@tangle-network/agent-eval 0.173.1 → 0.173.3
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +18 -0
- package/dist/analyst/index.d.ts +2 -2
- package/dist/analyst/index.js +4 -4
- package/dist/{benchmark-command-k-fOGzt8.js → benchmark-command-BY9oscke.js} +7 -7
- package/dist/{benchmark-command-k-fOGzt8.js.map → benchmark-command-BY9oscke.js.map} +1 -1
- package/dist/benchmarks/index.js +3 -3
- package/dist/campaign/index.js +5 -5
- package/dist/{campaign-pxS0wmo4.js → campaign-B3kPMU8S.js} +6 -6
- package/dist/{campaign-pxS0wmo4.js.map → campaign-B3kPMU8S.js.map} +1 -1
- package/dist/{chat-client-Db4bqYfA.js → chat-client-CkmjYlfB.js} +2 -2
- package/dist/{chat-client-Db4bqYfA.js.map → chat-client-CkmjYlfB.js.map} +1 -1
- package/dist/{chat-json-call-C26igCih.js → chat-json-call-B_Xv2oJK.js} +2 -2
- package/dist/{chat-json-call-C26igCih.js.map → chat-json-call-B_Xv2oJK.js.map} +1 -1
- package/dist/cli.js +3 -3
- package/dist/contract/index.js +5 -5
- package/dist/{define-agent-eval-D_i_s69h.js → define-agent-eval-8h3lXXee.js} +3 -3
- package/dist/{define-agent-eval-D_i_s69h.js.map → define-agent-eval-8h3lXXee.js.map} +1 -1
- package/dist/{dspy-rlm-engine-D5byiHn9.js → dspy-rlm-engine-CF0t2ITD.js} +2 -2
- package/dist/{dspy-rlm-engine-D5byiHn9.js.map → dspy-rlm-engine-CF0t2ITD.js.map} +1 -1
- package/dist/{external-optimizer-process-Cq_Pg15r.js → external-optimizer-process-BwITA9Jp.js} +2 -2
- package/dist/{external-optimizer-process-Cq_Pg15r.js.map → external-optimizer-process-BwITA9Jp.js.map} +1 -1
- package/dist/{external-optimizer-subprocess-DgNebftP.js → external-optimizer-subprocess-wBWeoG6A.js} +2 -2
- package/dist/{external-optimizer-subprocess-DgNebftP.js.map → external-optimizer-subprocess-wBWeoG6A.js.map} +1 -1
- package/dist/index.js +8 -8
- package/dist/{llm-client-CxQtdtd6.js → llm-client-CGlSi8sb.js} +2 -1
- package/dist/llm-client-CGlSi8sb.js.map +1 -0
- package/dist/{llm-judge-B2YxbAJb.js → llm-judge-BfqMFo4h.js} +3 -3
- package/dist/{llm-judge-B2YxbAJb.js.map → llm-judge-BfqMFo4h.js.map} +1 -1
- package/dist/openapi.json +1 -1
- package/dist/{produced-state-7VYDwtkk.js → produced-state-D91uDvQw.js} +3 -3
- package/dist/{produced-state-7VYDwtkk.js.map → produced-state-D91uDvQw.js.map} +1 -1
- package/dist/{semantic-concept-judge-Ct3QU7t5.js → semantic-concept-judge-Dok7_35a.js} +4 -4
- package/dist/{semantic-concept-judge-Ct3QU7t5.js.map → semantic-concept-judge-Dok7_35a.js.map} +1 -1
- package/dist/{server-BR6onwZB.js → server-D_cjseFN.js} +2 -2
- package/dist/{server-BR6onwZB.js.map → server-D_cjseFN.js.map} +1 -1
- package/dist/{skillopt-optimization-method-BzdphODy.js → skillopt-optimization-method-DDw3v3gA.js} +5 -5
- package/dist/{skillopt-optimization-method-BzdphODy.js.map → skillopt-optimization-method-DDw3v3gA.js.map} +1 -1
- package/dist/supervisor-run/index.d.ts.map +1 -1
- package/dist/supervisor-run/index.js +43 -18
- package/dist/supervisor-run/index.js.map +1 -1
- package/dist/types-gvRsyJLh.d.ts.map +1 -1
- package/dist/wire/index.js +1 -1
- package/package.json +2 -2
- package/dist/llm-client-CxQtdtd6.js.map +0 -1
|
@@ -1 +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;;;
|
|
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;;;UCiYc;;;;;;WAMN;;;;;;;;;;;;;;;;iBAsDW,yBACpB,gBACA,OAAM,uBACL,QAAQ;;iBAkEK,2BACd,gBACA,OAAM,uBACL;;iBAKmB,0BAA0B,iBAAiB"}
|
|
@@ -1189,13 +1189,14 @@ function renderSupervisorRollupMarkdown(rollup, title = "Round rollup") {
|
|
|
1189
1189
|
* Reader for agent-runtime's file-backed supervision context.
