@opensoftware/scorekit 0.4.0 → 0.5.0
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/README.md +2 -1
- package/dist/cli.js +2 -1
- package/dist/cli.js.map +1 -1
- package/dist/commands/score.js +2 -1
- package/dist/commands/score.js.map +1 -1
- package/package.json +4 -4
package/README.md
CHANGED
|
@@ -36,7 +36,7 @@ It appends two lines to `$OUTPUT/scores.jsonl`: one score per run in the format
|
|
|
36
36
|
|
|
37
37
|
| Command | What it does |
|
|
38
38
|
| --- | --- |
|
|
39
|
-
| `scorekit score --scorer <name or file> <runs>...` | Judge each sampled run that has no score of the scorer yet in `--out` (`$OUTPUT/scores.jsonl`) and append its score. `--seed <n>` (0) and `--sample <0..1>` change the sampling. |
|
|
39
|
+
| `scorekit score --scorer <name or file> <runs>...` | Judge each sampled run that has no score of the scorer yet in `--out` (`$OUTPUT/scores.jsonl`) and append its score. `--seed <n>` (0) and `--sample <0..1>` change the sampling; `--model <name>` sends the gateway this model in place of the scorer's. |
|
|
40
40
|
| `scorekit calibrate --scorer <name or file> --labels <file> <runs>...` | Judge every labeled run and write the agreement with the labels to `--out` (`$OUTPUT/calibration.json`). `--min-agreement <0..1>` exits 1 below it. |
|
|
41
41
|
| `scorekit report <scores>...` | Write each scorer's runs, passes, fails, pass rate, cost, pass rate per day and scored runs to `--out` (`$OUTPUT/report.json`); `--markdown <file>` writes the same as Markdown tables. |
|
|
42
42
|
| `scorekit push <scores>...` | Send the scores to Langfuse's score API (`POST /api/public/scores`). |
|
|
@@ -225,3 +225,4 @@ with a trusted publisher for `publish-scorekit.yml`. No credential is stored in
|
|
|
225
225
|
| SCN-SCORE-043 | implemented | Push finds an evaluator on the second page of Langfuse's list |
|
|
226
226
|
| SCN-SCORE-044 | implemented | An evaluator Langfuse refuses is named and fails the command |
|
|
227
227
|
| SCN-SCORE-045 | implemented | Push without Langfuse keys names the missing variables |
|
|
228
|
+
| SCN-SCORE-046 | implemented | The option "--model" overrides the scorer's judge model |
|
package/dist/cli.js
CHANGED
|
@@ -11,7 +11,7 @@ import { scorers } from "./commands/scorers.js";
|
|
|
11
11
|
const HELP = `Usage: scorekit <command> [options]
|
|
12
12
|
|
|
13
13
|
Commands:
|
|
14
|
-
score --scorer <name|file> [--seed <n>] [--sample <0..1>] [--out <file>] <runs file or folder>... Ask the scorer's judge about each sampled run without a score of the scorer and append the scores to --out ($OUTPUT/scores.jsonl).
|
|
14
|
+
score --scorer <name|file> [--seed <n>] [--sample <0..1>] [--model <name>] [--out <file>] <runs file or folder>... Ask the scorer's judge about each sampled run without a score of the scorer and append the scores to --out ($OUTPUT/scores.jsonl).
|
|
15
15
|
calibrate --scorer <name|file> --labels <file> [--min-agreement <0..1>] [--out <file>] <runs file or folder>... Judge every labeled run and write the agreement with the labels to --out ($OUTPUT/calibration.json). Exits 1 below --min-agreement.
|
|
16
16
|
report [--out <file>] [--markdown <file>] <scores file>... Write each scorer's pass rate, cost, pass rate per day and scored runs to --out ($OUTPUT/report.json), and as Markdown to --markdown.
|
|
17
17
|
push <scores file>... Send the scores to Langfuse's score API.
