@tangle-network/agent-runtime 0.273.5 → 0.277.0

package/dist/{runtime-CHxXrUeZ.js → runtime-DoFvrisL.js}

@@ -1,16 +1,16 @@
-import { $i as detachedSnapshot, Ai as readCommittedJsonLines, Bn as runAgentRounds, Bt as SpawnEventIndex, Ca as stringifySafe, Ct as rollingDispatch, Gn as readPromptOptions, Hn as probeSandboxCapabilities, Ht as encodeResultBlob, Ia as armDeadlineTimer, Jn as canonicalObservedModelParts, Kn as routerBrain, M as withDriverExecutor, O as verdictFromJudgeScore, Oa as zeroSpend, Ot as createExecutor, Qi as detachedFrozen, Rn as defaultSelectWinner, Rt as InMemoryResultBlobStore, Sa as sleep, Ut as insideCursorNamespace, Vn as createSandboxLineage, Vt as assertContentAddress, W as settledToIteration, Wn as assertBoxlessPromptOptions, Xi as executableAgentProfileSnapshot, Xt as isTraceAnalysisStore, Zi as executableAgentSpecSnapshot, Zn as LEAF_CONTINUATION_TASK, _ as CheckUnavailableError, ai as harnessInvocation, ar as createPushTraceSource, ga as cloneSpend, ha as chargedTokens, ir as unavailableSignalOfFailure, ji as writeAllBytes, jt as createWorktreeCliExecutor, ki as prepareJsonlAppend, kr as contentAddress, kt as createExecutorRegistry, ma as addSpend, nr as sandboxClientAsProvider, qn as runBrainLoop, r as createSupervisor, rr as unavailablePauseMs, ua as notifySandboxEventObserver, va as hasCompleteCacheBreakdown, xa as randomSuffix, ya as isAbortError, zt as InMemorySpawnJournal } from "./supervisor-DdKPGluv.js";
+import { $i as detachedSnapshot, Ai as readCommittedJsonLines, Bn as runAgentRounds, Bt as SpawnEventIndex, Ca as stringifySafe, Ct as rollingDispatch, Gn as readPromptOptions, Hn as probeSandboxCapabilities, Ht as encodeResultBlob, Ia as armDeadlineTimer, Jn as canonicalObservedModelParts, Kn as routerBrain, M as withDriverExecutor, O as verdictFromJudgeScore, Oa as zeroSpend, Ot as createExecutor, Qi as detachedFrozen, Rn as defaultSelectWinner, Rt as InMemoryResultBlobStore, Sa as sleep, Ut as insideCursorNamespace, Vn as createSandboxLineage, Vt as assertContentAddress, W as settledToIteration, Wn as assertBoxlessPromptOptions, Xi as executableAgentProfileSnapshot, Xt as isTraceAnalysisStore, Zi as executableAgentSpecSnapshot, Zn as LEAF_CONTINUATION_TASK, _ as CheckUnavailableError, ai as harnessInvocation, ar as createPushTraceSource, ga as cloneSpend, ha as chargedTokens, ir as unavailableSignalOfFailure, ji as writeAllBytes, jt as createWorktreeCliExecutor, ki as prepareJsonlAppend, kr as contentAddress, kt as createExecutorRegistry, ma as addSpend, nr as sandboxClientAsProvider, qn as runBrainLoop, r as createSupervisor, rr as unavailablePauseMs, ua as notifySandboxEventObserver, va as hasCompleteCacheBreakdown, xa as randomSuffix, ya as isAbortError, zt as InMemorySpawnJournal } from "./supervisor-CjNbOcLO.js";
 import { m as ValidationError, n as AnalystError, s as PlannerError } from "./errors-DodWX-cb.js";
-import { E as profileChatClient, S as ObservationError, T as renderReport, b as sample, k as strategyAuthorMethod, v as refine, w as observe, x as sampleThenRefine, y as runAgentic } from "./structural-rollout-DCUprNpH.js";
+import { E as profileChatClient, S as ObservationError, T as renderReport, b as sample, k as strategyAuthorMethod, v as refine, w as observe, x as sampleThenRefine, y as runAgentic } from "./structural-rollout-VR-j5t9d.js";
 import { a as concreteProfileModel, c as profileModelExecutionSettings, i as concreteModelId, l as profileProviderModel, o as enforceTokenLimits, t as assertExecutableAgentProfile } from "./model-policy-BbSCSak0.js";
 import { C as canonicalCandidateDigest$1, D as immutableCandidateValue, k as sha256Bytes$1, m as captureMaterializedWorkspace } from "./workspace-archive-C9lgf77y.js";
 import { i as notifyRuntimeHookEvent } from "./runtime-hooks-tXpAarhW.js";
 import { i as redactProtectedValue, r as redactProtectedReason } from "./protected-redaction-DylCC_ot.js";
-import { S as foldCoordinationRecords, y as createInMemoryRunContext } from "./coordination-driver-CkcW4dxU.js";
-import { C as gateOnDeliverable, r as supervise, w as mapExecutorResult } from "./supervise-Dej5aSDa.js";