|
|
1190
1190
|
*
|
|
1191
1191
|
* Runtime stores multiple recursive trees in one `spawn-journal.jsonl`.
|
|
1192
|
-
* Each line is an envelope whose `root` identifies the local tree. A
|
|
1193
|
-
*
|
|
1194
|
-
*
|
|
1195
|
-
*
|
|
1196
|
-
*
|
|
1197
|
-
*
|
|
1198
|
-
*
|
|
1192
|
+
* Each line is an envelope whose `root` identifies the local tree. A journal
|
|
1193
|
+
* can connect a nested tree with `spawned.ownedTreeRoot`. It can also use the
|
|
1194
|
+
* spawned child id as the nested root and repeat the spawn as a parentless
|
|
1195
|
+
* marker when no owned tree is recorded. Descendant spawns must occur in the
|
|
1196
|
+
* tree their parent owns. This reader removes a duplicate marker and preserves
|
|
1197
|
+
* the other envelopes for the supervisor-run analyzer. Runtime stores profile
|
|
1198
|
+
* identity below `identity` and does not emit Eval's role field. This boundary
|
|
1199
|
+
* projects those fields without changing Runtime's dialect.
|
|
1199
1200
|
*
|
|
1200
1201
|
* The run's terminal status is Runtime's own `result.json` `kind` — `winner`,
|
|
1201
1202
|
* `no-winner`, or whatever a later arm is called — read verbatim. The reader
|
|
@@ -1276,17 +1277,25 @@ function parseEnvelopeJournal(text, path) {
|
|
|
1276
1277
|
}
|
|
1277
1278
|
if (begins.length === 0) throw formatError(path, 1, "no begin record");
|
|
1278
1279
|
const parentSpawnsById = /* @__PURE__ */ new Map();
|
|
1280
|
+
const parentSpawnsByOwnedTreeRoot = /* @__PURE__ */ new Map();
|
|
1279
1281
|
const rootMarkersByTree = /* @__PURE__ */ new Map();
|
|
1280
1282
|
for (const entry of events) {
|
|
1281
1283
|
if (entry.event.kind !== "spawned") continue;
|
|
1282
1284
|
const id = nonEmptyString(entry.event.id);
|
|
1283
1285
|
if (id === null) continue;
|
|
1284
1286
|
if (nonEmptyString(entry.event.parent) !== null) {
|
|
1285
|
-
const
|
|
1286
|
-
|
|
1287
|
-
|
|
1287
|
+
const ownedTreeRoot = nonEmptyString(entry.event.ownedTreeRoot);
|
|
1288
|
+
if (ownedTreeRoot !== null) {
|
|
1289
|
+
const owners = parentSpawnsByOwnedTreeRoot.get(ownedTreeRoot) ?? [];
|
|
1290
|
+
owners.push(entry);
|
|
1291
|
+
parentSpawnsByOwnedTreeRoot.set(ownedTreeRoot, owners);
|
|
1292
|
+
} else {
|
|
1293
|
+
const matches = parentSpawnsById.get(id) ?? [];
|
|
1294
|
+
matches.push(entry);
|
|
1295
|
+
parentSpawnsById.set(id, matches);
|
|
1296
|
+
}
|
|
1288
1297
|
}
|
|
1289
|
-
if (entry.root === id && entry.event.parent === void 0) {
|
|
1298
|
+
if (entry.root === id && (entry.event.parent === void 0 || entry.event.parent === null)) {
|
|
1290
1299
|
const markers = rootMarkersByTree.get(entry.root) ?? [];
|
|
1291
1300
|
markers.push(entry);
|
|
1292
1301
|
rootMarkersByTree.set(entry.root, markers);
|
|
@@ -1295,7 +1304,7 @@ function parseEnvelopeJournal(text, path) {
|
|
|
1295
1304
|
const nestedRoots = /* @__PURE__ */ new Set();
|
|
1296
1305
|
const nestedParentSpawns = /* @__PURE__ */ new Map();
|
|
1297
1306
|
for (const begin of begins) {
|
|
1298
|
-
const parentSpawns = (parentSpawnsById.get(begin.root) ?? []).filter((entry) => entry.root !== begin.root);
|
|
1307
|
+
const parentSpawns = [.../* @__PURE__ */ new Set([...parentSpawnsByOwnedTreeRoot.get(begin.root) ?? [], ...parentSpawnsById.get(begin.root) ?? []])].filter((entry) => entry.root !== begin.root);
