@tangle-network/agent-runtime 0.132.0 → 0.132.3

package/dist/testing.d.ts

@@ -1,5 +1,5 @@
-import { Dt as RunGraphTestOptions, Ga as SuperviseTestOptions, an as DriverAgentOptions, fo as supervisorAgentWithTestBrain, jt as runGraphWithTestBrain, on as driverAgent, qa as superviseWithTestBrain, ro as SupervisorAgentTestDeps } from "./index-BLIEkf7Y.js";
-import { J as ToolLoopCallContext, Y as ToolLoopChat } from "./runtime-B0rxwNg7.js";
+import { Dt as RunGraphTestOptions, Ga as SuperviseTestOptions, an as DriverAgentOptions, fo as supervisorAgentWithTestBrain, jt as runGraphWithTestBrain, on as driverAgent, qa as superviseWithTestBrain, ro as SupervisorAgentTestDeps } from "./index-D9dVsozT.js";
+import { J as ToolLoopCallContext, Y as ToolLoopChat } from "./runtime-aXvRThzL.js";
 import { AgentImprovementProposal, AgentProfile, AgentProfileImprovementMeasuredComparison, SandboxSizePreset } from "@tangle-network/agent-interface";
 //#region src/testing/index.d.ts
 /** A proposal produced by Runtime's opaque profile-improvement path. */

package/dist/testing.js

@@ -1,12 +1,12 @@
 import { $ as parseExactAgentProfile, K as applyExactAgentProfileDiff, rt as canonicalCandidateDigest$1, st as immutableCandidateValue } from "./prepare-BQ82nG06.js";
-import { f as verifyAgentImprovementProposal } from "./improvement-cycle-0H8-10KE.js";
-import { c as supervisorAgentWithTestBrain, dt as driverAgent, r as superviseWithTestBrain } from "./supervise-C9iPh1y6.js";
-import { i as runGraphWithTestBrain } from "./graph-DyWVymtQ.js";
+import { f as verifyAgentImprovementProposal } from "./improvement-cycle-DgZtiPDT.js";
+import { c as supervisorAgentWithTestBrain, dt as driverAgent, r as superviseWithTestBrain } from "./supervise-BDj-uFfb.js";
+import { i as runGraphWithTestBrain } from "./graph-B9hFN-zX.js";
 import { SANDBOX_SIZE_PRESET_NAMES } from "@tangle-network/agent-interface";
 //#region src/testing/fixtures/agent-improvement-proposal.json
 var agent_improvement_proposal_default = {
 	changedSurfaces: ["prompt"],
-	digest: "sha256:db0f2ca349aa300bbfc93d6893c1c0b04e3f4471d4c64ed43f3d423ca28b9e94",
+	digest: "sha256:2215a3e3c31ee909f8b4c1cd445614c86e869638aff847268f89e6bf81a4953f",
 	evaluation: {
 		"decision": {
 			"contributingChecks": [
@@ -4565,7 +4565,7 @@ var agent_improvement_proposal_default = {
 		],
 		"metadata": {
 			"fixture": "agent-improvement-proposal",
-			"runtimeVersion": "0.132.0"
+			"runtimeVersion": "0.132.3"
 		},
 		"objectives": [
 			{
@@ -4676,8 +4676,8 @@ var agent_improvement_proposal_default = {
 			"baselineContentHash": "sha256:5c21ee53e513fc604cb09754e21c392b24a424da0ef37dbf8f1ee4a8a0b08f09",
 			"candidateContentHash": "sha256:60fcbb1c728194bd51d7d19cb732d1c3f1881dce7e0a6266b41c8b98cfd65693",
 			"kind": "agent-eval-loop",
-			"recordDigest": "sha256:c3ee11a1cc489f96ad93c59cfea30dc201c7e12a11bc2dbe5be9f7a23d54b58d",
-			"runId": "agent-runtime-0.132.0-proposal-fixture",
+			"recordDigest": "sha256:c3d399f3789ec252856a5e03699f77cb0c0062e579a6d1d717c8352f4adabca4",
+			"runId": "agent-runtime-0.132.3-proposal-fixture",
 			"schema": "agent-candidate-experiment"
 		}
 	},
@@ -4700,13 +4700,13 @@ var agent_improvement_proposal_default = {
 	}],
 	kind: "agent-improvement-proposal",
