@tangle-network/agent-runtime 0.133.8 → 0.134.2

package/dist/{supervisor-BA87pb7g.js → supervisor-4r4BYSti.js}

@@ -1,11 +1,11 @@
-import { f as ValidationError, r as BackendTransportError, u as RuntimeRunStateError } from "./errors-DRoXIc7E.js";
-import { t as detachedSnapshot } from "./snapshot-DY6VGliG.js";
-import { S as contentAddress, a as aggregateProviderModelEvidence, c as pendingWaits, g as runtimeOwnedNestedDriverTreeRoot, h as nestedDriverTreeRoot, l as replaySpawnTree, m as driverRuntime, o as loadSpawnForest, p as attestNestedDriverTreeOwner, s as materializeTreeView, v as captureWorkerTraceEvidence } from "./spawn-journal-BX3-4WDm.js";
-import { c as randomUuid, d as throwAbort, f as throwIfAborted, l as sleep, m as zeroTokenUsage, n as cloneTokenUsage, o as mapWithConcurrency, p as withTimeout, r as deleteBoxSafe, s as randomSuffix, t as addTokenUsage, u as stringifySafe } from "./util-Dr--8wAL.js";
+import { f as ValidationError, r as BackendTransportError, u as RuntimeRunStateError } from "./errors-CVIDr7GQ.js";
+import { t as detachedSnapshot } from "./snapshot-BOIprwBk.js";
+import { S as contentAddress, a as aggregateProviderModelEvidence, c as pendingWaits, g as runtimeOwnedNestedDriverTreeRoot, h as nestedDriverTreeRoot, l as replaySpawnTree, m as driverRuntime, o as loadSpawnForest, p as attestNestedDriverTreeOwner, s as materializeTreeView, v as captureWorkerTraceEvidence } from "./spawn-journal-TJSlDa2J.js";
+import { a as hasCompleteCacheBreakdown, c as randomSuffix, d as stringifySafe, f as throwAbort, h as zeroTokenUsage, i as deleteBoxSafe, l as randomUuid, m as withTimeout, n as chargedTokens, p as throwIfAborted, r as cloneTokenUsage, s as mapWithConcurrency, t as addTokenUsage, u as sleep } from "./util-CG3ghHfb.js";
 import { a as notifySandboxEventObserver, n as extractLlmCallEvent } from "./sandbox-events-vIwmuPdl.js";
-import { a as concreteProfileModel, l as resolveRouterRetryPolicy, o as profileBridgeWireModel, s as profileModelExecutionSettings, t as assertExecutableAgentProfile, u as agentHarness } from "./model-policy-Bv1DU2fw.js";
-import { C as unknownMaterializationReceipt, S as unknownExecutionBindingReceipt, T as executableAgentSpecSnapshot, _ as runtimeOwnedExecutorExecutionBinding, a as authoredProfileDigest, b as runtimeOwnedPendingExecutorMaterialization, c as knownExecutionBindingReceipt, f as recordRuntimeOwnedProviderAttemptStart, h as runtimeOwnedDeferredExecutorRuntime, l as knownMaterializationReceipt, m as recordRuntimeOwnedProviderModel, n as attestRuntimeOwnedExecutor, o as finalizeRuntimeOwnedPendingExecutor, p as recordRuntimeOwnedProviderIdentityConflict, r as attestRuntimeOwnedPendingExecutor, t as attestRuntimeOwnedDeferredExecutor, u as newExecutionAttemptId, v as runtimeOwnedExecutorMaterialization, w as executableAgentProfileSnapshot, x as runtimeOwnedScopeOwnerRuntime, y as runtimeOwnedExecutorProviderEvidence } from "./materialization-D2bdTfkK.js";
-import { i as resolveAgentEnvironmentProvider, n as providerAsExecutor, r as providerAsSandboxClient } from "./environment-provider-DTJA4ZiM.js";
+import { a as concreteProfileModel, l as resolveRouterRetryPolicy, o as profileBridgeWireModel, s as profileModelExecutionSettings, t as assertExecutableAgentProfile, u as agentHarness } from "./model-policy-UTTc9Z-A.js";
