+			};
+			const observer = new ObserverClient(observerConfigWithOverrides);
+			const result = opts.observerTierRouting === true ? await replayBatchExtractionWithTierRouting(resolveDbOpt(opts), observerConfigWithOverrides, {
+				batchId,
+				scenarioId: scenario.id,
+				transcriptBudget: transcriptBudget ?? void 0
+			}) : await replayBatchExtraction(resolveDbOpt(opts), observer, {
+				batchId,
+				scenarioId: scenario.id,
+				transcriptBudget: transcriptBudget ?? void 0
+			});
+			if (opts.json) {
+				console.log(JSON.stringify(result, null, 2));
+				return;
+			}
+			p.intro("codemem memory extraction-replay");
+			p.log.info([
+				`Scenario: ${result.scenario.id} — ${result.scenario.title}`,
+				`Batch: ${result.target.batchId}`,
+				`Session: ${result.target.sessionId}`,
+				`Observer: ${result.observer.provider}/${result.observer.model}`,
+				`Tier: ${result.observer.tier ?? "manual"}`,
+				`OpenAI Responses: ${result.observer.openaiUseResponses ? "yes" : "no"}`,
+				`Reasoning effort: ${result.observer.reasoningEffort ?? "none"}`,
+				`Classification: ${result.classification.status}`,
+				`Pass: ${result.evaluation.pass ? "yes" : "no"}`
+			].join("\n"));
+			if (result.classification.reason) p.log.message(`Classification reason: ${result.classification.reason}`);
+			if (result.evaluation.failureReasons.length > 0) {
+				p.log.warn("Failure reasons:");
+				for (const reason of result.evaluation.failureReasons) p.log.message(`  - ${reason}`);
+			}
+			p.log.info([
+				`Fresh summaries: ${result.evaluation.counts.summaries}`,
+				`Fresh observations: ${result.evaluation.counts.observations}`,
+				`Summary thread coverage: ${result.evaluation.coverage.summaryThreadCoverage}`,
+				`Observation thread coverage: ${result.evaluation.coverage.observationThreadCoverage}`,
+				`Total thread coverage: ${result.evaluation.coverage.totalThreadCoverage}`
+			].join("\n"));
+			p.outro("done");
+		} catch (error) {
+			const message = error instanceof Error ? error.message : "Extraction replay failed";
+			if (opts.json) emitJsonError("extraction_replay_failed", message);
+			else {
+				p.log.error(message);
+				process.exitCode = 1;
+			}
 			return;
 		}
-		p.intro("codemem memory extraction-replay");
-		p.log.info([
-			`Scenario: ${result.scenario.id} — ${result.scenario.title}`,
-			`Batch: ${result.target.batchId}`,
-			`Session: ${result.target.sessionId}`,
-			`Observer: ${result.observer.provider}/${result.observer.model}`,
-			`Tier: ${result.observer.tier ?? "manual"}`,
-			`OpenAI Responses: ${result.observer.openaiUseResponses ? "yes" : "no"}`,
-			`Reasoning effort: ${result.observer.reasoningEffort ?? "none"}`,
-			`Classification: ${result.classification.status}`,
-			`Pass: ${result.evaluation.pass ? "yes" : "no"}`
-		].join("\n"));
-		if (result.classification.reason) p.log.message(`Classification reason: ${result.classification.reason}`);
-		if (result.evaluation.failureReasons.length > 0) {
-			p.log.warn("Failure reasons:");
-			for (const reason of result.evaluation.failureReasons) p.log.message(`  - ${reason}`);
-		}
-		p.log.info([
-			`Fresh summaries: ${result.evaluation.counts.summaries}`,
-			`Fresh observations: ${result.evaluation.counts.observations}`,
-			`Summary thread coverage: ${result.evaluation.coverage.summaryThreadCoverage}`,
-			`Observation thread coverage: ${result.evaluation.coverage.observationThreadCoverage}`,
