@united-workforce/eval 0.1.1 → 0.1.3
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/dist/cli.js +3 -1
- package/dist/cli.js.map +1 -1
- package/dist/commands/run.js +1 -1
- package/dist/commands/run.js.map +1 -1
- package/package.json +2 -2
- package/src/cli.ts +4 -1
- package/src/commands/run.ts +1 -1
package/dist/cli.js
CHANGED
|
@@ -1,11 +1,13 @@
|
|
|
1
1
|
#!/usr/bin/env node
|
|
2
2
|
import { Command } from "commander";
|
|
3
3
|
import { registerDiffCommand, registerListCommand, registerReportCommand, registerRunCommand, } from "./commands/index.js";
|
|
4
|
+
// eslint-disable-next-line -- dynamic import for version
|
|
5
|
+
const pkg = await import("../package.json", { with: { type: "json" } });
|
|
4
6
|
const program = new Command();
|
|
5
7
|
program
|
|
6
8
|
.name("uwf-eval")
|
|
7
9
|
.description("Evaluate uwf workflow quality with real agents")
|
|
8
|
-
.version(
|
|
10
|
+
.version(pkg.default.version, "-V, --version");
|
|
9
11
|
registerRunCommand(program);
|
|
10
12
|
registerReportCommand(program);
|
|
11
13
|
registerDiffCommand(program);
|
package/dist/cli.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"cli.js","sourceRoot":"","sources":["../src/cli.ts"],"names":[],"mappings":";AACA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EACL,mBAAmB,EACnB,mBAAmB,EACnB,qBAAqB,EACrB,kBAAkB,GACnB,MAAM,qBAAqB,CAAC;AAE7B,MAAM,OAAO,GAAG,IAAI,OAAO,EAAE,CAAC;AAE9B,OAAO;KACJ,IAAI,CAAC,UAAU,CAAC;KAChB,WAAW,CAAC,gDAAgD,CAAC;KAC7D,OAAO,CAAC,OAAO,CAAC,CAAC;
|
|
1
|
+
{"version":3,"file":"cli.js","sourceRoot":"","sources":["../src/cli.ts"],"names":[],"mappings":";AACA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EACL,mBAAmB,EACnB,mBAAmB,EACnB,qBAAqB,EACrB,kBAAkB,GACnB,MAAM,qBAAqB,CAAC;AAE7B,yDAAyD;AACzD,MAAM,GAAG,GAAG,MAAM,MAAM,CAAC,iBAAiB,EAAE,EAAE,IAAI,EAAE,EAAE,IAAI,EAAE,MAAM,EAAE,EAAE,CAAC,CAAC;AAExE,MAAM,OAAO,GAAG,IAAI,OAAO,EAAE,CAAC;AAE9B,OAAO;KACJ,IAAI,CAAC,UAAU,CAAC;KAChB,WAAW,CAAC,gDAAgD,CAAC;KAC7D,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,OAAO,EAAE,eAAe,CAAC,CAAC;AAEjD,kBAAkB,CAAC,OAAO,CAAC,CAAC;AAC5B,qBAAqB,CAAC,OAAO,CAAC,CAAC;AAC/B,mBAAmB,CAAC,OAAO,CAAC,CAAC;AAC7B,mBAAmB,CAAC,OAAO,CAAC,CAAC;AAE7B,OAAO,CAAC,KAAK,EAAE,CAAC"}
|
package/dist/commands/run.js
CHANGED
|
@@ -32,7 +32,7 @@ export function registerRunCommand(program) {
|
|
|
32
32
|
program
|
|
33
33
|
.command("run <task>")
|
|
34
34
|
.description("Run eval on a task directory or tarball")
|
|
35
|
-
.option("--agent <name>", "agent adapter to use", "hermes")
|
|
35
|
+
.option("--agent <name>", "agent adapter to use", "uwf-hermes")
|
|
36
36
|
.option("--model <model>", "model override")
|
|
37
37
|
.option("--count <n>", "number of eval runs", "1")
|
|
38
38
|
.action(async (task, opts) => {
|