+import { C as unknownExecutionBindingReceipt, E as executableAgentSpecSnapshot, S as runtimeOwnedScopeOwnerRuntime, T as executableAgentProfileSnapshot, a as authoredProfileDigest, b as runtimeOwnedExecutorProviderEvidence, c as knownExecutionBindingReceipt, f as recordRuntimeOwnedProviderAttemptStart, g as runtimeOwnedDeferredExecutorRuntime, h as recordRuntimeOwnedProviderModel, l as knownMaterializationReceipt, m as recordRuntimeOwnedProviderIdentityConflict, n as attestRuntimeOwnedExecutor, o as finalizeRuntimeOwnedPendingExecutor, p as recordRuntimeOwnedProviderDispatchNotStarted, r as attestRuntimeOwnedPendingExecutor, t as attestRuntimeOwnedDeferredExecutor, u as newExecutionAttemptId, v as runtimeOwnedExecutorExecutionBinding, w as unknownMaterializationReceipt, x as runtimeOwnedPendingExecutorMaterialization, y as runtimeOwnedExecutorMaterialization } from "./materialization-BP5zspr2.js";
+import { i as resolveAgentEnvironmentProvider, n as providerAsExecutor, r as providerAsSandboxClient } from "./environment-provider-CPHwqLLm.js";
 import { i as notifyRuntimeHookEvent } from "./runtime-hooks-C7iJOWm3.js";
 import { estimateCost, isModelPriced } from "@tangle-network/agent-eval";
 import { AGENT_PROFILE_MATERIALIZATION_AXES, REASONING_EFFORTS, agentProfileSchema, canonicalAgentProfileDigest, canonicalCandidateDigest, profileMaterializationAxes, profileMaterializationAxes as profileMaterializationAxes$1, sha256DigestSchema } from "@tangle-network/agent-interface";
@@ -6668,6 +6668,7 @@ const sandboxExecutor = (spec, ctx) => {
 		}
 	};
 	if (seam.steering) {
+		if (seam.validator) throw new ValidationError("sandboxExecutor: validator is not representable with steering — the steerable session is not a runAgentRounds composition, so the score would be silently dropped");
 		const inbox = createInbox();
 		const session = createSteerableSandboxSession({
 			controller,
@@ -6719,6 +6720,7 @@ const sandboxExecutor = (spec, ctx) => {
 				maxIterations,
 				controller,
 				loopCtx: seam.loopCtx,
+				...seam.validator ? { validator: seam.validator } : {},
 				traceEnv,
 				onArtifact: (a) => {
 					artifact = a;
@@ -6758,6 +6760,7 @@ async function* streamSandboxLeaf(args) {
 		driver: args.driver,
 		agentRun,
 		output: args.output,
+		...args.validator ? { validator: args.validator } : {},
 		task: args.task,
 		maxIterations: args.maxIterations,
 		maxConcurrency: 1,
@@ -7135,6 +7138,7 @@ const bridgeExecutor = (spec, ctx) => {
 					trace.record(step);
 				},
 				onProviderAttemptStart: () => recordRuntimeOwnedProviderAttemptStart(executor),
+				onProviderDispatchNotStarted: () => recordRuntimeOwnedProviderDispatchNotStarted(executor),
 				onProviderModel: (model) => recordRuntimeOwnedProviderModel(executor, model),
 				onProviderIdentityConflict: () => recordRuntimeOwnedProviderIdentityConflict(executor),
 				onArtifact: (a) => {
@@ -7464,6 +7468,7 @@ async function* streamBridgeSession(args) {
 			publishProfileMaterialization();
 		} catch (error) {
 			publishProfileMaterialization();
+			if (isTrustedPreProviderRejection(error)) args.onProviderDispatchNotStarted();
 			if (interruptSig.aborted && !args.signal.aborted && !args.controller.signal.aborted) {
 				if (!await cancelBridgeRunToTerminal(seam, activeRun, "infinity", external)) throw new ValidationError(`bridgeExecutor: interrupted run ${activeRun.id} did not reach terminal state`);
 				interrupted = true;
@@ -7565,9 +7570,11 @@ async function* streamDurableBridgeRun(args) {
 		}
 		if (!res.ok) {
 			const body = (await res.text()).slice(0, 300);
+			const providerDispatch = providerDispatchFromErrorBody(body);
 			throw new BackendTransportError("bridge", `bridgeExecutor: bridge ${res.status}: ${body}`, {
 				status: res.status,
-				body
+				body,
+				...providerDispatch === "not_started" ? { providerDispatch } : {}