 	proposedAt: "2026-07-10T01:00:00.000Z",
-	runId: "agent-runtime-0.132.0-proposal-fixture"
+	runId: "agent-runtime-0.132.3-proposal-fixture"
 };
 //#endregion
 //#region src/testing/fixtures/agent-profile-improvement-proposal.json
 var agent_profile_improvement_proposal_default = {
 	changedSurfaces: ["prompt", "skills"],
-	digest: "sha256:dd7ac948fe129ee1cb46503595cf23fa8bf0ec04c0b549b69533a2fe6385e76d",
+	digest: "sha256:87c23c9c41322bd6b6878ea775d1bbe4869f5a4ddc447e81ed2ecbc293a3c1d5",
 	evaluation: {
 		"decision": {
 			"contributingChecks": [
@@ -6340,7 +6340,7 @@ var agent_profile_improvement_proposal_default = {
 		],
 		"metadata": {
 			"fixture": "agent-profile-improvement-proposal",
-			"runtimeVersion": "0.132.0"
+			"runtimeVersion": "0.132.3"
 		},
 		"objectives": [
 			{
@@ -6451,7 +6451,7 @@ var agent_profile_improvement_proposal_default = {
 			"baselineContentHash": "sha256:21c495a37c418c10bde64fbaa188beddeed31f1f051ea60a6a6582a9ee0db704",
 			"candidateContentHash": "sha256:103f77bc8481601eef1ad5fe6ba84a40dffabc3a44f421f8c8559121edab84e9",
 			"kind": "agent-eval-loop",
-			"recordDigest": "sha256:30bf175ad85605c16ef63c4fa585b9f5628e146524b3ae8712f4c0ee986ff1e4",
+			"recordDigest": "sha256:6d127c92643517af98a9bf25dcdf48a96fae2e45ee58404d1c8bf0fb9757cc71",
 			"runId": "profile-improvement-1",
 			"schema": "agent-profile-improvement-experiment"
 		}

package/dist/{types-CIg5DBNF.d.ts → types-B9KAen_0.d.ts}

@@ -721,6 +721,15 @@ declare function readWorkerTraceContext(ctx: WorkerTraceSeamCarrier): TraceConte
  * caller's own seam env so a deliberately-set id wins (see the precedence note above).
  */
 declare function workerTraceEnv(ctx: WorkerTraceSeamCarrier): Record<string, string>;
+/**
+ * The trace request headers for a worker dispatched over the cli-bridge HTTP transport — W3C
+ * `traceparent` plus the legacy `x-trace-id` / `x-parent-span-id` pair the bridge also reads —
+ * EMPTY when the run records no spans, which keeps the untraced request byte-identical. Derived
+ * from the same dual-write the env channel uses ({@link workerTraceEnv}), so the header and env
+ * spellings can never name different traces. `traceparent` is present only when a parent span id
+ * exists (the W3C grammar requires one); the legacy pair still carries a lone trace id.
+ */
+declare function workerTraceHeaders(ctx: WorkerTraceSeamCarrier): Record<string, string>;
 //#endregion
 //#region src/runtime/supervise/types.d.ts
 /** Options for `Scope.wait`. `label` is the wait's identity within its parent scope — it is what
@@ -1709,6 +1718,23 @@ interface NoWinnerError {
   message: string;
   stack?: string;
 }
+/** The accounting channels a usage gap leaves incomplete. */
+type SpendChannel = 'tokens' | 'usd';
+/**
+ * One journaled node whose usage accounting is incomplete — the named gap behind a `false`
+ * `tokensKnown`/`usdKnown` on a terminal `spentTotal`. `never-settled`: the spawn is durable but
+ * no terminal record landed, so the whole subtree is unaccounted on every channel and
+ * `spentTotal` charges its budget ceiling instead of a fabricated zero. `unreported`: a settled
+ * or metered record landed without a complete provider receipt, so the summed numbers are a
+ * floor on the named channels, never the measured total.
