effect-agent 0.1.0-beta.110 → 0.1.0-beta.112

package/dist/{DurableAgentRuntime-PyIgNYXF.mjs → DurableAgentRuntime-BIrgxwRl.mjs}

@@ -28,7 +28,7 @@ import { ExplainedEvidence, ExplainedSubmission, ExplainedUnknownCall, Integrity
 import { digestJson } from "./durable/Digest.mjs";
 import { a as definitionDigestsEqual, c as resolveWorkerBinding, i as compileRegistrations, l as toolReplayContracts, o as makeLegacyWorkerBinding, s as resolveDefinitionBinding, t as BindingUnavailable } from "./agent-registration-B5CI-Ygk.mjs";
 import { $ as workerOriginRecordId, A as subagentRequestedRecordId, B as toolCallUnknownRecordId, C as runStartedBatchId, D as subagentLineageBatchId, E as subagentJoinedRecordId, F as toolCallPreparedRecordId, G as turnCanonicalBatch, H as toolStepSettledRecordId, I as toolCallResolutionBatchId, J as turnResponseBatch, K as turnIdForRun, L as toolCallResolvedRecordId, M as subagentStartedRecordId, N as toolApprovalDecisionRecordId, O as subagentLineageRecordId, P as toolApprovalRequestRecordId, Q as workerInputRecordId, R as toolCallResultBatchId, S as runIdForSubmission, T as subagentJoinBatchId, U as turnApprovalsBatchId, V as toolStepSettledBatchId, X as turnResultsBatch, a as childIdempotencyKeyFor, b as runDurationBatchId, c as compactionRecordId, d as modelResponseInterruptedBatchId, et as workerReportRecordId, f as modelResponseInterruptedRecordId, g as projectRunJournalStream, i as approvalDecisionBatchId, j as subagentStartedBatchId, k as subagentRequestedBatchId, l as firstWorkerInputRecordId, m as peerMessageRecordId, o as childThreadIdFor, p as modelResponseRecordId, q as turnPreparedBatchId, r as agentUpdateRecordId, s as compactionBatchId, t as RunJournalError, tt as makeJournalMetadata, u as markUnknownBatchId, v as runCompletedRecordId, w as runStartedRecordId, x as runDurationRecordId, y as runCompletionDigest, z as toolCallSettledRecordId } from "./RunJournal-BXrq0VIK.mjs";
-import { FencedAppendRequest, LoadCheckpointRequest, MAX_THREAD_EXPORT_RECORDS, SaveRecoveryCheckpointRequest, ThreadCheckpoint, ThreadExportRequest, ThreadMaterialization, ThreadObservation, ThreadRead, ThreadStore, ThreadStoreError, ThreadTailRequest, getRecord, readWorkerState } from "./durable/ThreadStore.mjs";
+import { FencedAppendRequest, LoadCheckpointRequest, MAX_THREAD_EXPORT_RECORDS, SaveRecoveryCheckpointRequest, ThreadCheckpoint, ThreadExportRequest, ThreadMaterialization, ThreadObservation, ThreadRead, ThreadStore, ThreadStoreDiagnostic, ThreadStoreError, ThreadTailRequest, getRecord, readWorkerState } from "./durable/ThreadStore.mjs";
 import { DurableRuntimeFailpoint } from "./durable/DurableFailpoint.mjs";
 import { ScheduledInputRefused, ScheduledInputRetryable } from "./durable/Schedule.mjs";
 import { PreparedInputAdmission } from "./durable/PreparedInputAdmission.mjs";
@@ -42,7 +42,7 @@ import { verifyThreadInvariants } from "./durable/ThreadInvariants.mjs";
 import { PreparedToolCallEvidence, ToolReconciler } from "./durable/ToolReconciler.mjs";
 import { WakeScheduler } from "./durable/WakeScheduler.mjs";
 import { Prompt } from "effect/unstable/ai";
-import { Cause, Clock, Context, Crypto, DateTime, Deferred, Duration, Effect, Equal, Layer, Option, Ref, Schedule, Schema, Semaphore, Stream } from "effect";
+import { Cause, Clock, Context, Crypto, DateTime, Deferred, Duration, Effect, Equal, Exit, Layer, Option, Ref, Result, Schedule, Schema, Semaphore, Stream } from "effect";
