@tangle-network/agent-runtime 0.181.0 → 0.183.0

package/dist/{supervisor-CC3qqrVo.js → supervisor-DyRbmn1C.js}

@@ -3,10 +3,10 @@ import { c as RetainedInteractiveAdmissionError, d as RetainedRunDispatchBinding
 import { n as detachedSnapshot } from "./snapshot-CTAf4uuA.js";
 import { _ as captureWorkerTraceEvidence, a as aggregateProviderModelEvidence, c as loadSpawnForest, d as replaySpawnTree, f as attestNestedDriverTreeOwner, h as runtimeOwnedNestedDriverTreeRoot, l as materializeTreeView, m as nestedDriverTreeRoot, n as FileSpawnJournal, o as closesCursorSlot, p as driverRuntime, u as pendingWaits, x as contentAddress } from "./spawn-journal--N8Ten1q.js";
 import { _ as unmeteredSpend, a as cloneTokenUsage, b as zeroTokenUsage, d as randomSuffix, f as randomUuid, g as throwIfAborted, h as throwAbort, l as mapWithConcurrency, m as stringifySafe, n as addTokenUsage, o as deleteBoxSafe, p as sleep, r as chargedTokens, s as hasCompleteCacheBreakdown, t as addSpend, u as promptCacheTokenClasses, v as withTimeout, y as zeroSpend } from "./util-D6ZEuBMi.js";
-import { c as notifySandboxEventObserver, i as extractLlmCallEvent, l as sandboxEventServedBackend, m as parseCanonicalTransportEvent, p as extractTransportEventIdentity, r as createSandboxToolPartState, s as mapSandboxToolEvent, t as assertSandboxServedModel, u as sandboxProgressEvents } from "./sandbox-events-B2cWfjwu.js";
+import { _ as parseCodexUsageRecord, c as mapSandboxToolEvent, d as sandboxProgressEvents, h as parseCanonicalTransportEvent, i as createSandboxUsageLedger, l as notifySandboxEventObserver, m as extractTransportEventIdentity, r as createSandboxToolPartState, t as assertSandboxServedModel, u as sandboxEventServedBackend } from "./sandbox-events-DGXLxiuX.js";
 import { n as executableAgentSpecSnapshot, r as projectSandboxOutcome, t as executableAgentProfileSnapshot } from "./executable-spec-CkGvdPds.js";
 import { a as concreteProfileModel, c as profileModelExecutionSettings, d as agentHarness, o as enforceTokenLimits, s as profileBridgeWireModel, t as assertExecutableAgentProfile, u as resolveRouterRetryPolicy } from "./model-policy-Sw4ywhtL.js";
-import { c as linkAbort, i as resolveAgentEnvironmentProvider, l as runAbortable, n as providerAsExecutor, r as providerAsSandboxClient, s as abortError$1 } from "./environment-provider-B6HqC4mh.js";
+import { c as linkAbort, i as resolveAgentEnvironmentProvider, l as runAbortable, n as providerAsExecutor, r as providerAsSandboxClient, s as abortError$1 } from "./environment-provider-C4elwcpo.js";
 import { i as notifyRuntimeHookEvent } from "./runtime-hooks-tXpAarhW.js";
 import { n as assertNoSymlinkDescendant, r as publishExclusiveDurableFile } from "./durable-file-g3YdfEh4.js";
 import { AGENT_PROFILE_MATERIALIZATION_AXES, AgentEnvironmentCapabilitiesSchema, AgentExactRunControlRefSchema, AgentInteractiveSessionControlClaimAcknowledgementSchema, AgentInteractiveSessionControlClaimSchema, AgentInteractiveSessionPromptAcknowledgementSchema, AgentInteractiveSessionRefSchema, AgentInteractiveSessionStatusSchema, AgentInteractiveSessionStopAcknowledgementSchema, AgentNativeContextContinuationResultSchema, AgentTurnInputSchema, AgentTurnResultSchema, ContextTransferRequestSchema, InteractionAcknowledgementSchema, InteractionResponseCommandSchema, NativeContextBoundaryProofSchema, NativeContextContinuationRequestSchema, REASONING_EFFORTS, RuntimeEventEnvelopeSchema, TerminalReplayWindowSchema, TerminalSessionRefSchema, agentInteractiveSessionControlClaimAcknowledgementMatchesRequest, agentInteractiveSessionControlClaimMatchesRef, agentInteractiveSessionPromptAcknowledgementMatchesCommand, agentInteractiveSessionRefMatchesStart, agentInteractiveSessionRunRef, agentInteractiveSessionStatusMatchesRef, agentInteractiveSessionStopAcknowledgementMatchesCommand, agentNativeContextContinuationResultMatchesRequest, agentProfileSchema, canonicalAgentProfileDigest, canonicalCandidateDigest, exactAgentInteractiveSessionStart, harnessTypeSchema, nativeContextContinuationTurnDigest, nativeReasoningControl, profileMaterializationAxes, profileMaterializationAxes as profileMaterializationAxes$1, renderInputPartsAsText, sha256DigestSchema } from "@tangle-network/agent-interface";
@@ -1548,7 +1548,13 @@ function processGroupExists(processGroupId) {
 		return true;
 	}
 }
-/** Parse and validate the one terminal usage event emitted by `codex exec --json`. */
+/**
+* Parse and validate the one terminal usage event emitted by `codex exec --json`.
+*
+* The JSONL framing is this surface's own; the usage RECORD is read by `parseCodexUsageRecord`,
+* the one codex usage reader the sandbox decoder also calls, so both surfaces hold the same field
+* policy and the same two cross-field invariants.
+*/
 function parseCodexTokenUsage(stdout) {
 	const completed = [];
 	for (const [index, line] of stdout.split(/\r?\n/).entries()) {
@@ -1557,34 +1563,25 @@ function parseCodexTokenUsage(stdout) {
 		try {
 			event = JSON.parse(line);
 		} catch {
-			throw new Error(`runLocalHarness: Codex JSONL line ${index + 1} is not valid JSON`);
+			throw new ValidationError(`runLocalHarness: Codex JSONL line ${index + 1} is not valid JSON`);
 		}
 		if (isRecord(event) && event.type === "turn.completed") completed.push(event);
 	}
-	if (completed.length !== 1) throw new Error(`runLocalHarness: expected exactly one Codex turn.completed usage event, received ${completed.length}`);
+	if (completed.length !== 1) throw new ValidationError(`runLocalHarness: expected exactly one Codex turn.completed usage event, received ${completed.length}`);
 	const event = completed[0];
-	if (!isRecord(event) || !isRecord(event.usage)) throw new Error("runLocalHarness: Codex turn.completed event is missing usage");
-	const usage = event.usage;
-	const inputTokens = naturalNumber(usage.input_tokens, "input_tokens");
-	const cachedInputTokens = naturalNumber(usage.cached_input_tokens, "cached_input_tokens");
-	const outputTokens = naturalNumber(usage.output_tokens, "output_tokens");
-	const reasoningOutputTokens = naturalNumber(usage.reasoning_output_tokens, "reasoning_output_tokens");
-	if (cachedInputTokens > inputTokens) throw new Error("runLocalHarness: Codex cached_input_tokens exceeds input_tokens");
-	if (reasoningOutputTokens > outputTokens) throw new Error("runLocalHarness: Codex reasoning_output_tokens exceeds output_tokens");
+	if (!isRecord(event) || !isRecord(event.usage)) throw new ValidationError("runLocalHarness: Codex turn.completed event is missing usage");