+ */
+interface SpendGap {
+  readonly id: NodeId;
+  /** The spawn label, when the node's `spawned` event is in this journal tree. */
+  readonly label?: string;
+  readonly kind: 'never-settled' | 'unreported';
+  readonly channels: ReadonlyArray<SpendChannel>;
+}
 /** Typed terminal result (M2) — a no-winner is NEVER coerced to a best-effort output. */
 type SupervisedResult<Out> = {
   kind: 'winner';
@@ -1716,10 +1742,22 @@ type SupervisedResult<Out> = {
   outRef: string;
   verdict?: DefaultVerdict;
   tree: TreeView;
+  /** The run's terminal accounting. `iterations`/`tokens`/`usd` are per-channel journal sums;
+   *  `ms` is the wall clock from supervise start (the ORIGINAL root instant on a resumed run)
+   *  to this terminal state — executors under-report their own `ms` and parallel children
+   *  overlap, so a per-event sum cannot state the run's real duration. `tokensKnown`/`usdKnown`
+   *  are always explicit here: `true` is the checked claim that every spawn reached a terminal
+   *  record and every settled/metered record carried a complete receipt on that channel;
+   *  `false` comes with the unaccounted nodes named in `spendGaps`. */
   spentTotal: Spend;
+  /** The journaled nodes whose usage accounting is incomplete — the named gaps behind a
+   *  `false` `tokensKnown`/`usdKnown` on `spentTotal`. Present exactly when non-empty. */
+  spendGaps?: ReadonlyArray<SpendGap>;
   /** Where `spentTotal` went: `driverInference` = the drivers' own chat turns (metered via
    *  `Scope.meter`); `childWork` = every spawned child's reconciled spend (the journal sum).
-   *  `driverInference + childWork === spentTotal`. Present whenever any driver metered. */
+   *  `driverInference + childWork === spentTotal` on `iterations`/`tokens`/`usd`; the
+   *  breakdown's `ms` fields stay executor-reported sums while `spentTotal.ms` is wall clock.
+   *  Present whenever any driver metered. */
   spentBreakdown?: {
     driverInference: Spend;
     childWork: Spend;
@@ -1738,8 +1776,12 @@ type SupervisedResult<Out> = {
   downCount: number;
   /** The conserved spend incurred before the run failed — real cost is paid even when no
    *  worker delivers, so the caller always learns what the delegation actually spent. Summed
-   *  off the same journal the `winner` path reads. */
+   *  off the same journal the `winner` path reads, with the same contract: wall-clock `ms`,
+   *  explicit `tokensKnown`/`usdKnown`, gaps named in `spendGaps`. */
   spentTotal: Spend;
+  /** The journaled nodes whose usage accounting is incomplete — the named gaps behind a
+   *  `false` `tokensKnown`/`usdKnown` on `spentTotal`. Present exactly when non-empty. */
+  spendGaps?: ReadonlyArray<SpendGap>;
   /** Never present on a lifecycle arm — the discriminant, not prose, is what makes
    *  `if (r.reason === 'driver-failed') r.error.message` compile and every other arm refuse it. */
   error?: never;
@@ -1757,8 +1799,12 @@ type SupervisedResult<Out> = {
   downCount: number;
   /** The conserved spend incurred before the run failed — real cost is paid even when no
    *  worker delivers, so the caller always learns what the delegation actually spent. Summed
-   *  off the same journal the `winner` path reads. */
+   *  off the same journal the `winner` path reads, with the same contract: wall-clock `ms`,
+   *  explicit `tokensKnown`/`usdKnown`, gaps named in `spendGaps`. */
   spentTotal: Spend;
+  /** The journaled nodes whose usage accounting is incomplete — the named gaps behind a
+   *  `false` `tokensKnown`/`usdKnown` on `spentTotal`. Present exactly when non-empty. */
+  spendGaps?: ReadonlyArray<SpendGap>;
   /** The driver's own rejection, carried across the typed no-winner boundary so the failure is
    *  recoverable by the caller. A non-`Error` rejection is normalized, never dropped. */