-			`Total thread coverage: ${result.evaluation.coverage.totalThreadCoverage}`
-		].join("\n"));
-		p.outro("done");
 	});
 	return cmd;
 }
@@ -4671,140 +4811,150 @@ function createMemoryExtractionBenchmarkCommand() {
 	addDbOption(cmd);
 	addJsonOption(cmd);
 	cmd.action(async (opts) => {
-		const benchmarkId = opts.benchmark?.trim() ?? "";
-		const benchmark = getExtractionBenchmarkProfile(benchmarkId);
-		if (!benchmark) throw new Error(`Unknown extraction benchmark: ${benchmarkId || opts.benchmark}`);
-		const transcriptBudgetInput = opts.transcriptBudget?.trim() ?? "";
-		const transcriptBudget = transcriptBudgetInput.length > 0 ? parseStrictPositiveId(transcriptBudgetInput) : null;
-		if (transcriptBudgetInput.length > 0 && transcriptBudget === null) throw new Error(`Invalid transcript budget: ${transcriptBudgetInput || opts.transcriptBudget}`);
-		const observerTemperatureInput = opts.observerTemperature?.trim() ?? "";
-		let observerTemperature;
-		if (observerTemperatureInput.length > 0) {
-			const parsed = Number(observerTemperatureInput);
-			if (!Number.isFinite(parsed)) throw new Error(`Invalid observer temperature: ${observerTemperatureInput || opts.observerTemperature}`);
-			observerTemperature = parsed;
-		}
-		const maxOutputTokensInput = opts.maxOutputTokens?.trim() ?? "";
-		const maxOutputTokens = maxOutputTokensInput.length > 0 ? parseStrictPositiveId(maxOutputTokensInput) : null;
-		if (maxOutputTokensInput.length > 0 && maxOutputTokens === null) throw new Error(`Invalid max output tokens: ${maxOutputTokensInput || opts.maxOutputTokens}`);
-		const observerConfig = loadObserverConfig();
-		const observerConfigWithOverrides = {
-			...observerConfig,
-			observerProvider: opts.observerProvider?.trim() || observerConfig.observerProvider,
-			observerModel: opts.observerModel?.trim() || observerConfig.observerModel,
-			observerTemperature: observerTemperature ?? observerConfig.observerTemperature,
-			observerOpenAIUseResponses: opts.openaiResponses === true,
-			observerReasoningEffort: opts.reasoningEffort?.trim() || null,
-			observerReasoningSummary: opts.reasoningSummary?.trim() || null,
-			observerMaxOutputTokens: maxOutputTokens ?? observerConfig.observerMaxTokens
-		};
-		const observer = new ObserverClient(observerConfigWithOverrides);
-		const runs = [];
-		for (const batch of benchmark.batches) {
-			const scenarioId = batch.scenarioId ?? benchmark.scenarioId;
-			const result = opts.observerTierRouting === true ? await replayBatchExtractionWithTierRouting(resolveDbOpt(opts), observerConfigWithOverrides, {
-				batchId: batch.batchId,
-				scenarioId,
-				transcriptBudget: transcriptBudget ?? void 0
-			}) : await replayBatchExtraction(resolveDbOpt(opts), observer, {
-				batchId: batch.batchId,
-				scenarioId,
-				transcriptBudget: transcriptBudget ?? void 0
-			});
-			runs.push({
-				batchId: batch.batchId,
-				sessionId: batch.sessionId,
-				label: batch.label,
-				purpose: batch.purpose,
-				complexity: batch.complexity,
-				scenarioId,
-				expectedTier: batch.expectedTier ?? null,
-				analysis: {
-					eventSpan: result.analysis.eventSpan,
-					promptCount: result.analysis.promptCount,
-					toolCount: result.analysis.toolCount,
-					transcriptLength: result.analysis.transcriptLength
+		try {
+			const benchmarkId = opts.benchmark?.trim() ?? "";