 			});
 		}
 		if (!res.body) throw new ValidationError("bridgeExecutor: bridge response had no body to stream");
@@ -7820,6 +7827,27 @@ async function cancelBridgeRunToTerminal(seam, run, grace, stopSignal) {
 function errorMessage(error) {
 	return error instanceof Error ? error.message : String(error);
 }
+/** Only Router's exact one-sided fact proves that provider dispatch did not start. */
+function isTrustedPreProviderRejection(error) {
+	return error instanceof BackendTransportError && error.providerDispatch === "not_started";
+}
+/** Read only the Router-owned one-sided field from an error body. */
+function providerDispatchFromErrorBody(body) {
+	let parsed;
+	try {
+		parsed = JSON.parse(body);
+	} catch {
+		return;
+	}
+	if (typeof parsed !== "object" || parsed === null) return void 0;
+	const error = parsed.error;
+	if (typeof error !== "object" || error === null) return void 0;
+	return error.provider_dispatch === "not_started" ? "not_started" : void 0;
+}
+function bridgeUpstreamError(error, prefix) {
+	const providerDispatch = error.provider_dispatch === "not_started" ? "not_started" : void 0;
+	return new BackendTransportError("bridge", `${prefix}: ${error.message ?? error.type ?? "unknown"}`, providerDispatch === void 0 ? void 0 : { providerDispatch });
+}
 function mergeBridgeObservedModel(current, next) {
 	if (current === void 0 || current === next) return next;
 	const merged = mergeObservedModelIdentity(current, next);
@@ -8116,7 +8144,7 @@ function parseSseStreamTail(buf) {
 	} catch {
 		return;
 	}
-	if (parsed.error) throw new ValidationError(`bridgeExecutor: bridge upstream error: ${parsed.error.message ?? parsed.error.type ?? "unknown"}`);
+	if (parsed.error) throw bridgeUpstreamError(parsed.error, "bridgeExecutor: bridge upstream error");
 }
 /** Parse one SSE frame into a numbered run event, terminal marker, or unnumbered keepalive. */
 function parseSseFrame(frame) {
@@ -8151,7 +8179,7 @@ function parseSseFrame(frame) {
 	if (parsed.error) return {
 		kind: "event",
 		id,
-		error: new BackendTransportError("bridge", `bridgeExecutor: bridge stream error: ${parsed.error.message ?? parsed.error.type ?? "unknown"}`)
+		error: bridgeUpstreamError(parsed.error, "bridgeExecutor: bridge stream error")
 	};
 	const out = {};
 	if (parsed.model !== void 0) {
@@ -8530,14 +8558,19 @@ function snapshotExecutorConfig(config) {
 				"loopCtx",
 				"maxIterations",
 				"sandboxClient",
-				"steering"
+				"steering",
+				"validator"
 			]), "createExecutor sandbox config");
-			const { sandboxClient, loopCtx, ...decisionData } = config;
+			const { sandboxClient, loopCtx, validator, ...decisionData } = config;
+			const port = {
+				sandboxClient,
+				...validator === void 0 ? {} : { validator }
+			};
 			if (loopCtx === void 0) {
 				const snapshot = detachedSnapshot(decisionData, "createExecutor sandbox config");
 				return Object.freeze({
 					...snapshot,
-					sandboxClient
+					...port
 				});
 			}
 			const { hooks, traceEmitter, onSandboxEvent, runHandle, ...loopDecisionData } = loopCtx;
@@ -8548,7 +8581,7 @@ function snapshotExecutorConfig(config) {
 			const loopSnapshot = snapshot.loopCtx;
 			return Object.freeze({
 				...snapshot,
-				sandboxClient,
+				...port,
 				loopCtx: Object.freeze({
 					...loopSnapshot,
 					...hooks === void 0 ? {} : { hooks },
@@ -8944,6 +8977,19 @@ function mergeAbortSignals(...signals) {
 * spend, and refunds the unspent remainder to `free`. Tokens and usd are separate
 * channels (`LoopTokenUsage` has no `usd`); iterations are conserved alongside them.