|
|
@@ -46,6 +46,7 @@ const commands = {
|
|
|
46
46
|
scorer,
|
|
47
47
|
seed: values.seed ?? "0",
|
|
48
48
|
sample: values.sample,
|
|
49
|
+
model: values.model,
|
|
49
50
|
out: output("scores.jsonl"),
|
|
50
51
|
now: env.OSF_NOW ?? new Date().toISOString(),
|
|
51
52
|
}, env),
|
package/dist/cli.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"cli.js","sourceRoot":"","sources":["../src/cli.ts"],"names":[],"mappings":";AACA,oFAAoF;AACpF,OAAO,EAAE,SAAS,EAAE,MAAM,WAAW,CAAC;AACtC,OAAO,EAAgB,UAAU,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AACzE,OAAO,EAAE,SAAS,EAAE,MAAM,yBAAyB,CAAC;AACpD,OAAO,EAAE,gBAAgB,EAAE,MAAM,sBAAsB,CAAC;AACxD,OAAO,EAAE,IAAI,EAAE,MAAM,oBAAoB,CAAC;AAC1C,OAAO,EAAE,MAAM,EAAE,MAAM,sBAAsB,CAAC;AAC9C,OAAO,EAAE,KAAK,EAAE,MAAM,qBAAqB,CAAC;AAC5C,OAAO,EAAE,OAAO,EAAE,MAAM,uBAAuB,CAAC;AAEhD,MAAM,IAAI,GAAG;;;;;;;;;;CAUZ,CAAC;AAEF,MAAM,EAAE,WAAW,EAAE,MAAM,EAAE,GAAG,SAAS,CAAC;IACxC,gBAAgB,EAAE,IAAI;IACtB,OAAO,EAAE;QACP,IAAI,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE;QACzB,MAAM,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,QAAQ,EAAE,IAAI,EAAE;QAC1C,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QACxB,MAAM,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QAC1B,GAAG,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QACvB,MAAM,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QAC1B,eAAe,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QACnC,QAAQ,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QAC5B,QAAQ,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QAC5B,KAAK,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QACzB,IAAI,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE;QACzB,MAAM,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE;KAC5B;CACF,CAAC,CAAC;AACH,MAAM,CAAC,OAAO,EAAE,GAAG,KAAK,CAAC,GAAG,WAAW,CAAC;AACxC,MAAM,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC;AACxB,+DAA+D;AAC/D,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;AACrC,MAAM,MAAM,GAAG,CAAC,IAAY,EAAE,EAAE,CAAC,UAAU,CAAC,IAAI,EAAE,MAAM,CAAC,GAAG,CAAC,CAAC;AAE9D,MAAM,QAAQ,GAA4B;IACxC,KAAK,EAAE,GAAG,EAAE,CACV,KAAK,CACH,KAAK,EACL;QACE,MAAM;QACN,IAAI,EAAE,MAAM,CAAC,IAAI,IAAI,GAAG;QACxB,MAAM,EAAE,MAAM,CAAC,MAAM;QACrB,GAAG,EAAE,MAAM,CAAC,cAAc,CAAC;QAC3B,GAAG,EAAE,GAAG,CAAC,OAAO,IAAI,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;KAC7C,EACD,GAAG,CACJ;IACH,SAAS,EAAE,GAAG,EAAE,CACd,SAAS,CACP,KAAK,EACL;QACE,MAAM;QACN,MAAM,EAAE,MAAM,CAAC,MAAM;QACrB,YAAY,EAAE,MAAM,CAAC,eAAe,CAAC;QACrC,GAAG,EAAE,MAAM,CAAC,kBAAkB,CAAC;KAChC,EACD,GAAG,CACJ;IACH,MAAM,EAAE,GAAG,EAAE,CAAC,CAAC,EAAE,KAAK,EAAE,MAAM,CAAC,KAAK,EAAE,EAAE,GAAG,EAAE,MAAM,CAAC,aAAa,CAAC,EAAE,QAAQ,EAAE,MAAM,CAAC,QAAQ,EAAE,CAAC,EAAE,CAAC;IACnG,IAAI,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,KAAK,EAAE,GAAG,CAAC;IAC5B,