+			const benchmark = getExtractionBenchmarkProfile(benchmarkId);
+			if (!benchmark) throw new Error(`Unknown extraction benchmark: ${benchmarkId || opts.benchmark}`);
+			const transcriptBudgetInput = opts.transcriptBudget?.trim() ?? "";
+			const transcriptBudget = transcriptBudgetInput.length > 0 ? parseStrictPositiveId(transcriptBudgetInput) : null;
+			if (transcriptBudgetInput.length > 0 && transcriptBudget === null) throw new Error(`Invalid transcript budget: ${transcriptBudgetInput || opts.transcriptBudget}`);
+			const observerTemperatureInput = opts.observerTemperature?.trim() ?? "";
+			let observerTemperature;
+			if (observerTemperatureInput.length > 0) {
+				const parsed = Number(observerTemperatureInput);
+				if (!Number.isFinite(parsed)) throw new Error(`Invalid observer temperature: ${observerTemperatureInput || opts.observerTemperature}`);
+				observerTemperature = parsed;
+			}
+			const maxOutputTokensInput = opts.maxOutputTokens?.trim() ?? "";
+			const maxOutputTokens = maxOutputTokensInput.length > 0 ? parseStrictPositiveId(maxOutputTokensInput) : null;
+			if (maxOutputTokensInput.length > 0 && maxOutputTokens === null) throw new Error(`Invalid max output tokens: ${maxOutputTokensInput || opts.maxOutputTokens}`);
+			const observerConfig = loadObserverConfig();
+			const observerConfigWithOverrides = {
+				...observerConfig,
+				observerProvider: opts.observerProvider?.trim() || observerConfig.observerProvider,
+				observerModel: opts.observerModel?.trim() || observerConfig.observerModel,
+				observerTemperature: observerTemperature ?? observerConfig.observerTemperature,
+				observerOpenAIUseResponses: opts.openaiResponses === true,
+				observerReasoningEffort: opts.reasoningEffort?.trim() || null,
+				observerReasoningSummary: opts.reasoningSummary?.trim() || null,
+				observerMaxOutputTokens: maxOutputTokens ?? observerConfig.observerMaxTokens
+			};
+			const observer = new ObserverClient(observerConfigWithOverrides);
+			const runs = [];
+			for (const batch of benchmark.batches) {
+				const scenarioId = batch.scenarioId ?? benchmark.scenarioId;
+				const result = opts.observerTierRouting === true ? await replayBatchExtractionWithTierRouting(resolveDbOpt(opts), observerConfigWithOverrides, {
+					batchId: batch.batchId,
+					scenarioId,
+					transcriptBudget: transcriptBudget ?? void 0
+				}) : await replayBatchExtraction(resolveDbOpt(opts), observer, {
+					batchId: batch.batchId,
+					scenarioId,
+					transcriptBudget: transcriptBudget ?? void 0
+				});
+				runs.push({
+					batchId: batch.batchId,
+					sessionId: batch.sessionId,
+					label: batch.label,
+					purpose: batch.purpose,
+					complexity: batch.complexity,
+					scenarioId,
+					expectedTier: batch.expectedTier ?? null,
+					analysis: {
+						eventSpan: result.analysis.eventSpan,
+						promptCount: result.analysis.promptCount,
+						toolCount: result.analysis.toolCount,
+						transcriptLength: result.analysis.transcriptLength
+					},
+					status: result.classification.status,
+					reason: result.classification.reason,
+					tier: result.observer.tier ?? "manual",
+					provider: result.observer.provider,
+					model: result.observer.model,
+					openaiUseResponses: result.observer.openaiUseResponses,
+					reasoningEffort: result.observer.reasoningEffort,
+					reasoningSummary: result.observer.reasoningSummary,
