@tangle-network/agent-eval 0.173.2 → 0.173.3

package/dist/cli.js

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 import { i as runRolloutReleaseCli } from "./hf-dataset-D8_RNIis.js";
 import { t as createChatClient } from "./chat-client-CkmjYlfB.js";
-import { t as runAnalystBenchmarkCommand } from "./benchmark-command-CS6gVHVq.js";
+import { t as runAnalystBenchmarkCommand } from "./benchmark-command-BY9oscke.js";
 import { a as runRpcBatch, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi } from "./server-D_cjseFN.js";
 import { writeFileSync } from "node:fs";
 //#region src/cli-config.ts

package/dist/openapi.json

@@ -2,7 +2,7 @@
   "openapi": "3.1.0",
   "info": {
     "title": "@tangle-network/agent-eval — wire protocol",
-    "version": "0.173.2",
+    "version": "0.173.3",
     "description": "HTTP and stdio RPC interface to agent-eval. The TypeScript runtime is the source of truth; this spec is the contract that cross-language clients (Python, Rust, Go) generate from.\n\nWire-protocol version: 1.1.0. Bumps on breaking changes to request/response schemas.",
     "contact": {
       "name": "Tangle Network",

package/dist/supervisor-run/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/supervisor-run/source-facts.ts","../../src/supervisor-run/analyze.ts","../../src/supervisor-run/claude-code-reader.ts","../../src/supervisor-run/integrity-types.ts","../../src/supervisor-run/integrity.ts","../../src/supervisor-run/loops-reader.ts","../../src/supervisor-run/render.ts","../../src/supervisor-run/rollout-nodes.ts","../../src/supervisor-run/runtime-reader.ts"],"mappings":";;;UAwBU;EACR;EACA;EACA;EACA;;EAEA;;;;;;EAMA;;UAGe;EACf,QAAQ;;;;;;EAMR;EACA;;;;;;EAMA;;UAmFe;;EAEf;EACA;EACA;EACA;;EAEA,MAAM;;EAEN;;EAEA;EACA;;EAEA;EACA;;UAGe;EACf;EACA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;EACA;EACA,OAAO;;EAEP;;UAGe;;EAEf;EACA;EACA;;EAEA;EACA;EACA;EACA;EACA;;EAEA;EACA;EACA;;EAEA;;EAEA;EACA;;EAEA;;EAEA;EACA;EACA;EACA;EACA,eAAe;EACf,uBAAuB;;UAGR;;EAEf;EACA;;UAGe;;EAEf;;EAEA;EACA;;;;;;UAOe;WACN;WACA,iBAAiB;WACjB,iBAAiB;WACjB,uBAAuB;WACvB,uBAAuB;WACvB;;;;;;;IAOP;IACA;IACA;IACA;;IAEA;;IAEA;;IAEA;IACA;;IAEA;;IAEA;IACA;;IAEA;;WAEO,YAAY,oBAAoB;;WAEhC,wBAAwB;;WAExB;;;;;;;;;WASA;;WAEA;;;;;;WAMA,0BAA0B,SAAS;;;;;;;WAOnC;;WAEA,gBAAgB;;WAEhB,OAAO;WACP;WACA;;;;;;WAMA;WACA;;;;;;;;;;;;;;;;;;KAmDC;iBAiFI,oBAAoB,KAAK,uBAAuB;;;;;;;iBC9WhD,4BACd,KAAK,sBACL,qBACC;;iBAo2Ba,WAAW,eAAe;;;;;;iBA+C1B,qBAAqB,kBAAkB,wBAAwB;;;UCp4B9D;;WAEN;;;;;WAKA;WACA;WACA;WACA;WACA;WACA;WACA;;;;;;iBAmOW,4BACpB,MAAM,0BACL,QAAQ;;iBAiUK,8BAA8B,MAAM,0BAA0B;;;cCnoBjE;KAED;KAEA;KAKA;UAmCK;WACN;WACA;;UAGM;WACN,MAAM;WACN;WACA,UAAU;WACV;WACA;WACA;WACA,mBAAmB;WACnB;WACA,WAAW,SAAS;;UAGd;WACN,eAAe;WACf;WACA;WACA,MAAM;WACN,iBAAiB;;UAGX;;WAEN;;;;;;;;iBC5BK,8BACd,OAAO,uBAAuB,mBAC9B,UAAS,gCACR;;;;;;;iBCDmB,uBAAuB,aAAa;UAUzC;;WAEN;;WAEA;;WAEA;;WAEA;;;;;;;iBAQW,uBACpB,gBACA,OAAM,qBACL,QAAQ;;iBAmKK,yBACd,gBACA,OAAM,qBACL;;;;;;iBAkCmB,qBACpB,gBAAgB,sBAAsB,sBACtC,OAAM,qBACL,QAAQ;UAgBM,kCAAkC;;WAExC;;WAEA;;;;;;WAMA;;;;;;;;iBASW,yBACpB,gBACA,OAAM,4BACL,QAAQ;;;;;iBA0CW,6BACpB,gBACA,OAAM,4BACL,QAAQ;;;;;;iBAgBW,sBACpB,gBACA,OAAM;EAA8B;IACnC,QAAQ;;;;;;;iBA+BW,sBAAsB,eAAe;;;;;;;iBCzY3C,4BAA4B,GAAG;iBAiC/B,4BAA4B,GAAG;iBA+K/B,+BACd,QAAQ,qBACR;;;UC9Le;;WAEN;;WAEA,QAAQ;;WAER;;WAEA;;WAEA;;WAEA;;WAEA;;WAEA;;WAEA;WACA;WACA;WACA;WACA;;WAEA;;;;;;;iBAwBK,0BACd,KAAK,sBACL,OAAM,2BACL;;;UC6Vc;;;;;;WAMN;;;;;;;;;;;;;;;;iBAsDW,yBACpB,gBACA,OAAM,uBACL,QAAQ;;iBAkEK,2BACd,gBACA,OAAM,uBACL;;iBAKmB,0BAA0B,iBAAiB"}
+{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/supervisor-run/source-facts.ts","../../src/supervisor-run/analyze.ts","../../src/supervisor-run/claude-code-reader.ts","../../src/supervisor-run/integrity-types.ts","../../src/supervisor-run/integrity.ts","../../src/supervisor-run/loops-reader.ts","../../src/supervisor-run/render.ts","../../src/supervisor-run/rollout-nodes.ts","../../src/supervisor-run/runtime-reader.ts"],"mappings":";;;UAwBU;EACR;EACA;EACA;EACA;;EAEA;;;;;;EAMA;;UAGe;EACf,QAAQ;;;;;;EAMR;EACA;;;;;;EAMA;;UAmFe;;EAEf;EACA;EACA;EACA;;EAEA,MAAM;;EAEN;;EAEA;EACA;;EAEA;EACA;;UAGe;EACf;EACA;EACA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;EACA;EACA,OAAO;;EAEP;;UAGe;;EAEf;EACA;EACA;;EAEA;EACA;EACA;EACA;EACA;;EAEA;EACA;EACA;;EAEA;;EAEA;EACA;;EAEA;;EAEA;EACA;EACA;EACA;EACA,eAAe;EACf,uBAAuB;;UAGR;;EAEf;EACA;;UAGe;;EAEf;;EAEA;EACA;;;;;;UAOe;WACN;WACA,iBAAiB;WACjB,iBAAiB;WACjB,uBAAuB;WACvB,uBAAuB;WACvB;;;;;;;IAOP;IACA;IACA;IACA;;IAEA;;IAEA;;IAEA;IACA;;IAEA;;IAEA;IACA;;IAEA;;WAEO,YAAY,oBAAoB;;WAEhC,wBAAwB;;WAExB;;;;;;;;;WASA;;WAEA;;;;;;WAMA,0BAA0B,SAAS;;;;;;;WAOnC;;WAEA,gBAAgB;;WAEhB,OAAO;WACP;WACA;;;;;;WAMA;WACA;;;;;;;;;;;;;;;;;;KAmDC;iBAiFI,oBAAoB,KAAK,uBAAuB;;;;;;;iBC9WhD,4BACd,KAAK,sBACL,qBACC;;iBAo2Ba,WAAW,eAAe;;;;;;iBA+C1B,qBAAqB,kBAAkB,wBAAwB;;;UCp4B9D;;WAEN;;;;;WAKA;WACA;WACA;WACA;WACA;WACA;WACA;;;;;;iBAmOW,4BACpB,MAAM,0BACL,QAAQ;;iBAiUK,8BAA8B,MAAM,0BAA0B;;;cCnoBjE;KAED;KAEA;KAKA;UAmCK;WACN;WACA;;UAGM;WACN,MAAM;WACN;WACA,UAAU;WACV;WACA;WACA;WACA,mBAAmB;WACnB;WACA,WAAW,SAAS;;UAGd;WACN,eAAe;WACf;WACA;WACA,MAAM;WACN,iBAAiB;;UAGX;;WAEN;;;;;;;;iBC5BK,8BACd,OAAO,uBAAuB,mBAC9B,UAAS,gCACR;;;;;;;iBCDmB,uBAAuB,aAAa;UAUzC;;WAEN;;WAEA;;WAEA;;WAEA;;;;;;;iBAQW,uBACpB,gBACA,OAAM,qBACL,QAAQ;;iBAmKK,yBACd,gBACA,OAAM,qBACL;;;;;;iBAkCmB,qBACpB,gBAAgB,sBAAsB,sBACtC,OAAM,qBACL,QAAQ;UAgBM,kCAAkC;;WAExC;;WAEA;;;;;;WAMA;;;;;;;;iBASW,yBACpB,gBACA,OAAM,4BACL,QAAQ;;;;;iBA0CW,6BACpB,gBACA,OAAM,4BACL,QAAQ;;;;;;iBAgBW,sBACpB,gBACA,OAAM;EAA8B;IACnC,QAAQ;;;;;;;iBA+BW,sBAAsB,eAAe;;;;;;;iBCzY3C,4BAA4B,GAAG;iBAiC/B,4BAA4B,GAAG;iBA+K/B,+BACd,QAAQ,qBACR;;;UC9Le;;WAEN;;WAEA,QAAQ;;WAER;;WAEA;;WAEA;;WAEA;;WAEA;;WAEA;;WAEA;WACA;WACA;WACA;WACA;;WAEA;;;;;;;iBAwBK,0BACd,KAAK,sBACL,OAAM,2BACL;;;UCiYc;;;;;;WAMN;;;;;;;;;;;;;;;;iBAsDW,yBACpB,gBACA,OAAM,uBACL,QAAQ;;iBAkEK,2BACd,gBACA,OAAM,uBACL;;iBAKmB,0BAA0B,iBAAiB"}

