agentv 5.3.0-next.1 → 5.3.1-next.1
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/README.md +22 -20
- package/dist/{artifact-writer-JFNIPMKW.js → artifact-writer-7NBCOAYC.js} +4 -4
- package/dist/{chunk-T32NL3E6.js → chunk-ASIGJIOJ.js} +50 -30
- package/dist/chunk-ASIGJIOJ.js.map +1 -0
- package/dist/{chunk-6ZZCDZPD.js → chunk-ELCJ23K4.js} +12766 -12478
- package/dist/chunk-ELCJ23K4.js.map +1 -0
- package/dist/{chunk-FVR4RQFK.js → chunk-LKGARI3W.js} +333 -24
- package/dist/chunk-LKGARI3W.js.map +1 -0
- package/dist/{chunk-V52ATPTT.js → chunk-LXBI3SPX.js} +142 -14
- package/dist/chunk-LXBI3SPX.js.map +1 -0
- package/dist/{chunk-BHKQHG26.js → chunk-RKE7SSET.js} +2 -2
- package/dist/cli.js +5 -5
- package/dist/dashboard/assets/{index-DNgf3qJ2.js → index-CbEMiJSb.js} +1 -1
- package/dist/dashboard/assets/{index-r_jSJmlw.js → index-DTA6-l7q.js} +3 -3
- package/dist/dashboard/index.html +1 -1
- package/dist/{dist-6Z7U473R.js → dist-NMXMI5SK.js} +17 -5
- package/dist/index.js +5 -5
- package/dist/{interactive-RUY3OCBI.js → interactive-BN527UV3.js} +5 -5
- package/dist/skills/agentv-bench/SKILL.md +14 -13
- package/dist/skills/agentv-bench/agents/analyzer.md +1 -1
- package/dist/skills/agentv-bench/agents/executor.md +1 -1
- package/dist/skills/agentv-bench/references/autoresearch.md +9 -9
- package/dist/skills/agentv-bench/references/environment-adaptation.md +4 -4
- package/dist/skills/agentv-bench/references/eval-yaml-spec.md +30 -47
- package/dist/skills/agentv-bench/references/schemas.md +44 -60
- package/dist/skills/agentv-bench/references/subagent-pipeline.md +20 -18
- package/dist/skills/agentv-eval-migrations/SKILL.md +13 -0
- package/dist/skills/agentv-eval-migrations/references/breaking-changes.md +39 -21
- package/dist/skills/agentv-eval-writer/SKILL.md +75 -46
- package/dist/skills/agentv-eval-writer/references/custom-evaluators.md +10 -5
- package/dist/skills/agentv-eval-writer/references/eval.schema.json +749 -2158
- package/dist/skills/agentv-eval-writer/references/rubric-evaluator.md +1 -1
- package/dist/{ts-eval-loader-DQDYRULE-V3377FOL.js → ts-eval-loader-2RFVZHCT-7CZ3DCDD.js} +8 -4
- package/package.json +1 -1
- package/dist/chunk-6ZZCDZPD.js.map +0 -1
- package/dist/chunk-FVR4RQFK.js.map +0 -1
- package/dist/chunk-T32NL3E6.js.map +0 -1
- package/dist/chunk-V52ATPTT.js.map +0 -1
- /package/dist/{artifact-writer-JFNIPMKW.js.map → artifact-writer-7NBCOAYC.js.map} +0 -0
- /package/dist/{chunk-BHKQHG26.js.map → chunk-RKE7SSET.js.map} +0 -0
- /package/dist/{dist-6Z7U473R.js.map → dist-NMXMI5SK.js.map} +0 -0
- /package/dist/{interactive-RUY3OCBI.js.map → interactive-BN527UV3.js.map} +0 -0
- /package/dist/{ts-eval-loader-DQDYRULE-V3377FOL.js.map → ts-eval-loader-2RFVZHCT-7CZ3DCDD.js.map} +0 -0
|
@@ -18,7 +18,7 @@ Rubrics are defined as `assert` entries with plain strings or `type: llm-rubric`
|
|
|
18
18
|
| `outcome` | string | required* | Criterion being evaluated (*optional if `score_ranges` used) |
|
|
19
19
|
| `operator` | string | - | Optional intent: `correctness` or `contradiction` |
|
|
20
20
|
| `weight` | number | 1.0 | Relative importance |
|
|
21
|
-
| `required` | boolean | true | Failing forces
|
|
21
|
+
| `required` | boolean | true | Failing forces the case status to `fail` (checklist mode) |
|
|
22
22
|
| `min_score` | number | - | Minimum score (0–1) to pass this criterion |
|
|
23
23
|
| `score_ranges` | map or array | - | Score range definitions for analytic scoring |
|
|
24
24
|
|
|
@@ -1,14 +1,18 @@
|
|
|
1
1
|
import { createRequire } from 'node:module'; const require = createRequire(import.meta.url);
|
|
2
2
|
import {
|
|
3
|
+
isTypeScriptEvalConfigFileName,
|
|
3
4
|
loadTsEvalFile,
|
|
4
|
-
loadTsEvalSuite
|
|
5
|
-
|
|
5
|
+
loadTsEvalSuite,
|
|
6
|
+
typeScriptEvalConfigGlob
|
|
7
|
+
} from "./chunk-ELCJ23K4.js";
|
|
6
8
|
import "./chunk-7BGERE6L.js";
|
|
7
9
|
import "./chunk-PEUTJS7B.js";
|
|
8
10
|
import "./chunk-M7BUKBAF.js";
|
|
9
11
|
import "./chunk-QMRVH5ZP.js";
|
|
10
12
|
export {
|
|
13
|
+
isTypeScriptEvalConfigFileName,
|
|
11
14
|
loadTsEvalFile,
|
|
12
|
-
loadTsEvalSuite
|
|
15
|
+
loadTsEvalSuite,
|
|
16
|
+
typeScriptEvalConfigGlob
|
|
13
17
|
};
|
|
14
|
-
//# sourceMappingURL=ts-eval-loader-
|
|
18
|
+
//# sourceMappingURL=ts-eval-loader-2RFVZHCT-7CZ3DCDD.js.map
|