+					maxOutputTokens: result.observer.maxOutputTokens,
+					temperature: result.observer.temperature,
+					summaries: result.evaluation.counts.summaries,
+					observations: result.evaluation.counts.observations,
+					repairApplied: result.observer.repairApplied
+				});
+			}
+			const summary = {
+				total: runs.length,
+				shapeQualityTotal: runs.filter((run) => run.purpose === "shape_quality").length,
+				shapeQualityPasses: runs.filter((run) => run.purpose === "shape_quality" && run.status === "pass").length,
+				shapeQualityFails: runs.filter((run) => run.purpose === "shape_quality" && run.status === "shape_fail").length,
+				expectedTierTotal: runs.filter((run) => run.expectedTier != null).length,
+				expectedTierMatches: runs.filter((run) => run.expectedTier != null && run.expectedTier === run.tier).length,
+				robustnessNoOutput: runs.filter((run) => run.status === "observer_no_output").length
+			};
+			const uniqueObserverKeys = Array.from(new Set(runs.map((run) => `${run.provider}::${run.model}::${run.openaiUseResponses ? "responses" : "chat"}`)));
+			const observerSummary = opts.observerTierRouting === true ? {
+				provider: uniqueObserverKeys.length === 1 ? runs[0]?.provider ?? observer.provider : "mixed",
+				model: uniqueObserverKeys.length === 1 ? runs[0]?.model ?? observer.model : "mixed",
+				tierRouting: true,
+				openaiUseResponses: uniqueObserverKeys.length === 1 ? runs[0]?.openaiUseResponses ?? observer.openaiUseResponses : null,
+				reasoningEffort: uniqueObserverKeys.length === 1 ? runs[0]?.reasoningEffort ?? observer.reasoningEffort : "mixed",
+				reasoningSummary: uniqueObserverKeys.length === 1 ? runs[0]?.reasoningSummary ?? observer.reasoningSummary : "mixed",
+				maxOutputTokens: uniqueObserverKeys.length === 1 ? runs[0]?.maxOutputTokens ?? observer.maxOutputTokens : null,
+				temperature: uniqueObserverKeys.length === 1 ? runs[0]?.temperature ?? observer.temperature : null,
+				transcriptBudget: transcriptBudget ?? null,
+				selectedObservers: uniqueObserverKeys
+			} : {
+				provider: observer.provider,
+				model: observer.model,
+				tierRouting: false,
+				openaiUseResponses: observer.openaiUseResponses,
+				reasoningEffort: observer.reasoningEffort,
+				reasoningSummary: observer.reasoningSummary,
+				maxOutputTokens: observer.maxOutputTokens,
+				temperature: observer.temperature,
+				transcriptBudget: transcriptBudget ?? null,
+				selectedObservers: uniqueObserverKeys
+			};
+			const output = {
+				benchmark: {
+					id: benchmark.id,
+					title: benchmark.title,
+					scenarioId: benchmark.scenarioId
 				},
-				status: result.classification.status,
-				reason: result.classification.reason,
-				tier: result.observer.tier ?? "manual",
-				provider: result.observer.provider,
-				model: result.observer.model,
-				openaiUseResponses: result.observer.openaiUseResponses,
-				reasoningEffort: result.observer.reasoningEffort,
-				reasoningSummary: result.observer.reasoningSummary,
-				maxOutputTokens: result.observer.maxOutputTokens,
-				temperature: result.observer.temperature,
-				summaries: result.evaluation.counts.summaries,
-				observations: result.evaluation.counts.observations,
-				repairApplied: result.observer.repairApplied
-			});
-		}
-		const summary = {
-			total: runs.length,
-			shapeQualityTotal: runs.filter((run) => run.purpose === "shape_quality").length,