   error: NoWinnerError;
@@ -1808,5 +1854,5 @@ interface WidenGate<Out> {
   readonly judgeExempt?: boolean;
 }
 //#endregion
-export { WorkerTraceSeamCarrier as $, waitUntil as $t, RootMaterialization as A, RuntimeEventCollector as At, Spend as B, sanitizeKnowledgeReadinessReport as Bt, NodeStatus as C, createOtelExporter as Ct, ResumedKeyState as D, padSpanId as Dt, ResultBlobStore as E, loopEventToOtelSpan as Et, SpawnEvent as F, SanitizedKnowledgeReadinessReport as Ft, TreeView as G, WaitProbeRegistry as Gt, SupervisedResult as H, PendingWait as Ht, SpawnJournal as I, SanitizedKnowledgeRequirement as It, WaitOpts as J, createWaitProbes as Jt, UnknownMaterializationReason as K, WaitRejection as Kt, SpawnOpts as L, createRuntimeEventCollector as Lt, Runtime as M, RuntimeStreamEventSink as Mt, Scope as N, RuntimeStreamEventSummary as Nt, ResumedWork as O, padTraceId as Ot, Settled as P, RuntimeTelemetryOptions as Pt, WorkerTraceResolver as Q, validateWaitSpec as Qt, SpawnPrior as R, createRuntimeStreamEventCollector as Rt, NodeSnapshot as S, createOpenInferenceFileExporter as St, Restart as T, generateSpanId as Tt, Supervisor as U, WaitOutcome as Ut, SteerableRootHandle as V, sanitizeRuntimeStreamEvent as Vt, SupervisorOpts as W, WaitProbe as Wt, WorkerTraceEvidence as X, pollFor as Xt, WidenGate as Y, isWaitOutcome as Yt, WorkerTraceUnavailableReason as Z, timerAt as Zt, MaterializedExecutionIdentity as _, OtelSpan as _t, DefaultVerdict as a, decodeToolPart as an, mergeTraceEnv as at, NodeExecutionIdentity as b, buildLoopSpanNodes as bt, ExecutorAccounting as c, ActivityNote as cn, EvalRunEvent as ct, ExecutorFactory as d, ScopeProgressInput as dn, EvalRunsExportResult as dt, SessionMessageLike as en, readWorkerTraceContext as et, ExecutorMaterialization as f, WorkerProgress as fn, INTELLIGENCE_WIRE_VERSION as ft, Handle as g, OtelExporter as gt, ExecutorResult as h, OtelExportConfig as ht, Budget as i, createPushTraceSource as in, createPropagatingTraceEmitter as it, RootSignal as j, RuntimeStreamEventCollector as jt, RootHandle as k, toOtelAttributes as kt, ExecutorContext as l, DEFAULT_STALL_AFTER_MS as ln, EvalRunGeneration as lt, ExecutorRegistry as m, readWorkerProgress as mn, OtelAttribute as mt, AgentExecutionRef as n, ToolStepInput as nn, workerTraceSeamKey as nt, ExecutionBindingReceipt as o, sandboxSessionTraceSource as on, readTraceContextFromEnv as ot, ExecutorNodeContext as p, createActivityLog as pn, LoopSpanNode as pt, UsageEvent as q, WaitSpec as qt, AgentSpec as r, TraceSource as rn, TraceContext as rt, Executor as s, ActivityLog as sn, traceContextToEnv as st, Agent as t, SessionTraceBox as tn, workerTraceEnv as tt, ExecutorExecutionBinding as u, ExecutorProgress as un, EvalRunsExportConfig as ut, MaterializedModelIdentity as v, RuntimeEventOtelOptions as vt, ProfileMaterializationReceipt as w, exportEvalRuns as wt, NodeId as x, buildRuntimeEventOtelSpans as xt, NoWinnerError as y, buildLoopOtelSpans as yt, SpawnRejection as z, sanitizeAgentRuntimeEvent as zt };
-//# sourceMappingURL=types-CIg5DBNF.d.ts.map
+export { WorkerTraceUnavailableReason as $, pollFor as $t, RootMaterialization as A, padSpanId as At, Spend as B, createRuntimeEventCollector as Bt, NodeStatus as C, buildLoopSpanNodes as Ct, ResumedKeyState as D, exportEvalRuns as Dt, ResultBlobStore as E, createOtelExporter as Et, SpawnEvent as F, RuntimeStreamEventSink as Ft, Supervisor as G, PendingWait as Gt, SpendGap as H, sanitizeAgentRuntimeEvent as Ht, SpawnJournal as I, RuntimeStreamEventSummary as It, UnknownMaterializationReason as J, WaitProbeRegistry as Jt, SupervisorOpts as K, WaitOutcome as Kt, SpawnOpts as L, RuntimeTelemetryOptions as Lt, Runtime as M, toOtelAttributes as Mt, Scope as N, RuntimeEventCollector as Nt, ResumedWork