 *
+* The token unit is `chargedTokens`: every token counted ONCE, when it first enters the context
+* (`input − cacheRead + output`, which is `freshInput + cacheWrite + output` under a complete
+* split). A cached prefix is re-presented content that was already charged when it was written, so
+* charging a cache read again would charge the same tokens twice — on a real fleet cache is 98% of
+* the rolled-up prompt total, which made a declared cap bite about 56x early. No price weight
+* enters the token channel; money is budgeted separately on `maxUsd`. The unit is additive, so a
+* child's settlement charges what its turns charged and a rolled-up report agrees with the pool.
+* Prompt tokens the provider never classified are charged in full, and
+* `readout().cacheBreakdownKnown` then reads false, so the balance reads as an upper bound on
+* newly-presented work rather than a measurement. The pool trusts a reported cache read the same way
+* it trusts a reported `input`: the token channel is an accounting unit, not a trust boundary
+* against a provider that misreports its own usage.
+*
 * Pure and deterministic: `now()` is injected, there is no I/O, and no wall-clock or
 * RNG read. A `reserve`/`reconcile` ticket is single-use (fail-loud on double or
 * unknown reconcile) so a child can never refund twice. Reconciling an OPEN ticket always
@@ -8993,7 +9039,12 @@ function assertValidSpend(spend, label) {
 	}
 	for (const [field, value] of [["tokensKnown", spend.tokens.tokensKnown], ["cacheBreakdownKnown", spend.tokens.cacheBreakdownKnown]]) if (value !== void 0 && value !== false) throw new Error(`${label}.tokens.${field} must be false when present`);
 	const { freshInput, cacheRead, cacheWrite } = spend.tokens;
-	if (freshInput !== void 0 && cacheRead !== void 0 && cacheWrite !== void 0 && freshInput + cacheRead + cacheWrite !== spend.tokens.input) throw new Error(`${label}.tokens cache classes must sum to input`);
+	if (freshInput !== void 0 && cacheRead !== void 0 && cacheWrite !== void 0) {
+		const classified = freshInput + cacheRead + cacheWrite;
+		if (spend.tokens.cacheBreakdownKnown !== false) {
+			if (classified !== spend.tokens.input) throw new Error(`${label}.tokens cache classes must sum to input`);
+		} else if (classified > spend.tokens.input) throw new Error(`${label}.tokens cache classes must not exceed input`);
+	}
 	for (const [field, value] of [["usd", spend.usd], ["ms", spend.ms]]) if (!Number.isFinite(value) || value < 0) throw new Error(`${label}.${field} must be a non-negative finite number`);
 }
 /** Fold a normalized `UsageEvent` array into a `Spend`. Tokens and usd are separate
@@ -9044,9 +9095,6 @@ async function foldUsage(events) {
 		ms: 0
 	};
 }
-function totalTokens(usage) {
-	return usage.input + usage.output;
-}
 /**
 * Create a conserved reservation pool from a root `Budget`. `now()` is injected so the
 * deadline readout is deterministic; defaults to `Date.now` for non-test callers. The
@@ -9061,6 +9109,7 @@ function createBudgetPool(root, now = Date.now, restore = {}) {