-			shapeQualityPasses: runs.filter((run) => run.purpose === "shape_quality" && run.status === "pass").length,
-			shapeQualityFails: runs.filter((run) => run.purpose === "shape_quality" && run.status === "shape_fail").length,
-			expectedTierTotal: runs.filter((run) => run.expectedTier != null).length,
-			expectedTierMatches: runs.filter((run) => run.expectedTier != null && run.expectedTier === run.tier).length,
-			robustnessNoOutput: runs.filter((run) => run.status === "observer_no_output").length
-		};
-		const uniqueObserverKeys = Array.from(new Set(runs.map((run) => `${run.provider}::${run.model}::${run.openaiUseResponses ? "responses" : "chat"}`)));
-		const observerSummary = opts.observerTierRouting === true ? {
-			provider: uniqueObserverKeys.length === 1 ? runs[0]?.provider ?? observer.provider : "mixed",
-			model: uniqueObserverKeys.length === 1 ? runs[0]?.model ?? observer.model : "mixed",
-			tierRouting: true,
-			openaiUseResponses: uniqueObserverKeys.length === 1 ? runs[0]?.openaiUseResponses ?? observer.openaiUseResponses : null,
-			reasoningEffort: uniqueObserverKeys.length === 1 ? runs[0]?.reasoningEffort ?? observer.reasoningEffort : "mixed",
-			reasoningSummary: uniqueObserverKeys.length === 1 ? runs[0]?.reasoningSummary ?? observer.reasoningSummary : "mixed",
-			maxOutputTokens: uniqueObserverKeys.length === 1 ? runs[0]?.maxOutputTokens ?? observer.maxOutputTokens : null,
-			temperature: uniqueObserverKeys.length === 1 ? runs[0]?.temperature ?? observer.temperature : null,
-			transcriptBudget: transcriptBudget ?? null,
-			selectedObservers: uniqueObserverKeys
-		} : {
-			provider: observer.provider,
-			model: observer.model,
-			tierRouting: false,
-			openaiUseResponses: observer.openaiUseResponses,
-			reasoningEffort: observer.reasoningEffort,
-			reasoningSummary: observer.reasoningSummary,
-			maxOutputTokens: observer.maxOutputTokens,
-			temperature: observer.temperature,
-			transcriptBudget: transcriptBudget ?? null,
-			selectedObservers: uniqueObserverKeys
-		};
-		const output = {
-			benchmark: {
-				id: benchmark.id,
-				title: benchmark.title,
-				scenarioId: benchmark.scenarioId
-			},
-			observer: observerSummary,
-			summary,
-			runs
-		};
-		if (opts.json) {
-			console.log(JSON.stringify(output, null, 2));
+				observer: observerSummary,
+				summary,
+				runs
+			};
+			if (opts.json) {
+				console.log(JSON.stringify(output, null, 2));
+				return;
+			}
+			p.intro("codemem memory extraction-benchmark");
+			p.log.info([
+				`Benchmark: ${benchmark.id} — ${benchmark.title}`,
+				`Observer: ${observerSummary.provider}/${observerSummary.model}`,
+				`Tier routing: ${opts.observerTierRouting === true ? "yes" : "no"}`,
+				`OpenAI Responses: ${observerSummary.openaiUseResponses === null ? "mixed" : observerSummary.openaiUseResponses ? "yes" : "no"}`,
+				`Reasoning effort: ${observerSummary.reasoningEffort ?? "none"}`,
+				`Reasoning summary: ${observerSummary.reasoningSummary ?? "none"}`,
+				`Max output tokens: ${observerSummary.maxOutputTokens ?? "mixed"}`,
+				`Temperature: ${observerSummary.temperature ?? "mixed"}`,
+				`Transcript budget override: ${transcriptBudget ?? "default"}`,
+				`Shape-quality passes: ${summary.shapeQualityPasses}/${summary.shapeQualityTotal}`,
+				`Shape-quality fails: ${summary.shapeQualityFails}`,
+				`Expected-tier matches: ${summary.expectedTierMatches}/${summary.expectedTierTotal}`,