MAAM,EAAE,GAAG,EAAE,CACX,gBAAgB,CACd;QACE,OAAO,EAAE,MAAM,CAAC,MAAM,IAAI,EAAE;QAC5B,QAAQ,EAAE,MAAM,CAAC,QAAQ;QACzB,KAAK,EAAE,MAAM,CAAC,KAAK;QACnB,IAAI,EAAE,MAAM,CAAC,IAAI,IAAI,KAAK;QAC1B,MAAM,EAAE,MAAM,CAAC,MAAM,IAAI,KAAK;QAC9B,GAAG,EAAE,MAAM,CAAC,iBAAiB,CAAC;KAC/B,EACD,GAAG,CACJ;IACH,OAAO,EAAE,GAAG,EAAE,CAAC,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,EAAE,CAAC;CACtC,CAAC;AACF,MAAM,MAAM,GAAG,OAAO,CAAC,CAAC,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AACvD,OAAO,CAAC,QAAQ,GAAG,MAAM,UAAU,CAAC,EAAE,IAAI,EAAE,UAAU,EAAE,KAAK,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC"}
|
|
1
|
+
{"version":3,"file":"cli.js","sourceRoot":"","sources":["../src/cli.ts"],"names":[],"mappings":";AACA,oFAAoF;AACpF,OAAO,EAAE,SAAS,EAAE,MAAM,WAAW,CAAC;AACtC,OAAO,EAAgB,UAAU,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AACzE,OAAO,EAAE,SAAS,EAAE,MAAM,yBAAyB,CAAC;AACpD,OAAO,EAAE,gBAAgB,EAAE,MAAM,sBAAsB,CAAC;AACxD,OAAO,EAAE,IAAI,EAAE,MAAM,oBAAoB,CAAC;AAC1C,OAAO,EAAE,MAAM,EAAE,MAAM,sBAAsB,CAAC;AAC9C,OAAO,EAAE,KAAK,EAAE,MAAM,qBAAqB,CAAC;AAC5C,OAAO,EAAE,OAAO,EAAE,MAAM,uBAAuB,CAAC;AAEhD,MAAM,IAAI,GAAG;;;;;;;;;;CAUZ,CAAC;AAEF,MAAM,EAAE,WAAW,EAAE,MAAM,EAAE,GAAG,SAAS,CAAC;IACxC,gBAAgB,EAAE,IAAI;IACtB,OAAO,EAAE;QACP,IAAI,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE;QACzB,MAAM,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,QAAQ,EAAE,IAAI,EAAE;QAC1C,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QACxB,MAAM,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QAC1B,GAAG,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QACvB,MAAM,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QAC1B,eAAe,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QACnC,QAAQ,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QAC5B,QAAQ,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QAC5B,KAAK,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QACzB,IAAI,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE;QACzB,MAAM,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE;KAC5B;CACF,CAAC,CAAC;AACH,MAAM,CAAC,OAAO,EAAE,GAAG,KAAK,CAAC,GAAG,WAAW,CAAC;AACxC,MAAM,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC;AACxB,+DAA+D;AAC/D,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;AACrC,MAAM,MAAM,GAAG,CAAC,IAAY,EAAE,EAAE,CAAC,UAAU,CAAC,IAAI,EAAE,MAAM,CAAC,GAAG,CAAC,CAAC;AAE9D,MAAM,QAAQ,GAA4B;IACxC,KAAK,EAAE,GAAG,EAAE,CACV,KAAK,CACH,KAAK,EACL;QACE,MAAM;QACN,IAAI,EAAE,MAAM,CAAC,IAAI,IAAI,GAAG;QACxB,MAAM,EAAE,MAAM,CAAC,MAAM;QACrB,KAAK,EAAE,MAAM,CAAC,KAAK;QACnB,GAAG,EAAE,MAAM,CAAC,cAAc,CAAC;QAC3B,GAAG,EAAE,GAAG,CAAC,OAAO,IAAI,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;KAC7C,EACD,GAAG,CACJ;IACH,SAAS,EAAE,GAAG,EAAE,CACd,SAAS,CACP,KAAK,EACL;QACE,MAAM;QACN,MAAM,EAAE,MAAM,CAAC,MAAM;QACrB,YAAY,EAAE,MAAM,CAAC,eAAe,CAAC;QACrC,GAAG,EAAE,MAAM,CAAC,kBAAkB,CAAC;KAChC,EACD,GAAG,CACJ;IACH,MAAM,EAAE,GAAG,EAAE,CAAC,CAAC,EAAE,KAAK,EAAE,MAAM,CAAC,KAAK,EAAE,EAAE,GAAG,EAAE,MAAM,CAAC,aAAa,CAAC,EAAE,QAAQ,EAAE,MAAM,CAAC,QAAQ,EAAE,CAAC,EAAE,CAAC;IACnG,IAAI,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,KAAK,EAAE,GAAG,CAAC;IAC5B,MAAM,EAAE,GAAG,EAAE,CACX,gBAAgB,CACd;QACE,OAAO,EAAE,MAAM,CAAC,MAAM,IAAI,EAAE;QAC5B,QAAQ,EAAE,MAAM,CAAC,QAAQ;QACzB,KAAK,EAAE,MAAM,CAAC,KAAK;QACnB,IAAI,EAAE,MAAM,CAAC,IAAI,IAAI,KAAK;QAC1B,MAAM,EAAE,MAAM,CAAC,MAAM,IAAI,KAAK;QAC9B,GAAG,EAAE,MAAM,CAAC,iBAAiB,CAAC;KAC/B,EACD,GAAG,CACJ;IACH,OAAO,EAAE,GAAG,EAAE,CAAC,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,EAAE,CAAC;CACtC,CAAC;AACF,MAAM,MAAM,GAAG,OAAO,CAAC,CAAC,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AACvD,OAAO,CAAC,QAAQ,GAAG,MAAM,UAAU,CAAC,EAAE,IAAI,EAAE,UAAU,EAAE,KAAK,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC"}
|
package/dist/commands/score.js
CHANGED
|
@@ -7,7 +7,8 @@ import { isSampled } from "../domain/sample.js";
|
|
|
7
7
|
import { runOf, scoreOf } from "../domain/score.js";
|
|
8
8
|
import { gatewayFrom, judgeRuns } from "./judge-runs.js";
|
|
9
9
|
export async function score(paths, options, env) {
|
|
10
|
-
const
|
|
10
|
+
const read = readScorer(options.scorer);
|
|
11
|
+
const scorer = options.model === undefined ? read : { ...read, model: options.model };
|
|
11
12
|
const rate = options.sample === undefined ? scorer.sample : Number(options.sample);
|
|
12
13
|
if (!(rate >= 0 && rate <= 1))
|
|
13
14
|
throw new UsageError("--sample must be a number from 0 to 1");
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"score.js","sourceRoot":"","sources":["../../src/commands/score.ts"],"names":[],"mappings":"AAAA,mGAAmG;AAEnG,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EAAE,WAAW,EAAE,YAAY,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AACzF,OAAO,EAAE,KAAK,EAAE,MAAM,oBAAoB,CAAC;AAC3C,OAAO,EAAE,SAAS,EAAE,MAAM,mBAAmB,CAAC;AAC9C,OAAO,EAAE,SAAS,EAAE,MAAM,qBAAqB,CAAC;AAChD,OAAO,EAAE,KAAK,EAAE,OAAO,EAAE,MAAM,oBAAoB,CAAC;AACpD,OAAO,EAAE,WAAW,EAAE,SAAS,EAAE,MAAM,iBAAiB,CAAC;
|
|
1
|
+
{"version":3,"file":"score.js","sourceRoot":"","sources":["../../src/commands/score.ts"],"names":[],"mappings":"AAAA,mGAAmG;AAEnG,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EAAE,WAAW,EAAE,YAAY,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AACzF,OAAO,EAAE,KAAK,EAAE,MAAM,oBAAoB,CAAC;AAC3C,OAAO,EAAE,SAAS,EAAE,MAAM,mBAAmB,CAAC;AAC9C,OAAO,EAAE,SAAS,EAAE,MAAM,qBAAqB,CAAC;AAChD,OAAO,EAAE,KAAK,EAAE,OAAO,EAAE,MAAM,oBAAoB,CAAC;AACpD,OAAO,EAAE,WAAW,EAAE,SAAS,EAAE,MAAM,iBAAiB,CAAC;AAYzD,MAAM,CAAC,KAAK,UAAU,KAAK,CAAC,KAAe,EAAE,OAAqB,EAAE,GAAsB;IACxF,MAAM,IAAI,GAAG,UAAU,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;IACxC,MAAM,MAAM,GAAG,OAAO,