package/dist/supervisor-run/index.js

@@ -1192,10 +1192,11 @@ function renderSupervisorRollupMarkdown(rollup, title = "Round rollup") {
 * Each line is an envelope whose `root` identifies the local tree. A journal
 * can connect a nested tree with `spawned.ownedTreeRoot`. It can also use the
 * spawned child id as the nested root and repeat the spawn as a parentless
-* marker. This reader accepts both forms, removes a duplicate marker, and
-* preserves the other envelopes for the supervisor-run analyzer. Runtime
-* stores profile identity below `identity` and does not emit Eval's role field.
-* This boundary projects those fields without changing Runtime's dialect.
+* marker when no owned tree is recorded. Descendant spawns must occur in the
+* tree their parent owns. This reader removes a duplicate marker and preserves
+* the other envelopes for the supervisor-run analyzer. Runtime stores profile
+* identity below `identity` and does not emit Eval's role field. This boundary
+* projects those fields without changing Runtime's dialect.
 *
 * The run's terminal status is Runtime's own `result.json` `kind` — `winner`,
 * `no-winner`, or whatever a later arm is called — read verbatim. The reader
@@ -1283,14 +1284,15 @@ function parseEnvelopeJournal(text, path) {
 		const id = nonEmptyString(entry.event.id);
 		if (id === null) continue;
 		if (nonEmptyString(entry.event.parent) !== null) {
-			const matches = parentSpawnsById.get(id) ?? [];
-			matches.push(entry);
-			parentSpawnsById.set(id, matches);
 			const ownedTreeRoot = nonEmptyString(entry.event.ownedTreeRoot);
 			if (ownedTreeRoot !== null) {
 				const owners = parentSpawnsByOwnedTreeRoot.get(ownedTreeRoot) ?? [];
 				owners.push(entry);
 				parentSpawnsByOwnedTreeRoot.set(ownedTreeRoot, owners);
+			} else {
+				const matches = parentSpawnsById.get(id) ?? [];
+				matches.push(entry);
+				parentSpawnsById.set(id, matches);
 			}
 		}
 		if (entry.root === id && (entry.event.parent === void 0 || entry.event.parent === null)) {
@@ -1322,6 +1324,22 @@ function parseEnvelopeJournal(text, path) {
 		if (markerDigest !== null && parentDigest !== null && markerDigest !== parentDigest) throw formatError(path, markers[0]?.line ?? 1, `nested tree ${JSON.stringify(nestedRoot)} disagrees with its parent profile digest`);
 		if (parentDigest === null && markerDigest !== null) parentSpawn.event.profileDigest = markerDigest;
 	}
+	const ownedTreeBySupervisorId = /* @__PURE__ */ new Map([[top.root, top.root]]);
+	for (const [nestedRoot, parentSpawn] of nestedParentSpawns) {
+		const supervisorId = nonEmptyString(parentSpawn.event.id);
+		if (supervisorId === null) continue;
+		const priorTree = ownedTreeBySupervisorId.get(supervisorId);
+		if (priorTree !== void 0 && priorTree !== nestedRoot) throw formatError(path, parentSpawn.line, `spawn ${JSON.stringify(supervisorId)} owns both ${JSON.stringify(priorTree)} and ${JSON.stringify(nestedRoot)}`);
+		ownedTreeBySupervisorId.set(supervisorId, nestedRoot);
+	}
+	for (const entry of events) {
+		if (entry.event.kind !== "spawned") continue;
+		const parentId = nonEmptyString(entry.event.parent);
+		if (parentId === null) continue;
+		const parentTree = ownedTreeBySupervisorId.get(parentId);
+		if (parentTree === void 0) throw formatError(path, entry.line, `spawn ${JSON.stringify(entry.event.id)} names parent ${JSON.stringify(parentId)}, which owns no journal tree`);
+		if (entry.root !== parentTree) throw formatError(path, entry.line, `spawn ${JSON.stringify(entry.event.id)} is in tree ${JSON.stringify(entry.root)}, but parent ${JSON.stringify(parentId)} owns tree ${JSON.stringify(parentTree)}`);
+	}
 	const supervisorIds = /* @__PURE__ */ new Set([top.root, ...[...nestedParentSpawns.values()].map((entry) => nonEmptyString(entry.event.id)).filter((id) => id !== null)]);
 	const normalized = events.filter((entry) => !(nestedRoots.has(entry.root) && entry.event.kind === "spawned" && entry.event.id === entry.root && (entry.event.parent === void 0 || entry.event.parent === null))).map((entry) => {
 		const event = { ...entry.event };