+import { S as foldCoordinationRecords, y as createInMemoryRunContext } from "./coordination-driver-ZfUf-H4H.js";
+import { C as gateOnDeliverable, r as supervise, w as mapExecutorResult } from "./supervise-BoCOUAmr.js";
 import "./run-layout-vwd4SwX_.js";
-import "./provision-supervisor-BKAvsDtP.js";
-import "./delegate-Dvb1-mE-.js";
-import "./graph-C69Uxtzt.js";
+import "./provision-supervisor-BgjtrTlv.js";
+import "./delegate-DXOyzIq0.js";
+import "./graph-sGWwXC6E.js";
 import { agentProfileSchema, canonicalAgentProfileDigest, canonicalCandidateDigest, sha256Bytes, validateAgentProfileSecurity } from "@tangle-network/agent-interface";
 import { CODING_HARNESSES, DEFAULT_TRACE_ANALYST_KINDS, InMemoryTraceStore, OUTPUT_VALUE, benjaminiHochberg, buildTrajectory, computeFindingId as computeFindingId$1, confidenceInterval, createTraceAnalyst, expandProfileAxes, makeFinding as makeFinding$1, pairedBootstrap, paretoFrontier, wilcoxonSignedRank, wilson } from "@tangle-network/agent-eval";
 import { createAgentRunOutcomeTracker } from "@tangle-network/sandbox/runtime";
@@ -1957,9 +1957,9 @@ async function runInBox(options, placement) {
 		options.signal?.throwIfAborted();
 		box = await placement.client.createIsolated({
 			environment: placement.environment,
-			agent: false,
+			agent: placement.containers === true,
 			bare: false,
-			ephemeral: true,
+			ephemeral: placement.containers !== true,
 			egressPolicy: egressFor(placement),
 			...placement.resources ? { resources: placement.resources } : {},
 			maxLifetimeSeconds: Math.ceil(timeoutMs / 1e3) + BOX_SETUP_SECONDS,
@@ -2273,12 +2273,19 @@ async function removeScratch(path) {
 * pauses instead of blaming the director.
 *
 * The box receives the program's files at its working directory, the submitted result at
-* `_input/result.json` when there is one, and the sealed cases at `_sealed/` only when the read
-* scores a run or a version. `CHECK_SET` is `development` or `sealed`; `CHECK_RESULT` names the
-* result file when present. Sealed cases never reach an in-run read, so their lines never reach
-* the director.
+* `_input/result.json` when there is one, the run's state at `_input/state/` when the host
+* captures it, and the sealed cases at `_sealed/` only when the read scores a run or a version.
+* `CHECK_SET` is `development` or `sealed`; `CHECK_RESULT` names the result file and
+* `CHECK_STATE` the state directory when present. Sealed cases never reach an in-run read, so
+* their lines never reach the director.
+*
+* The state is what a check reads when the work is not in the submitted value. A Terminal-Bench
+* task is done when its container holds the right files, and the box that holds that container is
+* the run's, which the check box cannot reach. So the host copies the declared paths out before
+* each read ({@link DeclaredCheckStateCapture}), and the check reads those bytes.
 */
 const INPUT_DIR = "_input";
+const STATE_DIR = "state";
 const SEALED_DIR = "_sealed";
 const DEFAULT_TIMEOUT_MS = 15 * 6e4;
 /** Refuse a malformed declaration before any compute. */
@@ -2295,6 +2302,7 @@ function assertDeclaredCheck(check, context) {
 	if (check.feedback !== "verbatim" && check.feedback !== "pass-only") fail("feedback must be 'verbatim' or 'pass-only'");
 	if (check.sealed !== void 0 && !/^sha256:[0-9a-f]{64}$/u.test(check.sealed.digest ?? "")) fail("sealed.digest must be a sha256 digest");
 	for (const name of check.secrets ?? []) if (!/^[A-Z_][A-Z0-9_]*$/u.test(name)) fail(`secret name ${JSON.stringify(name)} is not an env name`);
+	if (check.containers !== void 0 && typeof check.containers !== "boolean") fail("containers must be a boolean");
 }
 /** The canonical digest of a directory's files: the digest a record names a program by. */
 async function checkProgramDigest(dir) {
@@ -2307,13 +2315,15 @@ function declaredCheckDigest(check) {
 		sealed: check.sealed?.digest ?? null,
 		command: check.command,
 		environment: check.environment,
-		pass: check.pass
+		pass: check.pass,
+		...check.containers === true ? { containers: true } : {}
 	})));
 }
 /**
 * Read the check once. `result` is the submitted result, absent for a read of the run's state.