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,GAAG,IAAI,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE,CAAC;IACtF,MAAM,IAAI,GAAG,OAAO,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;IACnF,IAAI,CAAC,CAAC,IAAI,IAAI,CAAC,IAAI,IAAI,IAAI,CAAC,CAAC;QAAE,MAAM,IAAI,UAAU,CAAC,uCAAuC,CAAC,CAAC;IAC7F,MAAM,IAAI,GAAG,SAAS,CAAC,YAAY,CAAC,KAAK,CAAC,CAAC,CAAC;IAC5C,MAAM,OAAO,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC;IACjC,MAAM,MAAM,GAAG,IAAI,GAAG,CACpB,UAAU,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC;SAC3C,MAAM,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,IAAI,KAAK,MAAM,CAAC,IAAI,CAAC;SACzC,GAAG,CAAC,KAAK,CAAC,CACd,CAAC;IACF,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,SAAS,CAAC,OAAO,CAAC,IAAI,EAAE,MAAM,CAAC,IAAI,EAAE,GAAG,CAAC,EAAE,EAAE,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC;IAC/G,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,IAAI,IAAwB,CAAC;IAC7B,MAAM,MAAM,GAAG,MAAM,SAAS,CAAC,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC,EAAE,GAAG,EAAE,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE,EAAE;QAC7F,WAAW,CAAC,OAAO,CAAC,GAAG,EAAE,OAAO,CAAC,MAAM,EAAE,GAAG,EAAE,OAAO,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,GAAG,EAAE,GAAG,EAAE,OAAO,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC;QAChG,IAAI,OAAO,CAAC,MAAM,KAAK,MAAM;YAAE,MAAM,IAAI,CAAC,CAAC;;YACtC,KAAK,IAAI,CAAC,CAAC;QAChB,IAAI,GAAG,KAAK,SAAS;YAAE,IAAI,GAAG,KAAK,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC;IACzD,CAAC,CAAC,CAAC;IACH,MAAM,OAAO,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC;IAChE,MAAM,KAAK,GAAG;QACZ,UAAU,MAAM,GAAG,KAAK,OAAO,IAAI,CAAC,MAAM,cAAc,MAAM,CAAC,IAAI,KAAK,MAAM,UAAU,KAAK,OAAO;YAClG,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,OAAO,iBAAiB,CAAC,CAAC,CAAC,EAAE,CAAC;YAC9C,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,UAAU,IAAI,MAAM,CAAC;KACnD,CAAC;IACF,OAAO,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC;AAC3B,CAAC"}
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@opensoftware/scorekit",
|
|
3
|
-
"version": "0.
|
|
3
|
+
"version": "0.5.0",
|
|
4
4
|
"description": "scorekit: grade agent sessions on concrete failure modes with LLM judges, calibrated against human labels.",
|
|
5
5
|
"license": "Apache-2.0",
|
|
6
6
|
"repository": {
|
|
@@ -20,12 +20,12 @@
|
|
|
20
20
|
"access": "public"
|
|
21
21
|
},
|
|
22
22
|
"devDependencies": {
|
|
23
|
-
"@opensoftware/testkit": "0.
|
|
23
|
+
"@opensoftware/testkit": "0.27.0"
|
|
24
24
|
},
|
|
25
25
|
"dependencies": {
|
|
26
26
|
"yaml": "2.8.1",
|
|
27
|
-
"@opensoftware/
|
|
28
|
-
"@opensoftware/
|
|
27
|
+
"@opensoftware/contracts": "1.10.0",
|
|
28
|
+
"@opensoftware/cli": "0.1.0"
|
|
29
29
|
},
|
|
30
30
|
"scripts": {
|
|
31
31
|
"build": "tsc -p tsconfig.build.json",
|