|
|
1299
1308
|
if (parentSpawns.length > 1) throw formatError(path, begin.line, `tree ${JSON.stringify(begin.root)} has ${parentSpawns.length} parent spawns`);
|
|
1300
1309
|
if (parentSpawns.length === 1) {
|
|
1301
1310
|
nestedRoots.add(begin.root);
|
|
@@ -1306,17 +1315,33 @@ function parseEnvelopeJournal(text, path) {
|
|
|
1306
1315
|
if (topRoots.length !== 1) throw formatError(path, topRoots[0]?.line ?? 1, `expected one top-level tree, found ${topRoots.length}`);
|
|
1307
1316
|
const top = topRoots[0];
|
|
1308
1317
|
for (const nestedRoot of nestedRoots) {
|
|
1309
|
-
const
|
|
1310
|
-
if (
|
|
1318
|
+
const markers = rootMarkersByTree.get(nestedRoot) ?? [];
|
|
1319
|
+
if (markers.length > 1) throw formatError(path, begun.get(nestedRoot)?.line ?? 1, `nested tree ${JSON.stringify(nestedRoot)} contains ${markers.length} root markers`);
|
|
1311
1320
|
const parentSpawn = nestedParentSpawns.get(nestedRoot);
|
|
1312
1321
|
if (parentSpawn === void 0) throw formatError(path, begun.get(nestedRoot)?.line ?? 1, `nested tree ${JSON.stringify(nestedRoot)} has no parent spawn`);
|
|
1313
|
-
const markerDigest = profileDigest(
|
|
1322
|
+
const markerDigest = profileDigest(markers[0]?.event ?? {});
|
|
1314
1323
|
const parentDigest = profileDigest(parentSpawn.event);
|
|
1315
|
-
if (markerDigest !== null && parentDigest !== null && markerDigest !== parentDigest) throw formatError(path,
|
|
1324
|
+
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`);
|
|
1316
1325
|
if (parentDigest === null && markerDigest !== null) parentSpawn.event.profileDigest = markerDigest;
|
|
1317
1326
|
}
|
|
1318
|
-
const
|
|
1319
|
-
|
|
1327
|
+
const ownedTreeBySupervisorId = /* @__PURE__ */ new Map([[top.root, top.root]]);
|
|
1328
|
+
for (const [nestedRoot, parentSpawn] of nestedParentSpawns) {
|
|
1329
|
+
const supervisorId = nonEmptyString(parentSpawn.event.id);
|
|
1330
|
+
if (supervisorId === null) continue;
|
|
1331
|
+
const priorTree = ownedTreeBySupervisorId.get(supervisorId);
|
|
1332
|
+
if (priorTree !== void 0 && priorTree !== nestedRoot) throw formatError(path, parentSpawn.line, `spawn ${JSON.stringify(supervisorId)} owns both ${JSON.stringify(priorTree)} and ${JSON.stringify(nestedRoot)}`);
|
|
1333
|
+
ownedTreeBySupervisorId.set(supervisorId, nestedRoot);
|
|
1334
|
+
}
|
|
1335
|
+
for (const entry of events) {
|
|
1336
|
+
if (entry.event.kind !== "spawned") continue;
|
|
1337
|
+
const parentId = nonEmptyString(entry.event.parent);
|
|
1338
|
+
if (parentId === null) continue;
|
|
1339
|
+
const parentTree = ownedTreeBySupervisorId.get(parentId);
|
|
1340
|
+
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`);
|
|
1341
|
+
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)}`);
|
|
1342
|
+
}
|
|
1343
|
+
const supervisorIds = /* @__PURE__ */ new Set([top.root, ...[...nestedParentSpawns.values()].map((entry) => nonEmptyString(entry.event.id)).filter((id) => id !== null)]);
|
|
1344
|
+
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) => {
|
|
1320
1345
|
const event = { ...entry.event };
|
|
1321
1346
|
if (event.kind === "spawned") {
|
|
1322
1347
|
const digest = profileDigest(event);
|