@ls-stack/agent-eval 0.58.3 → 0.58.4
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/dist/{app-ROCEce9X.mjs → app-sGeXC4AT.mjs} +63 -11
- package/dist/apps/web/dist/assets/index-BXFsxHVc.js +377 -0
- package/dist/apps/web/dist/index.html +1 -1
- package/dist/bin.mjs +1 -1
- package/dist/caseChild.mjs +4 -3
- package/dist/{cli-SP4kEtYL.mjs → cli-Bf5RzM8O.mjs} +7 -6
- package/dist/index.d.mts +62 -52
- package/dist/index.mjs +3 -3
- package/dist/runChild.mjs +5 -4
- package/dist/{runExecution-CFw0MQFs.mjs → runExecution-CLkC-4Z1.mjs} +35 -3
- package/dist/{runOrchestration-CxjiQmof.mjs → runOrchestration-BS-WxTee.mjs} +1 -1
- package/dist/{runner-BlFQyvN2.mjs → runner-Bz5ZPqmm.mjs} +1 -1
- package/dist/{runner-CY3bgsjU.mjs → runner-DW-11txl.mjs} +2 -2
- package/dist/{src-7GbQj1sb.mjs → src-BjMMDm_O.mjs} +2 -2
- package/package.json +1 -1
- package/skills/agent-eval/SKILL.md +2 -2
- package/dist/apps/web/dist/assets/index-PTikBbhf.js +0 -377
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import { Dt as
|
|
1
|
+
import { Dt as getCaseRowCaseKey, Lt as runWithEvalRegistry, Ot as caseDetailSchema, Q as runWithEvalClock, St as runSummarySchema, Tt as buildCaseKey, Y as runInEvalRuntimeScope, _t as dedupeEvalTags, bt as validateTagsFilterExpression, d as loadEvalModule, f as resolveEvalDefaultConfig, g as commitPendingCacheWrites, gt as deriveStatusFromChildStatuses, ht as deriveStatusFromCaseRows, i as isCaseChildMessage, kt as caseRowSchema, m as buildDeclaredColumnDefs, mt as deriveScopedSummaryFromCases, n as resolveRunnableEvalCases, o as stripTerminalControlCodes, t as filterEvalCases, u as runWithModuleIsolation, vt as matchesTagsFilter, xt as runManifestSchema, yt as validateEvalTagName } from "./runExecution-CLkC-4Z1.mjs";
|
|
2
2
|
import { readFile, readdir, rm, writeFile } from "node:fs/promises";
|
|
3
3
|
import { dirname, join } from "node:path";
|
|
4
4
|
import { existsSync } from "node:fs";
|
|
@@ -1,2 +1,2 @@
|
|
|
1
|
-
import { n as initRunner, t as getRunnerInstance } from "./runner-
|
|
1
|
+
import { n as initRunner, t as getRunnerInstance } from "./runner-DW-11txl.mjs";
|
|
2
2
|
export { getRunnerInstance, initRunner };
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
import { n as createRunner } from "./cli-
|
|
2
|
-
import "./src-
|
|
1
|
+
import { n as createRunner } from "./cli-Bf5RzM8O.mjs";
|
|
2
|
+
import "./src-BjMMDm_O.mjs";
|
|
3
3
|
//#region ../../apps/server/src/runner.ts
|
|
4
4
|
let runnerInstance = null;
|
|
5
5
|
function getRunnerInstance() {
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
import {
|
|
2
|
-
import "./cli-
|
|
1
|
+
import { Ft as defineEval$1, K as matchesEvalTags$1 } from "./runExecution-CLkC-4Z1.mjs";
|
|
2
|
+
import "./cli-Bf5RzM8O.mjs";
|
|
3
3
|
//#region src/index.ts
|
|
4
4
|
/** Register an eval definition with typed tag support. */
|
|
5
5
|
function defineEval(definition) {
|
package/package.json
CHANGED
|
@@ -546,8 +546,8 @@ Mental model:
|
|
|
546
546
|
JSON blob refs. Each namespace is capped at 100 entries by default. The runner
|
|
547
547
|
prunes least recently accessed indexed entries after a run finishes and the
|
|
548
548
|
runner stays idle for `cache.pruneIdleDelayMs ?? 5000` milliseconds. Configure
|
|
549
|
-
`cache.
|
|
550
|
-
`
|
|
549
|
+
`cache.maxEntries` as a number for the default cap, or as
|
|
550
|
+
`{ default, namespaces }` for exact namespace-specific caps.
|
|
551
551
|
- Unindexed legacy cache files are ignored by normal lookup/listing. Use
|
|
552
552
|
`agent-evals cache repair` to remove unindexed cache files, stale index rows,
|
|
553
553
|
debug sidecars, and unreferenced blob files.
|