 //#region src/durable/internal/journal-checkpoint.ts
 /** Retired accounting is additive; the latest Tool batch always remains replayable verbatim. */
 var JournalCheckpointSeed = class extends Schema.Class("@effect-agent/thread/internal/JournalCheckpointSeed")({
@@ -1746,8 +1746,12 @@ var DurableAgentRuntime_exports = /* @__PURE__ */ __exportAll({
 	DurableApprovalResolver: () => DurableApprovalResolver,
 	DurableRuntimeConfig: () => DurableRuntimeConfig,
 	Receipt: () => Receipt,
+	RecoveryBlocked: () => RecoveryBlocked,
+	RecoveryFailure: () => RecoveryFailure,
 	RecoveryReport: () => RecoveryReport,
+	RecoverySweepResult: () => RecoverySweepResult,
 	childReservationIdFor: () => childReservationIdFor,
+	isolateRecovery: () => isolateRecovery,
 	recoveryRepairBatchId: () => recoveryRepairBatchId,
 	recoveryRepairRecordId: () => recoveryRepairRecordId,
 	threadCreatedBatchId: () => threadCreatedBatchId,
@@ -1867,6 +1871,114 @@ const threadCreatedRecordId = (threadId) => decodeRecordId(`thread-created:${thr
 const recoveryRepairBatchId = (submissionId, decisionTag) => decodeBatchId(`repair:${submissionId}:${decisionTag}`);
 /** Deterministic record identity of one executed recovery decision's `RepairAnnotated` record. */
 const recoveryRepairRecordId = (submissionId, decisionTag) => decodeRecordId(`repair:${submissionId}:${decisionTag}`);
+const RecoveryCauseTag = Schema.Literals([
+	"AdmissionPolicyError",
+	"DigestError",
+	"LedgerError",
+	"OwnershipLost",
+	"SettlementConflict",
+	"ThreadStoreError",
+	"ThreadNotMaterialized",
+	"AppendConflict",
+	"FenceRejected",
+	"RunJournalError",
+	"DurableRuntimeFailpointError",
+	"SchemaError",
+	"Error",
+	"TypeError",
+	"RangeError",
+	"ReferenceError",
+	"SyntaxError",
+	"URIError",
+	"EvalError",
+	"AggregateError",
+	"ForeignError",
+	"NonErrorCause"
+]);
+const isRecoveryCauseTag = Schema.is(RecoveryCauseTag);
+/** Recovery could not establish execution authority. This is never a Settlement or replay grant. */
+var RecoveryFailure = class extends Schema.Class("@effect-agent/thread/RecoveryFailure")({
+	phase: Schema.Literals(["history", "recovery"]),
+	reason: Schema.Literals([
+		"failure",
+		"defect",
+		"timeout"
+	]),
+	errorTag: Schema.String.check(Schema.isMaxLength(128)),
+	/** Static operation names only; error messages and payloads are deliberately excluded. */
+	operation: Schema.String.check(Schema.isMaxLength(256)),
+	/** Ordered, bounded classifications. Foreign labels, operations and prose stay private. */
+	causes: Schema.Array(Schema.Struct({ errorTag: RecoveryCauseTag })).check(Schema.isMaxLength(16)),
+	causesTruncated: Schema.optionalKey(Schema.Literal(true)),
+	diagnostic: Schema.optionalKey(ThreadStoreDiagnostic)
+}) {};
+const decodeRecoveryCause = Schema.decodeUnknownOption(Schema.Struct({
+	diagnostic: Schema.optionalKey(Schema.Unknown),
+	cause: Schema.optionalKey(Schema.Unknown)
+}));
+/** Reuse the bounded foreign-error capture, then retain only content-free causal metadata. */
+const recoveryFailureDetails = (cause) => {
+	const pending = [capture(cause)];
+	const causes = [];
+	let truncated = false;
+	while (pending.length > 0 && causes.length < 16) {
+		const node = pending.shift();
+		if (node === void 0) break;
+		switch (node._tag) {
+			case "Cause":
+				pending.unshift(...node.reasons.flatMap((reason) => reason._tag === "Fail" ? [reason.error] : reason._tag === "Die" ? [reason.defect] : []));