+				`Observer no-output cases: ${summary.robustnessNoOutput}`
+			].join("\n"));
+			for (const run of runs) p.log.message(`  [${run.batchId}] ${run.status.padEnd(18)} ${run.complexity.padEnd(10)} tier=${run.tier.padEnd(6)} expected=${(run.expectedTier ?? "n/a").padEnd(6)} span=${String(run.analysis.eventSpan).padEnd(3)} prompts=${run.analysis.promptCount} tools=${String(run.analysis.toolCount).padEnd(2)} transcript=${run.analysis.transcriptLength} ${run.provider}/${run.model}${run.openaiUseResponses ? " [responses]" : ""} summaries=${run.summaries} observations=${run.observations} repair=${run.repairApplied ? "yes" : "no"} — ${run.label}`);
+			p.outro("done");
+		} catch (error) {
+			const message = error instanceof Error ? error.message : "Extraction benchmark failed";
+			if (opts.json) emitJsonError("extraction_benchmark_failed", message);
+			else {
+				p.log.error(message);
+				process.exitCode = 1;
+			}
 			return;
 		}
-		p.intro("codemem memory extraction-benchmark");
-		p.log.info([
-			`Benchmark: ${benchmark.id} — ${benchmark.title}`,
-			`Observer: ${observerSummary.provider}/${observerSummary.model}`,
-			`Tier routing: ${opts.observerTierRouting === true ? "yes" : "no"}`,
-			`OpenAI Responses: ${observerSummary.openaiUseResponses === null ? "mixed" : observerSummary.openaiUseResponses ? "yes" : "no"}`,
-			`Reasoning effort: ${observerSummary.reasoningEffort ?? "none"}`,
-			`Reasoning summary: ${observerSummary.reasoningSummary ?? "none"}`,
-			`Max output tokens: ${observerSummary.maxOutputTokens ?? "mixed"}`,
-			`Temperature: ${observerSummary.temperature ?? "mixed"}`,
-			`Transcript budget override: ${transcriptBudget ?? "default"}`,
-			`Shape-quality passes: ${summary.shapeQualityPasses}/${summary.shapeQualityTotal}`,
-			`Shape-quality fails: ${summary.shapeQualityFails}`,
-			`Expected-tier matches: ${summary.expectedTierMatches}/${summary.expectedTierTotal}`,
-			`Observer no-output cases: ${summary.robustnessNoOutput}`
-		].join("\n"));
-		for (const run of runs) p.log.message(`  [${run.batchId}] ${run.status.padEnd(18)} ${run.complexity.padEnd(10)} tier=${run.tier.padEnd(6)} expected=${(run.expectedTier ?? "n/a").padEnd(6)} span=${String(run.analysis.eventSpan).padEnd(3)} prompts=${run.analysis.promptCount} tools=${String(run.analysis.toolCount).padEnd(2)} transcript=${run.analysis.transcriptLength} ${run.provider}/${run.model}${run.openaiUseResponses ? " [responses]" : ""} summaries=${run.summaries} observations=${run.observations} repair=${run.repairApplied ? "yes" : "no"} — ${run.label}`);
-		p.outro("done");
 	});
 	return cmd;
 }
@@ -5023,6 +5173,14 @@ cmd.action((opts) => {
 		const items = store.recent(limit, filters);
 		if (opts.json) console.log(JSON.stringify(items));
 		else for (const item of items) console.log(`#${item.id} [${item.kind}] ${item.title}`);
+	} catch (error) {
+		const message = error instanceof Error ? error.message : "Recent lookup failed";
+		if (opts.json) emitJsonError("recent_failed", message);
+		else {
+			console.error(message);
+			process.exitCode = 1;
+		}
+		return;
 	} finally {
 		store.close();
 	}
@@ -5064,6 +5222,14 @@ var searchCommand = searchCmd.action((query, opts) => {
 			].join("\n"));
 		}
 		p.outro("done");
+	} catch (error) {
+		const message = error instanceof Error ? error.message : "Search failed";
+		if (opts.json) emitJsonError("search_failed", message);