+	const record = parseCodexUsageRecord(event.usage, "runLocalHarness: Codex");
 	return {
-		inputTokens,
-		cachedInputTokens,
-		outputTokens,
-		reasoningOutputTokens
+		inputTokens: record.inputTokens,
+		cachedInputTokens: record.cachedInputTokens,
+		outputTokens: record.outputTokens,
+		reasoningOutputTokens: record.reasoningOutputTokens,
+		...record.cacheWriteInputTokens === void 0 ? {} : { cacheWriteInputTokens: record.cacheWriteInputTokens }
 	};
 }
 function isRecord(value) {
 	return value !== null && typeof value === "object" && !Array.isArray(value);
 }
-function naturalNumber(value, field) {
-	if (typeof value !== "number" || !Number.isSafeInteger(value) || value < 0) throw new Error(`runLocalHarness: Codex usage.${field} must be a non-negative safe integer`);
-	return value;
-}
 //#endregion
 //#region src/mcp/worktree.ts
 /**
@@ -4960,6 +4957,25 @@ async function executeIteration(args) {
 	});
 	let box;
 	let lineageOwned = false;
+	let sawLlmCall = false;
+	const usageLedger = createSandboxUsageLedger(spec.profile.harness);
+	const creditLlmCall = (llmCall) => {
+		sawLlmCall = true;
+		slot.costUsd += llmCall.costUsd ?? 0;
+		if (llmCall.usdKnown === false) slot.costUsdKnown = false;
+		if (llmCall.estimatedCostUsd !== void 0) slot.estimatedCostUsd = (slot.estimatedCostUsd ?? 0) + llmCall.estimatedCostUsd;
+		if (llmCall.promptCache) {
+			slot.promptCache ??= {};
+			mergeUsageMetadata(slot.promptCache, llmCall.promptCache);
+		}
+		addTokenUsage(slot.tokenUsage, {
+			input: llmCall.tokensIn,
+			output: llmCall.tokensOut,
+			...llmCall.tokensKnown === false ? { tokensKnown: false } : {},
+			...promptCacheTokenClasses(llmCall.tokensIn, llmCall.promptCache)
+		});
+		args.ctx.runHandle?.observe(llmCall);
+	};
 	try {
 		let stream;
 		const source = args.lineagePlan?.[args.item.index - args.baseIndex];
@@ -5001,7 +5017,6 @@ async function executeIteration(args) {
 			...spec.profile.model?.provider !== void 0 ? { provider: spec.profile.model.provider } : {},
 			...requestedModel !== void 0 ? { model: requestedModel } : {}
 		};
-		let sawLlmCall = false;
 		const outcomeTracker = createAgentRunOutcomeTracker();
 		for await (const event of stream) {
 			events.push(event);
@@ -5010,26 +5025,12 @@ async function executeIteration(args) {
 				iterationIndex: args.item.index,
 				agentRunName: slot.agentRunName
 			});
-			const llmCall = extractLlmCallEvent(event, slot.agentRunName);
-			if (llmCall) {
-				sawLlmCall = true;
-				slot.costUsd += llmCall.costUsd ?? 0;
-				if (llmCall.usdKnown === false) slot.costUsdKnown = false;
-				if (llmCall.estimatedCostUsd !== void 0) slot.estimatedCostUsd = (slot.estimatedCostUsd ?? 0) + llmCall.estimatedCostUsd;
-				if (llmCall.promptCache) {
-					slot.promptCache ??= {};
-					mergeUsageMetadata(slot.promptCache, llmCall.promptCache);
-				}
-				addTokenUsage(slot.tokenUsage, {
-					input: llmCall.tokensIn,
-					output: llmCall.tokensOut,
-					...llmCall.tokensKnown === false ? { tokensKnown: false } : {},
-					...promptCacheTokenClasses(llmCall.tokensIn, llmCall.promptCache)
-				});
-				args.ctx.runHandle?.observe(llmCall);
-			}
+			const llmCall = usageLedger.observe(event, slot.agentRunName);
+			if (llmCall) creditLlmCall(llmCall);
 			assertSandboxServedModel(event, requested);