+				truncated ||= node.truncated === true;
+				break;
+			case "Error": {
+				const tag = node.errorTag ?? node.name;
+				causes.push({ errorTag: isRecoveryCauseTag(tag) ? tag : "ForeignError" });
+				pending.unshift(...node.cause === void 0 ? [] : [node.cause], ...node.reason === void 0 ? [] : [node.reason], ...node.errors ?? []);
+				truncated ||= node.truncated === true;
+				break;
+			}
+			case "Omitted":
+				truncated = true;
+				break;
+			case "Value": causes.push({ errorTag: "NonErrorCause" });
+		}
+	}
+	let nested = Option.getOrUndefined(Cause.findErrorOption(cause));
+	let diagnostic;
+	for (let depth = 0; depth < 16 && nested !== void 0; depth++) {
+		const decoded = decodeRecoveryCause(nested);
+		if (Option.isNone(decoded)) break;
+		if (decoded.value.diagnostic instanceof ThreadStoreDiagnostic) {
+			diagnostic = decoded.value.diagnostic;
+			break;
+		}
+		nested = decoded.value.cause;
+	}
+	return {
+		causes,
+		...truncated || pending.length > 0 ? { causesTruncated: true } : {},
+		...diagnostic === void 0 ? {} : { diagnostic }
+	};
+};
+/** @internal One bounded failure boundary for execution history and host recovery control reads. */
+const isolateRecovery = Effect.fnUntraced(function* (body, options) {
+	const outcome = yield* body.pipe(Effect.scoped, Effect.timeoutOption(options.timeout), Effect.exit);
+	if (Exit.isSuccess(outcome) && Option.isSome(outcome.value)) return Result.succeed(outcome.value.value);
+	if (Exit.isFailure(outcome) && (Cause.hasInterrupts(outcome.cause) || outcome.cause.reasons.some((reason) => reason._tag === "Fail" && reason.error._tag === "DurableRuntimeFailpointError"))) return yield* Effect.failCause(outcome.cause);
+	const error = Exit.isFailure(outcome) ? Cause.findErrorOption(outcome.cause) : Option.none();
+	const phase = options.phase();
+	return Result.fail(RecoveryFailure.make({
+		phase,
+		reason: Exit.isSuccess(outcome) ? "timeout" : Cause.hasDies(outcome.cause) ? "defect" : "failure",
+		errorTag: Option.isSome(error) ? error.value._tag : Exit.isSuccess(outcome) ? "RecoveryTimeout" : "Defect",
+		operation: Option.isSome(error) && "operation" in error.value ? error.value.operation.slice(0, 256) : options.operation ?? (phase === "history" ? "read recovery history" : "recover submission"),
+		...Exit.isFailure(outcome) ? recoveryFailureDetails(outcome.cause) : { causes: [] }
+	}));
+});
+/** One failed Thread remains pending; the host must not claim it before recovery succeeds. */
+var RecoveryBlocked = class extends Schema.TaggedError()("RecoveryBlocked", {
+	threadId: ThreadId,
+	failure: RecoveryFailure
+}) {};
 /** One executed (or deliberately deferred) recovery decision (durability §14, DUR-013). */
 var RecoveryReport = class extends Schema.Class("@effect-agent/thread/RecoveryReport")({
 	submissionId: SubmissionId,
@@ -1885,6 +1997,12 @@ var RecoveryReport = class extends Schema.Class("@effect-agent/thread/RecoveryRe
 		"unknown"
 	])
 }) {};
+/** Submission decisions and operational faults have different ownership and settlement semantics. */
+var RecoverySweepResult = class extends Schema.Class("RecoverySweepResult")({
+	reports: Schema.Array(RecoveryReport),
+	/** Exactly one fault per failed Thread, regardless of its queued Submission count. */
+	blocked: Schema.Array(RecoveryBlocked)
+}) {};
 /**
 * Optional policy-auto approval delegate consulted by the durable approval hook (plan §2.6 step