-* `set: 'sealed'` adds the sealed cases. Throws {@link CheckUnavailableError} when the program
-* could not run or printed no score.
+* `state` is a local directory of the run's state; the box receives a copy of its files at
+* `_input/state/`. `set: 'sealed'` adds the sealed cases. Throws {@link CheckUnavailableError} when
+* the program could not run or printed no score.
 */
 async function readDeclaredCheck(check, placement, read) {
 	if (read.set === "sealed" && check.sealed === void 0) throw new ValidationError("readDeclaredCheck: a sealed read needs check.sealed");
@@ -2335,10 +2345,15 @@ async function readDeclaredCheck(check, placement, read) {
 			CHECK_SET: read.set
 		};
 		if (read.result !== void 0) {
-			await mkdir(join(tree, INPUT_DIR));
+			await mkdir(join(tree, INPUT_DIR), { recursive: true });
 			await writeFile(join(tree, INPUT_DIR, "result.json"), `${JSON.stringify(read.result)}\n`);
 			env.CHECK_RESULT = `${INPUT_DIR}/result.json`;
 		}
+		if (read.state !== void 0) {
+			await mkdir(join(tree, INPUT_DIR, STATE_DIR), { recursive: true });
+			await writeFiles((await captureMaterializedWorkspace(read.state)).files, join(tree, INPUT_DIR, STATE_DIR));
+			env.CHECK_STATE = `${INPUT_DIR}/${STATE_DIR}`;
+		}
 		if (read.set === "sealed" && check.sealed !== void 0) await copyVerified(check.sealed.dir, check.sealed.digest, join(tree, SEALED_DIR), "sealed cases");
 		return verdictOf(await runIsolatedCheck({
 			workspaceRoot: workspace,
@@ -2352,7 +2367,8 @@ async function readDeclaredCheck(check, placement, read) {
 				builderAccounts: placement.builderAccounts,
 				environment: check.environment,
 				...check.egress === void 0 ? {} : { egress: check.egress },
-				...check.resources === void 0 ? {} : { resources: check.resources }
+				...check.resources === void 0 ? {} : { resources: check.resources },
+				...check.containers === true ? { containers: true } : {}
 			}
 		}), check.pass);
 	} finally {
@@ -2362,15 +2378,42 @@ async function readDeclaredCheck(check, placement, read) {
 		});
 	}
 }
-/** The declared check as a manager's completion check: every in-run read uses development cases. */
-function declaredCheckDeliverable(check, placement) {
+/**
+* The declared check as a manager's completion check: every in-run read uses development cases.
+* With `state`, each read first captures the run's state and the check reads it beside the
+* submitted result, on `submit_result` and at a turn end alike.
+*/
+function declaredCheckDeliverable(check, placement, options = {}) {
 	assertDeclaredCheck(check, "declaredCheckDeliverable");
-	return {
-		check: (result) => readDeclaredCheck(check, placement, {
+	const { state } = options;
+	if (state !== void 0 && typeof state !== "function") throw new ValidationError("declaredCheckDeliverable: state must be a capture function");
+	const read = async (result) => {
+		if (state === void 0) return readDeclaredCheck(check, placement, {
 			result,
 			set: "development"
-		}),
-		checkState: () => readDeclaredCheck(check, placement, { set: "development" }),
+		});
+		const into = await mkdtemp(join(tmpdir(), "declared-check-state-"));
+		try {
+			try {
+				await state(into);
+			} catch (error) {
+				throw new CheckUnavailableError(`the run's state could not be captured: ${error instanceof Error ? error.message : String(error)}`, { cause: error });
+			}
+			return await readDeclaredCheck(check, placement, {
+				result,
+				state: into,
+				set: "development"
+			});
+		} finally {
+			await rm(into, {
+				recursive: true,
+				force: true
+			});
+		}
+	};
+	return {
+		check: (result) => read(result),
+		checkState: () => read(),
 		feedback: check.feedback,
 		...check.sealed === void 0 ? {} : { sealed: true },
 		...check.describe === void 0 ? {} : { describe: check.describe }
@@ -2434,7 +2477,10 @@ async function copyVerified(from, digest, to, label) {
 	const captured = await captureMaterializedWorkspace(from);