 		}
+		const harnessCall = usageLedger.settleTurn(slot.agentRunName);
+		if (harnessCall) creditLlmCall(harnessCall);
 		const outcome = outcomeTracker.finish();
 		slot.sandboxOutcome = outcome;
 		if (!sawLlmCall) {
@@ -5048,6 +5049,8 @@ async function executeIteration(args) {
 		else if (projection.status !== "completed") slot.verdict = projection.verdict;
 	} catch (err) {
 		slot.error = err instanceof Error ? err : new Error(String(err));
+		const abortedCall = usageLedger.settleTurn(slot.agentRunName);
+		if (abortedCall) creditLlmCall(abortedCall);
 	} finally {
 		slot.endedAt = args.now();
 		await emitTrace(args.ctx.traceEmitter, {
@@ -6536,6 +6539,50 @@ function createSteerableSandboxSession(args) {
 		let handle;
 		state.session = () => handle;
 		let nextPrompt = args.taskToPrompt(task);
+		const usageLedger = createSandboxUsageLedger(args.harness);
+		let usageFailure;
+		const readUsage = (read) => {
+			try {
+				return read();
+			} catch (err) {
+				tokensKnown = false;
+				const message = err instanceof Error ? err.message : String(err);
+				usageFailure ??= message;
+				activity.push({
+					at: now(),
+					kind: "note",
+					label: "usage-unreadable",
+					detail: message
+				});
+				return;
+			}
+		};
+		function* creditCall(call) {
+			const callTokensKnown = call.tokensKnown !== false && typeof call.tokensIn === "number" && typeof call.tokensOut === "number";
+			if (!callTokensKnown) tokensKnown = false;
+			const input = call.tokensIn ?? 0;
+			const output = call.tokensOut ?? 0;
+			if (input || output || !callTokensKnown) {
+				const usage = {
+					kind: "tokens",
+					input,
+					output,
+					...callTokensKnown ? {} : { tokensKnown: false },
+					...promptCacheTokenClasses(call.tokensIn, call.promptCache)
+				};
+				addTokenUsage(tokens, usage);
+				yield usage;
+			}
+			if (typeof call.costUsd === "number" && call.costUsd > 0) {
+				usd += call.costUsd;
+				yield {
+					kind: "cost",
+					usd: call.costUsd,
+					usdKnown: false,
+					provenance: "uncaptured"
+				};
+			}
+		}
 		try {
 			for (let turn = 0; maxTurns === 0 || turn < maxTurns; turn += 1) {
 				const pending = args.inbox.drain();
@@ -6584,39 +6631,23 @@ function createSteerableSandboxSession(args) {
 							kind: "progress",
 							progress
 						};
-						const call = extractLlmCallEvent(event, spec.name ?? String(args.harness));
+						const call = readUsage(() => usageLedger.observe(event, spec.name ?? String(args.harness)));
 						if (call) {
 							sawLlmCall = true;
-							const callTokensKnown = call.tokensKnown !== false && typeof call.tokensIn === "number" && typeof call.tokensOut === "number";
-							if (!callTokensKnown) tokensKnown = false;
-							const input = call.tokensIn ?? 0;
-							const output = call.tokensOut ?? 0;
-							if (input || output || !callTokensKnown) {
-								const usage = {
-									kind: "tokens",
-									input,
-									output,
-									...callTokensKnown ? {} : { tokensKnown: false },
-									...promptCacheTokenClasses(call.tokensIn, call.promptCache)
-								};
-								addTokenUsage(tokens, usage);
-								yield usage;
-							}
-							if (typeof call.costUsd === "number" && call.costUsd > 0) {
-								usd += call.costUsd;
-								yield {
