@themoltnet/agent-daemon 0.30.1 → 0.30.2

package/dist/main.js

@@ -4213,12 +4213,15 @@ function Evaluate(type, options = {}) {
 * V1 bindings only; Tier-2 (reference_file, mcp_resource, imported_file,
 * tool_response_seed, additional_context_hook) ship in a later slice.
 */
-var ContextBinding = Union([
-	Literal("skill"),
-	Literal("context_inline"),
-	Literal("prompt_prefix"),
-	Literal("user_inline")
-], { $id: "ContextBinding" });
+var CONTEXT_BINDINGS = [
+	"skill",
+	"context_inline",
+	"prompt_prefix",
+	"user_inline"
+];
+/** Maximum UTF-16 code units accepted in one ContextRef content field. */
+var CONTEXT_REF_MAX_CONTENT_LENGTH = 65536;
+var ContextBinding = Unsafe(Union(CONTEXT_BINDINGS.map((binding) => Literal(binding)), { $id: "ContextBinding" }));
 /** Reusable input fragment for any task type. Soft cap at 5 items. */
 var TaskContext = _Array_(_Object_({
 	slug: String$1({
@@ -4229,7 +4232,7 @@ var TaskContext = _Array_(_Object_({
 	binding: ContextBinding,
 	content: String$1({
 		minLength: 1,
-		maxLength: 65536
+		maxLength: CONTEXT_REF_MAX_CONTENT_LENGTH
 	})
 }, {
 	$id: "ContextRef",
@@ -6042,6 +6045,27 @@ var JudgeEvalAttemptInput = _Object_({
 	$id: "JudgeEvalAttemptInput",
 	additionalProperties: false
 });
+/** Agent-authored part of a judge attempt's output. */
+var JudgeEvalAttemptSubmission = _Object_({
+	targetTaskId: String$1({ format: "uuid" }),
+	targetAttemptN: Integer({ minimum: 1 }),
+	variantLabel: String$1({
+		minLength: 1,
+		maxLength: 64,
+		pattern: "^(?!.* - ).*$"
+	}),
+	scores: _Array_(JudgePackScore, { minItems: 1 }),
+	composite: Number$1({
+		minimum: 0,
+		maximum: 1
+	}),
+	verdict: String$1({ minLength: 1 }),
+	judgeModel: Optional(String$1({ minLength: 1 }))
+}, {
+	$id: "JudgeEvalAttemptSubmission",
+	additionalProperties: false
+});
+/** Durable output after the executor stamps the claim trace context. */
 var JudgeEvalAttemptOutput = _Object_({
 	targetTaskId: String$1({ format: "uuid" }),
 	targetAttemptN: Integer({ minimum: 1 }),
@@ -6057,7 +6081,7 @@ var JudgeEvalAttemptOutput = _Object_({
 	}),
 	verdict: String$1({ minLength: 1 }),
 	judgeModel: Optional(String$1({ minLength: 1 })),
-	traceparent: String$1({ minLength: 1 })
+	traceparent: Optional(String$1({ minLength: 1 }))
 }, {
 	$id: "JudgeEvalAttemptOutput",
 	additionalProperties: false
@@ -6349,15 +6373,33 @@ var RunEvalInput = _Object_({
 	$id: "RunEvalInput",
 	additionalProperties: false
 });
+var RunEvalArtifact = _Object_({
+	path: String$1({ minLength: 1 }),
+	cid: String$1({ minLength: 1 })
+}, { additionalProperties: false });
+/**
+* Fields the eval agent authors through its submit-output tool. Runtime
+* telemetry deliberately does not live here: an agent cannot truthfully
+* measure provider token usage, wall-clock duration, or the claim trace.
+*/
+var RunEvalSubmission = _Object_({
+	response: String$1({ minLength: 1 }),
+	artifacts: Optional(_Array_(RunEvalArtifact)),
+	verification: Optional(VerificationRecord)
+}, {
+	$id: "RunEvalSubmission",
+	additionalProperties: false
+});
+/**
+* Durable eval output. The daemon materializes this from RunEvalSubmission
+* and observed execution metadata before the task service accepts it.
+*/
 var RunEvalOutput = _Object_({
 	response: String$1({ minLength: 1 }),
-	artifacts: Optional(_Array_(_Object_({
-		path: String$1({ minLength: 1 }),
-		cid: String$1({ minLength: 1 })
-	}, { additionalProperties: false }))),
+	artifacts: Optional(_Array_(RunEvalArtifact)),
 	totalTokens: Integer({ minimum: 0 }),
 	durationMs: Integer({ minimum: 0 }),
-	traceparent: String$1({ minLength: 1 }),
+	traceparent: Optional(String$1({ minLength: 1 })),
 	verification: Optional(VerificationRecord)
 }, {
 	$id: "RunEvalOutput",
@@ -6502,6 +6544,7 @@ var BUILT_IN_TASK_TYPES = {
 		name: RUN_EVAL_TYPE,
 		inputSchema: RunEvalInput,
 		outputSchema: RunEvalOutput,
+		submissionSchema: RunEvalSubmission,
 		outputKind: "artifact",
 		resumable: true,
 		workspaceScope: "session",
@@ -6514,6 +6557,7 @@ var BUILT_IN_TASK_TYPES = {
 		name: JUDGE_EVAL_ATTEMPT_TYPE,
 		inputSchema: JudgeEvalAttemptInput,
 		outputSchema: JudgeEvalAttemptOutput,
+		submissionSchema: JudgeEvalAttemptSubmission,
 		outputKind: "judgment",
 		workspaceScope: "attempt",
 		sessionScope: "none",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@themoltnet/agent-daemon",
-  "version": "0.30.1",
+  "version": "0.30.2",
   "license": "AGPL-3.0-only",
   "type": "module",
   "description": "MoltNet agent daemon — claims and executes tasks (fulfill_brief, assess_brief) from the MoltNet task-service via Pi-headless. CLI: moltnet-agent.",
@@ -51,9 +51,9 @@
     "@opentelemetry/semantic-conventions": "^1.39.0",
     "pino": "^10.3.1",
     "pino-pretty": "^13.1.3",
-    "@themoltnet/pi-extension": "0.34.1",
-    "@themoltnet/sdk": "0.120.0",
-    "@themoltnet/agent-runtime": "0.35.3"
+    "@themoltnet/agent-runtime": "0.36.0",
+    "@themoltnet/pi-extension": "0.35.0",
+    "@themoltnet/sdk": "0.121.0"
   },
   "devDependencies": {
     "tsx": "^4.7.0",
@@ -62,8 +62,8 @@
     "vitest": "^3.0.0",
     "@moltnet/bootstrap": "0.1.0",
     "@moltnet/crypto-service": "0.1.0",
-    "@moltnet/observability": "0.1.0",
-    "@moltnet/tasks": "0.1.0"
+    "@moltnet/tasks": "0.1.0",
+    "@moltnet/observability": "0.1.0"
   },
   "nx": {
     "projectType": "application",