 	let reservedTokens = 0;
 	let committedTokens = 0;
 	let tokensTainted = false;
+	let cacheBreakdownTainted = false;
 	const usdCapped = root.maxUsd !== void 0;
 	let freeUsd = root.maxUsd ?? 0;
 	let reservedUsd = 0;
@@ -9126,7 +9175,7 @@ function createBudgetPool(root, now = Date.now, restore = {}) {
 		assertValidSpend(spent, `budget pool ticket ${ticket.id} spend`);
 		const { tokens: rTokens, usd: rUsd, iterations: rIterations } = ticket.reserved;
 		const unknownUnderCap = usdCapped && spent.usdKnown === false;
-		const spentTokens = totalTokens(spent.tokens);
+		const spentTokens = chargedTokens(spent.tokens);
 		const usdBudgeted = ticket.reserved.usdBudgeted !== false;
 		let violation;
 		if (spentTokens > rTokens) violation = `ticket ${ticket.id} spent ${spentTokens} tokens > reserved ${rTokens}`;
@@ -9135,6 +9184,7 @@ function createBudgetPool(root, now = Date.now, restore = {}) {
 		else if (unknownUnderCap) violation = `ticket ${ticket.id} reported unknown dollar cost under a dollar-capped budget`;
 		open.delete(ticket.id);
 		if (spent.tokensKnown === false) tokensTainted = true;
+		if (!hasCompleteCacheBreakdown(spent.tokens)) cacheBreakdownTainted = true;
 		if (spent.usdKnown === false) usdMeasured = false;
 		reservedTokens -= rTokens;
 		committedTokens += spentTokens;
@@ -9156,8 +9206,9 @@ function createBudgetPool(root, now = Date.now, restore = {}) {
 		assertValidSpend(spend, "observed spend");
 		if (usdCapped && spend.usdKnown === false) throw new ValidationError("budget pool: cannot observe unknown dollar cost under a dollar-capped budget");
 		if (spend.tokensKnown === false) tokensTainted = true;
+		if (!hasCompleteCacheBreakdown(spend.tokens)) cacheBreakdownTainted = true;
 		if (spend.usdKnown === false) usdMeasured = false;
-		const tokens = totalTokens(spend.tokens);
+		const tokens = chargedTokens(spend.tokens);
 		freeTokens -= tokens;
 		committedTokens += tokens;
 		freeIterations -= spend.iterations;
@@ -9169,6 +9220,7 @@ function createBudgetPool(root, now = Date.now, restore = {}) {
 		return {
 			tokensLeft: freeTokens,
 			tokensKnown: !tokensTainted,
+			cacheBreakdownKnown: !cacheBreakdownTainted,
 			usdLeft: usdCapped ? usdTainted ? 0 : freeUsd : 0,
 			usdCapped,
 			usdKnown: usdMeasured && !usdTainted,
@@ -9184,8 +9236,9 @@ function createBudgetPool(root, now = Date.now, restore = {}) {
 		assertValidSpend(restore.committed, "budget restore committed");
 		const committed = restore.committed;
 		if (committed.tokensKnown === false) tokensTainted = true;
+		if (!hasCompleteCacheBreakdown(committed.tokens)) cacheBreakdownTainted = true;
 		if (committed.usdKnown === false) usdMeasured = false;
-		const tokens = totalTokens(committed.tokens);
+		const tokens = chargedTokens(committed.tokens);
 		freeTokens -= tokens;
 		committedTokens += tokens;
 		freeIterations -= committed.iterations;
@@ -12396,4 +12449,4 @@ function isNonEmptySpend(s) {
 //#endregion