 	const actual = canonicalCandidateDigest(captured.manifest);
 	if (actual !== digest) throw new ValidationError(`readDeclaredCheck: the ${label} at ${from} is ${actual}; the record names ${digest}`);
-	for (const file of captured.files) {
+	await writeFiles(captured.files, to);
+}
+async function writeFiles(files, to) {
+	for (const file of files) {
 		const path = join(to, file.path);
 		await mkdir(dirname(path), { recursive: true });
 		await writeFile(path, file.bytes, { mode: file.mode });
@@ -2894,6 +2940,7 @@ function localSandboxClient(opts) {
 					type: "llm_call",
 					data: {
 						model,
+						tools: mcp.tools.map((t) => t.function.name),
 						...r.tokensKnown === false ? { tokensKnown: false } : {
 							tokensIn: r.usage.input,
 							tokensOut: r.usage.output
@@ -9146,4 +9193,4 @@ function tail(s) {
 //#endregion
 export { equalKOnCost as $, SqlResultBlobStore as $t, unsafeInProcessRunner as A, loopDispatch as At, strategyAuthorContract as B, deterministicCompletion as Bt, withUntrackedArtifacts as C, HarvestError as Ct, patchDelivered as D, localSandboxClient as Dt, createSqlRunContext as E, resolveSandboxClient as Et, pickChampion as F, declaredCheckDigest as Ft, sharedBoxPlacement as G, renderLeaderboardHtml as Gt, DEFAULT_SHARED_BOX_RESOURCES as H, stopSentinel as Ht, runStrategyEvolution as I, declaredCheckJudge as It, SandboxRunAbortError as J, renderPairwiseMarkdown as Jt, sharedBoxRefusal as K, renderLeaderboardMarkdown as Kt, selectChampion as L, readDeclaredCheck as Lt, chatWorkerSeam as M, assertDeclaredCheck as Mt, createChatSessionStore as N, checkProgramDigest as Nt, runCoderChecks as O, inlineSandboxClient as Ot, discriminatingMeans as P, declaredCheckDeliverable as Pt, promotionGate as Q, openSqlRunStore as Qt, assertStrategyContract as R, runIsolatedCheck as Rt, copyUntrackedIntoClone as S, inProcessSandboxClient as St, createFencedSqlRunContext as T, defineLeaderboard as Tt, DEFAULT_SHARED_BOX_WORKERS as U, leaderboard as Ut, strategyAuthorSystemPrompt as V, sentinelCompletion as Vt, ROUTER_CLIENT_HEADER as W, pairwiseSignificance as Wt, printBenchmarkReport as X, defaultAuditorInstruction as Xt, openSandboxRun as Y, auditIntent as Yt, runBenchmark as Z, SqlRunOwnershipError as Zt, NOTE_MAX_CHARS as _, assertTraceDerivedFindings as _t, localShell as a, sanitizeMcpToolSchema as an, registerShape as at, composeWorkerEvidence as b, registryScopeAnalyst as bt, createVerifierEnvironment as c, resolveMcpServerLaunch as cn, renderCorpusToInstructions as ct, harvestSurfaceDiffs as d, createTangleSandboxExactProcessProvider as dn, loopUntil as dt, SqlSpawnJournal as en, trajectoryReport as et, analystsFromRegistry as f, panel as ft, EVIDENCE_MAX_CHARS as g, widen as gt, worktreeFanout as h, verify as ht, jjWorkspace as i, createMcpEnvironment as in, createShapeRegistry as it, chatTransportExecutor as j, superviseDispatch as jt, codeModeSupervisorTools as k, loopCampaignDispatch as kt, boxSurfaceReader as l, resolveSecretEnv as ln, fanout as lt, superviseSurface as m, selectValidWinner as mt, makeFinding$1 as n, connectStdioMcp as nn, runPersonified as nt, runInWorkspace as o, envKeyProvider as on, FileCorpus as ot, failuresAnalyst as p, pipeline as pt, sharedWorkerClientName as q, renderLeaderboardSvg as qt, gitWorkspace as r, materializeLocalMcp as rn, builtinShapes as rt, createWaterfallCollector as s, mcpSecretEnvMetadataKey as sn, InMemoryCorpus as st, computeFindingId$1 as t, McpSpawnFault as tn, definePersona as tt, fsSurfaceReader as u, secretEnvOfMcpServer as un, flatWidenGate as ut, VERIFY_TAIL_CHARS as v, buildSteerContext as vt, analyzeTrace as w, harvestCorpus as wt, settledWorkerOut as x, observationFromRegistry as xt, closingWorkerNote as y, createScopeAnalyst as yt, authorStrategy as z, completionAuthorizes as zt };
 
-//# sourceMappingURL=runtime-CHxXrUeZ.js.map
+//# sourceMappingURL=runtime-DoFvrisL.js.map