as O, generateSpanId as Ot, Settled as P, RuntimeStreamEventCollector as Pt, WorkerTraceEvidence as Q, isWaitOutcome as Qt, SpawnPrior as R, SanitizedKnowledgeReadinessReport as Rt, NodeSnapshot as S, buildLoopOtelSpans as St, Restart as T, createOpenInferenceFileExporter as Tt, SteerableRootHandle as U, sanitizeKnowledgeReadinessReport as Ut, SpendChannel as V, createRuntimeStreamEventCollector as Vt, SupervisedResult as W, sanitizeRuntimeStreamEvent as Wt, WaitOpts as X, WaitSpec as Xt, UsageEvent as Y, WaitRejection as Yt, WidenGate as Z, createWaitProbes as Zt, MaterializedExecutionIdentity as _, readWorkerProgress as _n, OtelAttribute as _t, DefaultVerdict as a, ToolStepInput as an, workerTraceSeamKey as at, NodeExecutionIdentity as b, OtelSpan as bt, ExecutorAccounting as c, decodeToolPart as cn, mergeTraceEnv as ct, ExecutorFactory as d, ActivityNote as dn, EvalRunEvent as dt, timerAt as en, WorkerTraceResolver as et, ExecutorMaterialization as f, DEFAULT_STALL_AFTER_MS as fn, EvalRunGeneration as ft, Handle as g, createActivityLog as gn, LoopSpanNode as gt, ExecutorResult as h, WorkerProgress as hn, INTELLIGENCE_WIRE_VERSION as ht, Budget as i, SessionTraceBox as in, workerTraceHeaders as it, RootSignal as j, padTraceId as jt, RootHandle as k, loopEventToOtelSpan as kt, ExecutorContext as l, sandboxSessionTraceSource as ln, readTraceContextFromEnv as lt, ExecutorRegistry as m, ScopeProgressInput as mn, EvalRunsExportResult as mt, AgentExecutionRef as n, waitUntil as nn, readWorkerTraceContext as nt, ExecutionBindingReceipt as o, TraceSource as on, TraceContext as ot, ExecutorNodeContext as p, ExecutorProgress as pn, EvalRunsExportConfig as pt, TreeView as q, WaitProbe as qt, AgentSpec as r, SessionMessageLike as rn, workerTraceEnv as rt, Executor as s, createPushTraceSource as sn, createPropagatingTraceEmitter as st, Agent as t, validateWaitSpec as tn, WorkerTraceSeamCarrier as tt, ExecutorExecutionBinding as u, ActivityLog as un, traceContextToEnv as ut, MaterializedModelIdentity as v, OtelExportConfig as vt, ProfileMaterializationReceipt as w, buildRuntimeEventOtelSpans as wt, NodeId as x, RuntimeEventOtelOptions as xt, NoWinnerError as y, OtelExporter as yt, SpawnRejection as z, SanitizedKnowledgeRequirement as zt };
+//# sourceMappingURL=types-B9KAen_0.d.ts.map

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tangle-network/agent-runtime",
-  "version": "0.132.0",
+  "version": "0.132.3",
   "description": "Shared task-lifecycle skeleton for agents: a recursive loop kernel for chat turns, one-shot tasks, and multi-attempt loops, with trace capture and eval-gated self-improvement. Domain behavior lives in adapters; scoring and ship-gates in @tangle-network/agent-eval.",
   "homepage": "https://github.com/tangle-network/agent-runtime#readme",
   "repository": {
@@ -118,7 +118,7 @@
     "@arethetypeswrong/cli": "0.18.5",
     "@modelcontextprotocol/sdk": "1.30.0",
     "@biomejs/biome": "^2.5.5",
-    "@tangle-network/agent-eval": "0.144.10",
+    "@tangle-network/agent-eval": "0.144.12",
     "@tangle-network/agent-interface": "0.46.1",
     "@tangle-network/sandbox": "0.19.4",
     "@types/node": "26.1.1",
@@ -139,7 +139,7 @@
   },
   "license": "MIT",
   "peerDependencies": {
-    "@tangle-network/agent-eval": ">=0.144.10 <0.145.0",
+    "@tangle-network/agent-eval": ">=0.144.12 <0.145.0",
     "@tangle-network/agent-interface": ">=0.46.1 <0.47.0",
     "@tangle-network/sandbox": ">=0.19.4 <0.20.0"
   },
@@ -150,7 +150,7 @@
   },
   "dependencies": {
     "@tangle-network/agent-core": "0.5.4",
-    "@tangle-network/agent-knowledge": "7.2.0",
+    "@tangle-network/agent-knowledge": "7.2.1",
     "@tangle-network/agent-profile-materialize": "0.13.1",
     "@tangle-network/agent-trace-contract": "^1.0.2",
     "tar-stream": "3.2.0"