-									kind: "cost",
-									usd: call.costUsd,
-									usdKnown: false,
-									provenance: "uncaptured"
-								};
-							}
+							yield* creditCall(call);
 						}
 						assertSandboxServedModel(event, requested);
 					}
+					const harnessCall = readUsage(() => usageLedger.settleTurn(spec.name ?? String(args.harness)));
+					if (harnessCall) {
+						sawLlmCall = true;
+						yield* creditCall(harnessCall);
+					}
 					state.latestOutcome = outcomeTracker.finish();
 				} catch (e) {
 					cleanup();
+					const abortedCall = readUsage(() => usageLedger.settleTurn(spec.name ?? String(args.harness)));
+					if (abortedCall) yield* creditCall(abortedCall);
 					if (isInterruptAbort(e, interruptSig, signal, args.controller.signal)) {
 						activity.push({
 							at: now(),
@@ -6672,7 +6703,8 @@ function createSteerableSandboxSession(args) {
 			...state.servedBackend === void 0 ? {} : { servedBackend: state.servedBackend },
 			turns: state.turns,
 			toolCalls: state.toolCalls,
-			...state.latestOutcome ? { outcome: state.latestOutcome } : {}
+			...state.latestOutcome ? { outcome: state.latestOutcome } : {},
+			...usageFailure === void 0 ? {} : { tokensUnknownReason: usageFailure }
 		};
 		const verdict = state.latestOutcome ? projectSandboxOutcome(state.latestOutcome).verdict : void 0;
 		state.artifact = {
@@ -10735,7 +10767,7 @@ async function teardownExecutor(executor, grace, deadlineAtMs, now) {
 *
 *   1. `SuperviseOptions.maxLiveWorkers` (`src/runtime/supervise/supervise.ts`) — supervised-tree
 *      level. One shared Scope counter bounds every spawned manager and leaf in the recursive tree;
-*      `spawn_agent` fails closed with `error: 'max-live-workers'` past it. Unset ⇒ NO tree cap.
+*      `spawn_worker` fails closed with `error: 'max-live-workers'` past it. Unset ⇒ NO tree cap.
 *      `CoordinationToolsOptions.maxLiveWorkers` remains the local form for a toolbox mounted on a
 *      caller-owned Scope that has no tree limit.
 *   2. `SandboxLineage`'s `maxConcurrency` / `DEFAULT_FORK_CONCURRENCY = 4`
@@ -15506,4 +15538,4 @@ function isNonEmptySpend(s) {
 //#endregion
 export { bridgeModelRouteRefusal as $, INTELLIGENCE_WIRE_VERSION as $t, readWorkerInteractiveBinding as A, controlProfileMaterialization as An, peerMailVerbNames as At, retainedCreateMaterial as B, sandboxActProfileMaterialization as Bn, createSandboxLineage as Bt, timerAt as C, LOCAL_HARNESSES as Cn, AUTHORITY_MARKERS as Ct, readWorkerInteractiveAdmissions as D, CodexExecutionDiagnosticError as Dn, createPeerMailbox as Dt, interactiveAdmissionSeamKey as E, parseCodexTokenUsage as En, claimsAuthority as Et, startRetainedInteractiveRun as F, promptModelProfileMaterialization as Fn, TERMINAL_DECISIONS as Ft, rollingDispatch as G, routerBrain as Gt, effectiveConcurrency as H, validateProfileMaterialization as Hn, acquireSandbox as Ht, reconnectRetainedRun as I, promptOnlyProfileMaterialization as In, createSandboxForSpec as It, assertValidBudget as J, observedModelMatchesDeclared as Jt, DEFAULT_SUCCESSFUL_SHUTDOWN_MS as K, runBrainLoop as Kt, recoverRetainedRun as L, promptResourceProfileMaterialization as Ln, defaultSelectWinner as Lt, workerInteractiveBindingsDir as M, fullProfileMaterialization as Mn, isTerminalNodeStatus as