+		else {
+			p.log.error(message);
+			process.exitCode = 1;
+		}
+		return;
 	} finally {
 		store.close();
 	}
@@ -6260,6 +6426,14 @@ var statsCommand = statsCmd.action((opts) => {
 			p.log.step(`Usage\n${lines.join("\n")}`);
 		}
 		p.outro("done");
+	} catch (error) {
+		const message = error instanceof Error ? error.message : "Stats failed";
+		if (opts.json) emitJsonError("stats_failed", message);
+		else {
+			p.log.error(message);
+			process.exitCode = 1;
+		}
+		return;
 	} finally {
 		store.close();
 	}
@@ -6431,6 +6605,14 @@ attemptsCmd.action((opts) => {
 			return;
 		}
 		for (const row of rows) console.log(formatSyncAttempt(row));
+	} catch (error) {
+		const message = error instanceof Error ? error.message : "Failed to read sync attempts";
+		if (opts.json) emitJsonError("invalid_input", message);
+		else {
+			p.log.error(message);
+			process.exitCode = 1;
+		}
+		return;
 	} finally {
 		store.close();
 	}
@@ -6926,6 +7108,14 @@ enableCmd.action((opts) => {
 			`Interval:    ${config.sync_interval_s ?? 120}s`
 		].join("\n"));
 		p.outro("Sync enabled — restart `codemem serve` to activate");
+	} catch (error) {
+		const message = error instanceof Error ? error.message : "Failed to enable sync";
+		if (opts.json) emitJsonError("invalid_input", message);
+		else {
+			p.log.error(message);
+			process.exitCode = 1;
+		}
+		return;
 	} finally {
 		store.close();
 	}
@@ -7265,22 +7455,25 @@ if (hasRootFlag("--cleanup-completion")) {
 	process.exit(0);
 }
 {
+	const forwardedTail = (subcommand) => {
+		const argv = process.argv.slice(2);
+		const groupIdx = argv.indexOf("memory");
+		const searchFrom = groupIdx >= 0 ? groupIdx + 1 : 0;
+		const subIdx = argv.indexOf(subcommand, searchFrom);
+		return subIdx >= 0 ? argv.slice(subIdx + 1) : [];
+	};
 	const memExport = new Command("export").description("Export memories to a JSON file for sharing or backup").argument("<output>", "output file path (use '-' for stdout)").option("--db <path>", "database path").option("--db-path <path>", "database path").option("--project <project>", "filter by project").option("--all-projects", "export all projects").option("--include-inactive", "include deactivated memories").option("--since <iso>", "only export memories created after this ISO timestamp").configureHelp(helpStyle).allowUnknownOption(true).allowExcessArguments(true).action(async () => {
-		const idx = process.argv.indexOf("export");
-		const tail = idx >= 0 ? process.argv.slice(idx + 1) : [];
 		await exportMemoriesCommand.parseAsync([
 			"node",
 			"export-memories",
-			...tail
+			...forwardedTail("export")
 		]);
 	});
 	const memImport = new Command("import").description("Import memories from an exported JSON file").argument("<inputFile>", "input JSON file (use '-' for stdin)").option("--db <path>", "database path").option("--db-path <path>", "database path").option("--remap-project <path>", "remap all projects to this path on import").option("--dry-run", "preview import without writing").configureHelp(helpStyle).allowUnknownOption(true).allowExcessArguments(true).action(async () => {
-		const idx = process.argv.indexOf("import");
-		const tail = idx >= 0 ? process.argv.slice(idx + 1) : [];
 		await importMemoriesCommand.parseAsync([
 			"node",
 			"import-memories",
-			...tail
+			...forwardedTail("import")
 		]);
 	});
 	memoryCommand.addCommand(memExport);