 * 2) AFTER the recorded-decision lookup misses. An immediate `approved`/`denied` answer becomes
@@ -1904,6 +2022,7 @@ var DurableRuntimeConfig = class DurableRuntimeConfig extends Context.Service()(
 			settlementPollInterval: options.settlementPollInterval ?? Duration.millis(500),
 			leaseRenewalInterval: options.leaseRenewalInterval ?? Duration.seconds(10),
 			abortPollInterval: options.abortPollInterval ?? Duration.millis(500),
+			recoveryTimeout: options.recoveryTimeout ?? Duration.seconds(30),
 			...options.estimateCostMicrousd === void 0 ? {} : { estimateCostMicrousd: options.estimateCostMicrousd }
 		};
 	}
@@ -6103,31 +6222,44 @@ const make = Effect.fn("DurableAgentRuntime.make")(function* (bindings) {
 		if (history.materialized && found.value.workerAdmission?.origin.reporting?.mode === "standard") yield* updateRuntime.repair(found.value.threadId);
 		return yield* recoverSnapshot(found.value, history);
 	});
-	const runRecoveryImpl = Effect.fn("DurableAgentRuntime.runRecovery")(function* () {
+	const runRecovery = Effect.fn("DurableAgentRuntime.runRecovery")(function* (options) {
 		const nonterminal = yield* Stream.runCollect(ledger.scanNonterminal);
 		const reports = [];
+		const blocked = [];
 		let index = 0;
 		while (index < nonterminal.length) {
 			const first = nonterminal[index];
 			if (first === void 0) break;
 			const submissionIds = [];
-			let hasStandardReporting = false;
 			for (let offset = index; offset < nonterminal.length; offset += 1) {
 				const entry = nonterminal[offset];
 				if (entry === void 0 || entry.threadId !== first.threadId) break;
 				submissionIds.push(entry.submissionId);
-				hasStandardReporting ||= entry.workerAdmission?.origin.reporting?.mode === "standard";
-			}
-			const history = yield* readRecoveryHistory(first.threadId, submissionIds);
-			if (history.materialized && hasStandardReporting) yield* updateRuntime.repair(first.threadId);
-			while (index < nonterminal.length) {
-				const submission = nonterminal[index];
-				if (submission === void 0 || submission.threadId !== first.threadId) break;
-				reports.push(yield* recoverSnapshot(submission, history));
-				index += 1;
 			}
+			index += submissionIds.length;
+			if (options?.threadId !== void 0 && options.threadId !== first.threadId) continue;
+			let phase = "recovery";
+			const outcome = yield* isolateRecovery(Effect.gen(function* () {
+				const group = yield* Effect.forEach(submissionIds, (submissionId) => lookupKnownSubmission("recover submission", submissionId));
+				phase = "history";
+				const history = yield* readRecoveryHistory(first.threadId, submissionIds);
+				phase = "recovery";
+				if (history.materialized && group.some((row) => row.workerAdmission?.origin.reporting?.mode === "standard")) yield* updateRuntime.repair(first.threadId);
+				return yield* Effect.forEach(group, (submission) => recoverSnapshot(submission, history));
+			}), {
+				timeout: config.recoveryTimeout,
+				phase: () => phase
+			});
+			if (Result.isSuccess(outcome)) reports.push(...outcome.success);
+			else blocked.push(RecoveryBlocked.make({
+				threadId: first.threadId,
+				failure: outcome.failure
+			}));
 		}
-		return reports;
+		return RecoverySweepResult.make({
+			reports,
+			blocked
+		});
 	});
 	const submit = Effect.fn("DurableAgentRuntime.submit")(function* (agent, input, options) {
