@tangle-network/agent-eval 0.118.3 → 0.119.0
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 +21 -0
- package/README.md +26 -0
- package/dist/analyst/index.d.ts +143 -21
- package/dist/analyst/index.js +44 -10
- package/dist/analyst/index.js.map +1 -1
- package/dist/benchmarks/index.d.ts +50 -40
- package/dist/benchmarks/index.js +6 -6
- package/dist/campaign/index.d.ts +231 -156
- package/dist/campaign/index.js +5 -5
- package/dist/{chunk-JTMFT5QZ.js → chunk-4I2E3LLO.js} +2 -2
- package/dist/{chunk-VNLWDTDA.js → chunk-7A4LIMMY.js} +845 -252
- package/dist/chunk-7A4LIMMY.js.map +1 -0
- package/dist/{chunk-KK2O4VWA.js → chunk-D7AEXSM5.js} +2 -2
- package/dist/{chunk-EKHHBKS6.js → chunk-F6YUH3L4.js} +39 -28
- package/dist/{chunk-EKHHBKS6.js.map → chunk-F6YUH3L4.js.map} +1 -1
- package/dist/{chunk-267UCQWI.js → chunk-GERDAIAL.js} +2 -2
- package/dist/{chunk-267UCQWI.js.map → chunk-GERDAIAL.js.map} +1 -1
- package/dist/{chunk-MLHRFWQK.js → chunk-JHOJHHU7.js} +85 -16
- package/dist/chunk-JHOJHHU7.js.map +1 -0
- package/dist/{chunk-F5A4GDQW.js → chunk-LMJ2TGWJ.js} +3 -15
- package/dist/chunk-LMJ2TGWJ.js.map +1 -0
- package/dist/{chunk-EGNRE7VA.js → chunk-PAHNGS65.js} +2 -2
- package/dist/{chunk-QWRLW4CT.js → chunk-RLCJQ6VZ.js} +5 -5
- package/dist/{chunk-7V3QDWNL.js → chunk-XJ7JVCHB.js} +2 -2
- package/dist/cli.js +2 -2
- package/dist/contract/index.d.ts +257 -160
- package/dist/contract/index.js +6 -6
- package/dist/fuzz.d.ts +19 -1
- package/dist/fuzz.js +1 -1
- package/dist/index.d.ts +154 -48
- package/dist/index.js +12 -10
- package/dist/index.js.map +1 -1
- package/dist/openapi.json +1 -1
- package/dist/primeintellect/index.d.ts +311 -0
- package/dist/primeintellect/index.js +348 -0
- package/dist/primeintellect/index.js.map +1 -0
- package/dist/rl.d.ts +4 -0
- package/dist/{run-campaign-DWC67KJP.js → run-campaign-OBXSN5WK.js} +3 -3
- package/dist/wire/index.d.ts +19 -1
- package/dist/wire/index.js +2 -2
- package/package.json +6 -1
- package/dist/chunk-F5A4GDQW.js.map +0 -1
- package/dist/chunk-MLHRFWQK.js.map +0 -1
- package/dist/chunk-VNLWDTDA.js.map +0 -1
- /package/dist/{chunk-JTMFT5QZ.js.map → chunk-4I2E3LLO.js.map} +0 -0
- /package/dist/{chunk-KK2O4VWA.js.map → chunk-D7AEXSM5.js.map} +0 -0
- /package/dist/{chunk-EGNRE7VA.js.map → chunk-PAHNGS65.js.map} +0 -0
- /package/dist/{chunk-QWRLW4CT.js.map → chunk-RLCJQ6VZ.js.map} +0 -0
- /package/dist/{chunk-7V3QDWNL.js.map → chunk-XJ7JVCHB.js.map} +0 -0
- /package/dist/{run-campaign-DWC67KJP.js.map → run-campaign-OBXSN5WK.js.map} +0 -0
|
@@ -15,7 +15,7 @@ import {
|
|
|
15
15
|
runImprovementLoop,
|
|
16
16
|
surfaceContentHash,
|
|
17
17
|
surfaceHash
|
|
18
|
-
} from "./chunk-
|
|
18
|
+
} from "./chunk-RLCJQ6VZ.js";
|
|
19
19
|
import {
|
|
20
20
|
SearchLedgerConflictError,
|
|
21
21
|
SearchLedgerError,
|
|
@@ -30,7 +30,7 @@ import {
|
|
|
30
30
|
runCampaign,
|
|
31
31
|
summarizeBackendIntegrity,
|
|
32
32
|
withSearchLedgerFileLock
|
|
33
|
-
} from "./chunk-
|
|
33
|
+
} from "./chunk-PAHNGS65.js";
|
|
34
34
|
import {
|
|
35
35
|
Mutex
|
|
36
36
|
} from "./chunk-3YYRZDON.js";
|
|
@@ -48,7 +48,7 @@ import {
|
|
|
48
48
|
makePolicyEditCandidateRecord,
|
|
49
49
|
policyEditsFromFindings,
|
|
50
50
|
validatePolicyEditCandidateRecord
|
|
51
|
-
} from "./chunk-
|
|
51
|
+
} from "./chunk-7A4LIMMY.js";
|
|
52
52
|
import {
|
|
53
53
|
callLlm,
|
|
54
54
|
callLlmJson,
|
|
@@ -67,7 +67,7 @@ import {
|
|
|
67
67
|
import {
|
|
68
68
|
CostAccountingIncompleteError,
|
|
69
69
|
CostLedger
|
|
70
|
-
} from "./chunk-
|