package/dist/commands/run.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"run.js","sourceRoot":"","sources":["../../src/commands/run.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAIpC,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,gBAAgB,EAAE,OAAO,EAAE,MAAM,oBAAoB,CAAC;AAEjF,OAAO,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAC;AAQtD,KAAK,UAAU,OAAO,CACpB,OAAe,EACf,KAAa,EACb,KAAa,EACb,aAAqB;IAErB,MAAM,QAAQ,GAAG,MAAM,OAAO,CAAC,OAAO,CAAC,CAAC;IACxC,MAAM,EAAE,QAAQ,EAAE,OAAO,EAAE,GAAG,QAAQ,CAAC;IAEvC,MAAM,EAAE,QAAQ,EAAE,GAAG,MAAM,OAAO,CAAC;QACjC,OAAO;QACP,QAAQ,EAAE,QAAQ,CAAC,QAAQ;QAC3B,MAAM,EAAE,QAAQ,CAAC,MAAM;QACvB,KAAK;QACL,QAAQ,EAAE,QAAQ,CAAC,MAAM,CAAC,QAAQ;KACnC,CAAC,CAAC;IAEH,MAAM,SAAS,GAAG,MAAM,eAAe,EAAE,CAAC;IAC1C,MAAM,MAAM,GAAkB,EAAE,KAAK,EAAE,KAAK,EAAE,aAAa,EAAE,CAAC;IAC9D,MAAM,SAAS,GAAG,MAAM,OAAO,CAAC;QAC9B,SAAS;QACT,OAAO,EAAE,QAAQ,CAAC,OAAO;QACzB,OAAO;QACP,QAAQ;QACR,QAAQ;QACR,MAAM;KACP,CAAC,CAAC;IAEH,OAAO;QACL,OAAO,EAAE,SAAS,CAAC,OAAO;QAC1B,OAAO,EAAE,SAAS,CAAC,OAAO;QAC1B,IAAI,EAAE,QAAQ,CAAC,IAAI;QACnB,MAAM,EAAE,SAAS,CAAC,MAAM;KACzB,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,kBAAkB,CAAC,OAAgB;IACjD,OAAO;SACJ,OAAO,CAAC,YAAY,CAAC;SACrB,WAAW,CAAC,yCAAyC,CAAC;SACtD,MAAM,CAAC,gBAAgB,EAAE,sBAAsB,EAAE,
|
|
1
|
+
{"version":3,"file":"run.js","sourceRoot":"","sources":["../../src/commands/run.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAIpC,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,gBAAgB,EAAE,OAAO,EAAE,MAAM,oBAAoB,CAAC;AAEjF,OAAO,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAC;AAQtD,KAAK,UAAU,OAAO,CACpB,OAAe,EACf,KAAa,EACb,KAAa,EACb,aAAqB;IAErB,MAAM,QAAQ,GAAG,MAAM,OAAO,CAAC,OAAO,CAAC,CAAC;IACxC,MAAM,EAAE,QAAQ,EAAE,OAAO,EAAE,GAAG,QAAQ,CAAC;IAEvC,MAAM,EAAE,QAAQ,EAAE,GAAG,MAAM,OAAO,CAAC;QACjC,OAAO;QACP,QAAQ,EAAE,QAAQ,CAAC,QAAQ;QAC3B,MAAM,EAAE,QAAQ,CAAC,MAAM;QACvB,KAAK;QACL,QAAQ,EAAE,QAAQ,CAAC,MAAM,CAAC,QAAQ;KACnC,CAAC,CAAC;IAEH,MAAM,SAAS,GAAG,MAAM,eAAe,EAAE,CAAC;IAC1C,MAAM,MAAM,GAAkB,EAAE,KAAK,EAAE,KAAK,EAAE,aAAa,EAAE,CAAC;IAC9D,MAAM,SAAS,GAAG,MAAM,OAAO,CAAC;QAC9B,SAAS;QACT,OAAO,EAAE,QAAQ,CAAC,OAAO;QACzB,OAAO;QACP,QAAQ;QACR,QAAQ;QACR,MAAM;KACP,CAAC,CAAC;IAEH,OAAO;QACL,OAAO,EAAE,SAAS,CAAC,OAAO;QAC1B,OAAO,EAAE,SAAS,CAAC,OAAO;QAC1B,IAAI,EAAE,QAAQ,CAAC,IAAI;QACnB,MAAM,EAAE,SAAS,CAAC,MAAM;KACzB,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,kBAAkB,CAAC,OAAgB;IACjD,OAAO;SACJ,OAAO,CAAC,YAAY,CAAC;SACrB,WAAW,CAAC,yCAAyC,CAAC;SACtD,MAAM,CAAC,gBAAgB,EAAE,sBAAsB,EAAE,YAAY,CAAC;SAC9D,MAAM,CAAC,iBAAiB,EAAE,gBAAgB,CAAC;SAC3C,MAAM,CAAC,aAAa,EAAE,qBAAqB,EAAE,GAAG,CAAC;SACjD,MAAM,CAAC,KAAK,EAAE,IAAY,EAAE,IAAmB,EAAE,EAAE;QAClD,MAAM,OAAO,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;QAC9B,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC;QACzB,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,IAAI,EAAE,