 		const encodedInput = yield* Schema.encodeEffect(agent.definition.input)(input).pipe(Effect.mapError((cause) => AgentInputError.make({ message: `Unable to encode Agent input: ${cause.message}` })));
@@ -6436,7 +6568,7 @@ const make = Effect.fn("DurableAgentRuntime.make")(function* (bindings) {
 		const explanations = [];
 		for (const submission of nonterminal) {
 			if (submission.threadId !== threadId) continue;
-			explanations.push(yield* explainSubmission(submission));
+			explanations.push(yield* explainSubmission(yield* lookupKnownSubmission("explain Thread", submission.submissionId)));
 		}
 		return explanations;
 	});
@@ -6448,8 +6580,8 @@ const make = Effect.fn("DurableAgentRuntime.make")(function* (bindings) {
 		const exported = yield* store.export(ThreadExportRequest.make({ threadId }));
 		const rows = /* @__PURE__ */ new Map();
 		const nonterminal = yield* Stream.runCollect(ledger.scanNonterminal);
-		for (const submission of nonterminal) if (submission.threadId === threadId) rows.set(submission.submissionId, submission);
 		const named = /* @__PURE__ */ new Set();
+		for (const submission of nonterminal) if (submission.threadId === threadId) named.add(submission.submissionId);
 		for (const envelope of exported.records) {
 			const payload = envelope.record.payload;
 			if (payload._tag === "UserInputRecorded" || payload._tag === "SubmissionSettled" || payload._tag === "AbortRequested") {
@@ -6564,8 +6696,9 @@ const make = Effect.fn("DurableAgentRuntime.make")(function* (bindings) {
 		const nowMillis = yield* Clock.currentTimeMillis;
 		const generatedAt = yield* nowUtc;
 		const entries = [];
-		for (const submission of nonterminal) {
-			const snapshot = yield* ledger.loadRecoverySnapshot(RecoverySnapshotRequest.make({ submissionId: submission.submissionId }));
+		for (const entry of nonterminal) {
+			const snapshot = yield* ledger.loadRecoverySnapshot(RecoverySnapshotRequest.make({ submissionId: entry.submissionId }));
+			const submission = snapshot.submission;
 			let blockedOn;
 			let since = submission.readyAt ?? submission.createdAt;
 			switch (submission.state) {
@@ -6685,7 +6818,7 @@ const make = Effect.fn("DurableAgentRuntime.make")(function* (bindings) {
 		processThreadHead: processThreadHeadImpl,
 		runWorker: runWorkerImpl,
 		runResolvedWorker: runResolvedWorkerImpl,
-		runRecovery: runRecoveryImpl(),
+		runRecovery,
 		recoverSubmission
 	});
 });
@@ -6767,6 +6900,6 @@ var DurableAgentRuntime = class DurableAgentRuntime extends Context.Service()("@
 	static layer = DurableAgentRuntime.layerWithServices.pipe(Layer.provide([RunContextPreparationPassthrough, RunToolAuthorization.allowAll]));
 };
 //#endregion
-export { RecoveryReport as a, recoveryRepairRecordId as c, DurableRuntimeConfig as i, threadCreatedBatchId as l, DurableAgentRuntime_exports as n, childReservationIdFor as o, DurableApprovalResolver as r, recoveryRepairBatchId as s, DurableAgentRuntime as t, threadCreatedRecordId as u };
+export { RecoveryBlocked as a, RecoverySweepResult as c, recoveryRepairBatchId as d, recoveryRepairRecordId as f, DurableRuntimeConfig as i, childReservationIdFor as l, threadCreatedRecordId as m, DurableAgentRuntime_exports as n, RecoveryFailure as o, threadCreatedBatchId as p, DurableApprovalResolver as r, RecoveryReport as s, DurableAgentRuntime as t, isolateRecovery as u };
 
-//# sourceMappingURL=DurableAgentRuntime-PyIgNYXF.mjs.map
+//# sourceMappingURL=DurableAgentRuntime-BIrgxwRl.mjs.map