@tangle-network/agent-eval 0.125.0 → 0.126.1

package/dist/control.js

@@ -10,11 +10,10 @@ import {
   stopOnNoProgress,
   stopOnRepeatedAction,
   subjectiveEval
-} from "./chunk-R226UZOI.js";
+} from "./chunk-474LBSOX.js";
 import "./chunk-VQMK5FMP.js";
-import "./chunk-LOW3U7JZ.js";
+import "./chunk-IILEIWGW.js";
 import "./chunk-MA6HLL3S.js";
-import "./chunk-GC4ATIKK.js";
 import "./chunk-VSMTAMNK.js";
 import "./chunk-ONWEPEDO.js";
 import "./chunk-PZ5AY32C.js";

package/dist/fuzz.d.ts

@@ -31,29 +31,40 @@ interface CostReceipt extends CostCallBase, CostUsage {
     costUsd: number;
     costUnknown: boolean;
     usageUnknown?: boolean;
+    /** Rates used to estimate cost locally. Absent when cost is provider-reported or unknown. */
     pricing?: {
         inputUsdPerThousand: number;
+        cachedInputUsdPerThousand?: number;
+        cacheWriteUsdPerThousand?: number;
         outputUsdPerThousand: number;
     };
+    /** Cost reported by the provider, not a local token-price calculation. */
     actualCostUsd?: number;
     error?: string;
 }
 interface CostReceiptInput extends CostUsage {
     model: string;
+    /** Caller-supplied rates for a local estimate when the provider does not report billed cost. */
+    customTokenPricing?: CustomTokenPricing;
     actualCostUsd?: number;
     costUnknown?: boolean;
     usageUnknown?: boolean;
 }
 /** Per-million token rates for a model or endpoint not covered by package pricing. */
 interface CustomTokenPricing {
+    /** Non-cached input tokens. */
     inputUsdPerMillion: number;
+    /** Cache-read tokens. Falls back to the normal input rate when omitted. */
+    cachedInputUsdPerMillion?: number;
+    /** Cache-creation or cache-write tokens. Falls back to the normal input rate when omitted. */
+    cacheWriteUsdPerMillion?: number;
     outputUsdPerMillion: number;
 }
 type MaximumCharge = {
     externallyEnforcedMaximumUsd: number;
 } | ({
     customTokenPricing: CustomTokenPricing;
-} & Pick<CostUsage, 'inputTokens' | 'outputTokens'>) | ({
+} & Pick<CostUsage, 'inputTokens' | 'outputTokens' | 'cachedTokens' | 'cacheWriteTokens'>) | ({
     model: string;
 } & CostUsage);
 interface RunPaidCallInput<T> {
@@ -121,6 +132,8 @@ interface CostLedgerFilter {
 interface CostLedgerWaitOptions {
     /** Maximum time to wait for active provider calls. Default 5 seconds. */
     timeoutMs?: number;
+    /** Wait only for calls matching this attribution filter. */
+    filter?: CostLedgerFilter;
 }
 /** Append-only storage. `append` must atomically reject stale revisions. */
 interface CostLedgerPersistence {

package/dist/fuzz.js

@@ -4,7 +4,7 @@ import {
 import {
   CostLedger,
   CostReceiptCaptureError
-} from "./chunk-GRCDRKII.js";
+} from "./chunk-WS3NZZQQ.js";
 import "./chunk-VI2UW6B6.js";
 import {
   ValidationError

package/dist/hosted/index.d.ts

@@ -46,14 +46,21 @@ interface CodeSurface {
     /** Human summary of what changed — rendered into the auto-PR body. */
     readonly summary?: string;
 }
+/** Named text components optimized together as one candidate. */
+interface ComponentSurface {
+    readonly kind: 'components';
+    readonly components: Readonly<Record<string, string>>;
+}
 /** The mutable surface a proposer changes. Tiers (see
  *  `docs/design/loop-taxonomy.md`):
  *   - `string`      — tiers 1-2: system-prompt addendum / serialized tool
  *                     config. Cheap, reversible, text-diffable.
+ *   - `ComponentSurface` — named prompts, policies, and tool descriptions
+ *                          optimized together.
  *   - `CodeSurface` — tier 4: an implementation change behind a worktree ref.
  *  Tier 3 (knowledge) is owned by agent-knowledge and rides its own adapter,
  *  not this type. */
-type MutableSurface = string | CodeSurface;
+type MutableSurface = string | ComponentSurface | CodeSurface;
 /** Five-valued verdict taxonomy (MOSS-paper alignment). */
 type GateDecision = 'ship' | 'hold' | 'need_more_work' | 'model_ceiling' | 'arch_ceiling';