|
70
|
+
} from "./chunk-JHOJHHU7.js";
|
|
71
71
|
import {
|
|
72
72
|
analyzeTraces
|
|
73
73
|
} from "./chunk-FIUFRXP3.js";
|
|
@@ -4912,29 +4912,34 @@ function analysisEditProposer(opts) {
|
|
|
4912
4912
|
const tracePath = join4(dir, "traces.jsonl");
|
|
4913
4913
|
writeFileSync2(tracePath, traces.endsWith("\n") ? traces : `${traces}
|
|
4914
4914
|
`);
|
|
4915
|
-
|
|
4916
|
-
|
|
4917
|
-
|
|
4918
|
-
|
|
4919
|
-
|
|
4920
|
-
|
|
4921
|
-
|
|
4922
|
-
|
|
4923
|
-
actor: `${opts.kind}.analyze`,
|
|
4924
|
-
model: opts.analysisModel,
|
|
4925
|
-
maximumCharge: opts.analysisMaximumCharge,
|
|
4926
|
-
tags: { generation: String(ctx.generation) },
|
|
4927
|
-
signal: ctx.signal,
|
|
4928
|
-
execute: (signal) => opts.analyze(tracePath, { ...ctx, signal }),
|
|
4929
|
-
receipt: (report2) => opts.analysisReceipt?.(report2) ?? {
|
|
4930
|
-
model: opts.analysisModel,
|
|
4931
|
-
inputTokens: 0,
|
|
4932
|
-
outputTokens: 0,
|
|
4933
|
-
costUnknown: true
|
|
4915
|
+
let report;
|
|
4916
|
+
if (opts.analysisAlreadyMetered) {
|
|
4917
|
+
report = await opts.analyze(tracePath, { ...ctx, costLedger, costPhase: phase });
|
|
4918
|
+
} else {
|
|
4919
|
+
if (costLedger.costCeilingUsd !== void 0 && !opts.analysisReceipt) {
|
|
4920
|
+
throw new CostAccountingIncompleteError(
|
|
4921
|
+
`${opts.kind}: capped analysis requires analysisReceipt before external execution`
|
|
4922
|
+
);
|
|
4934
4923
|
}
|
|
4935
|
-
|
|
4936
|
-
|
|
4937
|
-
|
|
4924
|
+
const analysis = await costLedger.runPaidCall({
|
|
4925
|
+
channel: "analyst",
|
|
4926
|
+
phase,
|
|
4927
|
+
actor: `${opts.kind}.analyze`,
|
|
4928
|
+
model: opts.analysisModel,
|
|
4929
|
+
maximumCharge: opts.analysisMaximumCharge,
|
|
4930
|
+
tags: { generation: String(ctx.generation) },
|
|
4931
|
+
signal: ctx.signal,
|
|
4932
|
+
execute: (signal) => opts.analyze(tracePath, { ...ctx, signal, costLedger, costPhase: phase }),
|
|
4933
|
+
receipt: (value) => opts.analysisReceipt?.(value) ?? {
|
|
4934
|
+
model: opts.analysisModel,
|
|
4935
|
+
inputTokens: 0,
|
|
4936
|
+
outputTokens: 0,
|
|
4937
|
+
costUnknown: true
|
|
4938
|
+
}
|
|
4939
|
+
});
|
|
4940
|
+
if (!analysis.succeeded) throw analysis.error;
|
|
4941
|
+
report = analysis.value;
|
|
4942
|
+
}
|
|
4938
4943
|
const request = {
|
|
4939
4944
|
model: opts.applyModel,
|
|
4940
4945
|
messages: [
|
|
@@ -6316,7 +6321,12 @@ function traceAnalystProposer(opts) {
|
|
|
6316
6321
|
const result = await registry.run(
|
|
6317
6322
|
`trace-analyst-gen-${c.generation}`,
|
|
6318
6323
|
{ traceStore: new OtlpFileTraceStore({ path }) },
|
|
6319
|
-
{
|
|
6324
|
+
{
|
|
6325
|
+
signal: c.signal,
|
|
6326
|
+
chainFindings: true,
|
|
6327
|
+
costLedger: c.costLedger,
|
|
6328
|
+
costPhase: c.costPhase
|
|
6329
|
+
}
|
|
6320
6330
|
);
|
|
6321
6331
|
return result.findings;
|
|
6322
6332
|
});
|
|
@@ -6330,6 +6340,7 @@ function traceAnalystProposer(opts) {
|
|
|
6330
6340
|
costLedger: opts.costLedger,
|
|
6331
6341
|
analysisMaximumCharge: opts.analysisMaximumCharge,
|
|
6332
6342
|
analysisReceipt: opts.analysisReceipt,
|
|
6343
|
+
analysisAlreadyMetered: opts.analyze === void 0,
|
|
6333
6344
|
applyMaxTokens: opts.applyMaxTokens,
|
|
6334
6345
|
fetchImpl: opts.fetchImpl,
|
|
6335
6346
|
resolveTraces: opts.resolveTraces,
|
|
@@ -8098,4 +8109,4 @@ export {
|
|
|
8098
8109
|
verifyCodeSurface,
|
|
8099
8110
|
resolveWorktreePath
|
|
8100
8111
|
};
|
|
8101
|
-
//# sourceMappingURL=chunk-
|
|
8112
|
+
//# sourceMappingURL=chunk-F6YUH3L4.js.map
|