Mt, reconnectRetainedInteractiveRun as N, profileMaterializationAxes$1 as Nn, createInPlaceCliExecutor as Nt, workerInteractiveAdmissionFile as O, AGENT_PROFILE_MATERIALIZATION_AXES as On, isPeerMailEnvelope as Ot, recoverRetainedInteractiveRun as P, promptControlProfileMaterialization as Pn, taskToPrompt as Pt, bridgeAdmissionRead as Q, traceContextToEnv as Qt, startRetainedRun as R, renderProfileMaterializationIssues as Rn, isTerminalDecision as Rt, pollFor as S, DEFAULT_LOCAL_HARNESS as Sn, createInbox as St, waitUntil as T, localHarnessExecutable as Tn, PEER_MAIL_WIRE_KEY as Tt, freeSlots as U, worktreeCliProfileMaterialization as Un, assertBoxlessPromptOptions as Ut, destroyInteractiveEnvironment as V, unsupportedProfileDimensions as Vn, probeSandboxCapabilities as Vt, queueOf as W, readPromptOptions as Wt, spendFromUsageEvents as X, mergeTraceEnv as Xt, createBudgetPool as Y, createPropagatingTraceEmitter as Yt, bindReusableExecutorExecutionId as Z, readTraceContextFromEnv as Zt, recordScopeOwnerMaterialization as _, runSettledCommand as _n, decodeToolPart as _t, uncertainSpawnBudgets as a, exportEvalRuns as an, createExecutor as at, createWaitProbes as b, createWorktree as bn, createActivityLog as bt, pickBestDelivered as c, loopEventToOtelSpan as cn, createWorktreeCliExecutor as ct, driverChild as d, toOtelAttributes as dn, workerTraceEnv as dt, buildLoopOtelSpans as en, bridgeRuntimeAttachmentsKey as et, withDriverExecutor as f, createRuntimeEventCollector as fn, workerTraceHeaders as ft, meterRuntimeOwnedProviderAttempt as g, sanitizeRuntimeStreamEvent as gn, createPushTraceSource as gt, meterRuntimeOwnedAccounting as h, sanitizeKnowledgeReadinessReport as hn, createSteerableSandboxSession as ht, sumMeasuredSpendFromEvents as i, createOtelExporter as in, cliWorktreeExecutor as it, workerInteractiveBindingFile as j, defineProfileMaterializationContract as jn, isLiveNodeStatus as jt, attachWorker as k, assertProfileMaterialization as kn, peerMailTools as kt, runFinalizer as l, padSpanId as ln, WORKER_TRACE_PROPAGATION as lt, deriveNodeExecutionIdentity as m, sanitizeAgentRuntimeEvent as mn, DEFAULT_SANDBOX_STEERING_MAX_TURNS as mt, createSupervisor as n, buildRuntimeEventOtelSpans as nn, captureReusableExecutorConfig as nt, bestDelivered as o, flatOtelSpan as on, createExecutorRegistry as ot, createScope as p, createRuntimeStreamEventCollector as pn, workerTraceSeamKey as pt, teardownExecutor as q, canonicalObservedModel as qt, maxSeqOf as r, createOpenInferenceFileExporter as rn, cliInPlaceExecutor as rt, collectDelivered as s, generateSpanId as sn, snapshotExecutorConfig as st, createRootHandle as t, buildLoopSpanNodes as tn, bridgeStopSignalKey as tt, runTree as u, padTraceId as un, readWorkerTraceContext as ut, scopeOwnerExecutorNodeContext as v, runWorktreeHarness as vn, sandboxSessionTraceSource as vt, validateWaitSpec as w, harnessSupportsReasoningEffort as wn, DEFAULT_PEER_MAIL_LIMITS as wt, isWaitOutcome as x, removeWorktree as xn, readWorkerProgress as xt, settledToIteration as y, captureWorktreeDiff as yn, DEFAULT_STALL_AFTER_MS as yt, startRetainedRunInEnvironment as z, renderUnsupported as zn, runAgentRounds as zt };
 
-//# sourceMappingURL=supervisor-CC3qqrVo.js.map
+//# sourceMappingURL=supervisor-DyRbmn1C.js.map