 export { createSandboxForSpec as $, sandboxActProfileMaterialization as $t, spendFromUsageEvents as A, sanitizeRuntimeStreamEvent as At, readWorkerTraceContext as B, parseCodexTokenUsage as Bt, freeSlots as C, padSpanId as Ct, teardownExecutor as D, createRuntimeStreamEventCollector as Dt, DEFAULT_SUCCESSFUL_SHUTDOWN_MS as E, createRuntimeEventCollector as Et, createExecutor as F, removeWorktree as Ft, createSteerableSandboxSession as G, defineProfileMaterializationContract as Gt, workerTraceHeaders as H, AGENT_PROFILE_MATERIALIZATION_AXES as Ht, createExecutorRegistry as I, DEFAULT_LOCAL_HARNESS as It, sandboxSessionTraceSource as J, promptControlProfileMaterialization as Jt, createPushTraceSource as K, fullProfileMaterialization as Kt, snapshotExecutorConfig as L, LOCAL_HARNESSES as Lt, bridgeStopSignalKey as M, runWorktreeHarness as Mt, captureReusableExecutorConfig as N, captureWorktreeDiff as Nt, assertValidBudget as O, sanitizeAgentRuntimeEvent as Ot, cliWorktreeExecutor as P, createWorktree as Pt, createInbox as Q, renderProfileMaterializationIssues as Qt, createWorktreeCliExecutor as R, harnessSupportsReasoningEffort as Rt, effectiveConcurrency as S, loopEventToOtelSpan as St, rollingDispatch as T, toOtelAttributes as Tt, workerTraceSeamKey as U, assertProfileMaterialization as Ut, workerTraceEnv as V, CodexExecutionDiagnosticError as Vt, DEFAULT_SANDBOX_STEERING_MAX_TURNS as W, controlProfileMaterialization as Wt, createActivityLog as X, promptOnlyProfileMaterialization as Xt, DEFAULT_STALL_AFTER_MS as Y, promptModelProfileMaterialization as Yt, readWorkerProgress as Z, promptResourceProfileMaterialization as Zt, isWaitOutcome as _, createOpenInferenceFileExporter as _t, pickBestDelivered as a, routerBrain as at, validateWaitSpec as b, flatOtelSpan as bt, driverChild as c, observedModelMatchesDeclared as ct, deriveNodeExecutionIdentity as d, readTraceContextFromEnv as dt, validateProfileMaterialization as en, defaultSelectWinner as et, meterRuntimeOwnedProviderAttempt as f, traceContextToEnv as ft, createWaitProbes as g, buildRuntimeEventOtelSpans as gt, settledToIteration as h, buildLoopSpanNodes as ht, collectDelivered as i, acquireSandbox as it, bindReusableExecutorExecutionId as j, runSettledCommand as jt, createBudgetPool as k, sanitizeKnowledgeReadinessReport as kt, withDriverExecutor as l, createPropagatingTraceEmitter as lt, scopeOwnerExecutorNodeContext as m, buildLoopOtelSpans as mt, createSupervisor as n, createSandboxLineage as nt, runFinalizer as o, runBrainLoop as ot, recordScopeOwnerMaterialization as p, INTELLIGENCE_WIRE_VERSION as pt, decodeToolPart as q, profileMaterializationAxes$1 as qt, bestDelivered as r, probeSandboxCapabilities as rt, runTree as s, canonicalObservedModel as st, createRootHandle as t, worktreeCliProfileMaterialization as tn, runAgentRounds as tt, createScope as u, mergeTraceEnv as ut, pollFor as v, createOtelExporter as vt, queueOf as w, padTraceId as wt, waitUntil as x, generateSpanId as xt, timerAt as y, exportEvalRuns as yt, WORKER_TRACE_PROPAGATION as z, localHarnessExecutable as zt };
 
-//# sourceMappingURL=supervisor-BA87pb7g.js.map
+//# sourceMappingURL=supervisor-4r4BYSti.js.map