CAAC;QAC/B,MAAM,KAAK,GAAG,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;QAC9C,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,IAAI,KAAK,GAAG,CAAC,EAAE,CAAC;YAC1C,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,sCAAsC,CAAC,CAAC;YAC7D,OAAO,CAAC,QAAQ,GAAG,CAAC,CAAC;YACrB,OAAO;QACT,CAAC;QAED,MAAM,aAAa,GAAG,gBAAgB,EAAE,CAAC;QAEzC,IAAI,CAAC;YACH,MAAM,OAAO,GAAgB,EAAE,CAAC;YAChC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,EAAE,CAAC,EAAE,EAAE,CAAC;gBAC/B,OAAO,CAAC,IAAI,CAAC,MAAM,OAAO,CAAC,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,aAAa,CAAC,CAAC,CAAC;YACpE,CAAC;YACD,MAAM,MAAM,GAAG,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC;YAClD,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;QACtD,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,MAAM,OAAO,GAAG,CAAC,YAAY,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;YAC3D,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,OAAO,IAAI,CAAC,CAAC;YACrC,OAAO,CAAC,QAAQ,GAAG,CAAC,CAAC;QACvB,CAAC;IACH,CAAC,CAAC,CAAC;AACP,CAAC"}
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@united-workforce/eval",
|
|
3
|
-
"version": "0.1.
|
|
3
|
+
"version": "0.1.3",
|
|
4
4
|
"private": false,
|
|
5
5
|
"files": [
|
|
6
6
|
"src",
|
|
@@ -23,7 +23,7 @@
|
|
|
23
23
|
"commander": "^14.0.3",
|
|
24
24
|
"yaml": "^2.9.0",
|
|
25
25
|
"@united-workforce/protocol": "^0.1.0",
|
|
26
|
-
"@united-workforce/util": "^0.1.
|
|
26
|
+
"@united-workforce/util": "^0.1.1"
|
|
27
27
|
},
|
|
28
28
|
"devDependencies": {
|
|
29
29
|
"typescript": "^5.8.3"
|
package/src/cli.ts
CHANGED
|
@@ -7,12 +7,15 @@ import {
|
|
|
7
7
|
registerRunCommand,
|
|
8
8
|
} from "./commands/index.js";
|
|
9
9
|
|
|
10
|
+
// eslint-disable-next-line -- dynamic import for version
|
|
11
|
+
const pkg = await import("../package.json", { with: { type: "json" } });
|
|
12
|
+
|
|
10
13
|
const program = new Command();
|
|
11
14
|
|
|
12
15
|
program
|
|
13
16
|
.name("uwf-eval")
|
|
14
17
|
.description("Evaluate uwf workflow quality with real agents")
|
|
15
|
-
.version(
|
|
18
|
+
.version(pkg.default.version, "-V, --version");
|
|
16
19
|
|
|
17
20
|
registerRunCommand(program);
|
|
18
21
|
registerReportCommand(program);
|
package/src/commands/run.ts
CHANGED
|
@@ -52,7 +52,7 @@ export function registerRunCommand(program: Command): void {
|
|
|
52
52
|
program
|
|
53
53
|
.command("run <task>")
|
|
54
54
|
.description("Run eval on a task directory or tarball")
|
|
55
|
-
.option("--agent <name>", "agent adapter to use", "hermes")
|
|
55
|
+
.option("--agent <name>", "agent adapter to use", "uwf-hermes")
|
|
56
56
|
.option("--model <model>", "model override")
|
|
57
57
|
.option("--count <n>", "number of eval runs", "1")
|
|
58
58
|
.action(async (task: string, opts: RunCliOptions) => {
|