@united-workforce/eval 0.1.7 → 0.2.1

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 ADDED
@@ -0,0 +1,13 @@
1
+ # @united-workforce/eval
2
+
3
+ Evaluation harness for workflow runs. Installs the `uwf-eval` CLI binary.
4
+
5
+ ## Usage
6
+
7
+ ```bash
8
+ uwf-eval run <task-dir> [options]
9
+ ```
10
+
11
+ ### Debugging
12
+
13
+ `uwf-eval run` now cleans up `/tmp` workdirs after each run. Use `--keep-workdir` to retain them for debugging.
@@ -1 +1 @@
1
- f5c52636c23211ac74569ed6f0d9d1fcd247f2f17d9b8800fa47cbe82e05410f
1
+ {"hash":"a185a72834e0bc03535f8eb1205c4670a48cdfe4f836cb10006da5b0dfe8c706","outputs":["cli.d.ts","cli.d.ts.map","cli.js","cli.js.map","commands/diff.d.ts","commands/diff.d.ts.map","commands/diff.js","commands/diff.js.map","commands/format.d.ts","commands/format.d.ts.map","commands/format.js","commands/format.js.map","commands/index.d.ts","commands/index.d.ts.map","commands/index.js","commands/index.js.map","commands/list.d.ts","commands/list.d.ts.map","commands/list.js","commands/list.js.map","commands/read.d.ts","commands/read.d.ts.map","commands/read.js","commands/read.js.map","commands/report.d.ts","commands/report.d.ts.map","commands/report.js","commands/report.js.map","commands/run.d.ts","commands/run.d.ts.map","commands/run.js","commands/run.js.map","commands/types.d.ts","commands/types.d.ts.map","commands/types.js","commands/types.js.map","index.d.ts","index.d.ts.map","index.js","index.js.map","judge/builtin/frontmatter.d.ts","judge/builtin/frontmatter.d.ts.map","judge/builtin/frontmatter.js","judge/builtin/frontmatter.js.map","judge/builtin/hallucination.d.ts","judge/builtin/hallucination.d.ts.map","judge/builtin/hallucination.js","judge/builtin/hallucination.js.map","judge/builtin/index.d.ts","judge/builtin/index.d.ts.map","judge/builtin/index.js","judge/builtin/index.js.map","judge/builtin/read-steps.d.ts","judge/builtin/read-steps.d.ts.map","judge/builtin/read-steps.js","judge/builtin/read-steps.js.map","judge/builtin/token-stats.d.ts","judge/builtin/token-stats.d.ts.map","judge/builtin/token-stats.js","judge/builtin/token-stats.js.map","judge/builtin/types.d.ts","judge/builtin/types.d.ts.map","judge/builtin/types.js","judge/builtin/types.js.map","judge/builtin/upstream.d.ts","judge/builtin/upstream.d.ts.map","judge/builtin/upstream.js","judge/builtin/upstream.js.map","judge/index.d.ts","judge/index.d.ts.map","judge/index.js","judge/index.js.map","judge/types.d.ts","judge/types.d.ts.map","judge/types.js","judge/types.js.map","runner/collect.d.ts","runner/collect.d.ts.map","runner/collect.js","runner/collect.js.map","runner/execute.d.ts","runner/execute.d.ts.map","runner/execute.js","runner/execute.js.map","runner/index.d.ts","runner/index.d.ts.map","runner/index.js","runner/index.js.map","runner/prepare.d.ts","runner/prepare.d.ts.map","runner/prepare.js","runner/prepare.js.map","runner/types.d.ts","runner/types.d.ts.map","runner/types.js","runner/types.js.map","storage/index.d.ts","storage/index.d.ts.map","storage/index.js","storage/index.js.map","storage/schemas.d.ts","storage/schemas.d.ts.map","storage/schemas.js","storage/schemas.js.map","storage/store.d.ts","storage/store.d.ts.map","storage/store.js","storage/store.js.map","storage/types.d.ts","storage/types.d.ts.map","storage/types.js","storage/types.js.map","task/index.d.ts","task/index.d.ts.map","task/index.js","task/index.js.map","task/loader.d.ts","task/loader.d.ts.map","task/loader.js","task/loader.js.map","task/types.d.ts","task/types.d.ts.map","task/types.js","task/types.js.map"]}
@@ -1 +1 @@
1
- {"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../src/commands/run.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAgDzC,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,OAAO,GAAG,IAAI,CAiCzD"}
1
+ {"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../src/commands/run.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AA6DzC,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,OAAO,GAAG,IAAI,CAkCzD"}
@@ -1,40 +1,55 @@
1
+ import { rm } from "node:fs/promises";
1
2
  import { resolve } from "node:path";
3
+ import { createLogger } from "@united-workforce/util";
2
4
  import { collect, execute, getEngineVersion, prepare } from "../runner/index.js";
3
5
  import { createEvalStore } from "../storage/index.js";
4
- async function runOnce(taskDir, agent, model, engineVersion) {
6
+ const log = createLogger({ sink: { kind: "stderr" } });
7
+ const LOG_CLEANUP = "EVC1NP7Q";
8
+ async function runOnce(taskDir, agent, model, engineVersion, keepWorkdir) {
5
9
  const prepared = await prepare(taskDir);
6
10
  const { manifest, workDir } = prepared;
7
- const { threadId } = await execute({
8
- workDir,
9
- workflow: manifest.workflow,
10
- prompt: manifest.prompt,
11
- agent,
12
- maxSteps: manifest.limits.maxSteps,
13
- });
14
- const evalStore = await createEvalStore();
15
- const config = { agent, model, engineVersion };
16
- const collected = await collect({
17
- evalStore,
18
- taskDir: prepared.taskDir,
19
- workDir,
20
- threadId,
21
- manifest,
22
- config,
23
- });
24
- return {
25
- runHash: collected.runHash,
26
- overall: collected.overall,
27
- task: manifest.name,
28
- judges: collected.judges,
29
- };
11
+ try {
12
+ const { threadId } = await execute({
13
+ workDir,
14
+ workflow: manifest.workflow,
15
+ prompt: manifest.prompt,
16
+ agent,
17
+ maxSteps: manifest.limits.maxSteps,
18
+ });
19
+ const evalStore = await createEvalStore();
20
+ const config = { agent, model, engineVersion };
21
+ const collected = await collect({
22
+ evalStore,
23
+ taskDir: prepared.taskDir,
24
+ workDir,
25
+ threadId,
26
+ manifest,
27
+ config,
28
+ });
29
+ return {
30
+ runHash: collected.runHash,
31
+ overall: collected.overall,
32
+ task: manifest.name,
33
+ judges: collected.judges,
34
+ };
35
+ }
36
+ finally {
37
+ if (keepWorkdir) {
38
+ log(LOG_CLEANUP, `keeping workDir: ${workDir}`);
39
+ }
40
+ else {
41
+ await rm(workDir, { recursive: true, force: true });
42
+ }
43
+ }
30
44
  }
31
45
  export function registerRunCommand(program) {
32
46
  program
33
47
  .command("run <task>")
34
48
  .description("Run eval on a task directory or tarball")
35
- .option("--agent <name>", "agent adapter to use", "uwf-hermes")
49
+ .option("--agent <name>", "agent adapter to use", "uwf-builtin")
36
50
  .option("--model <model>", "model override")
37
51
  .option("--count <n>", "number of eval runs", "1")
52
+ .option("--keep-workdir", "keep the temporary workDir after run (for debugging)", false)
38
53
  .action(async (task, opts) => {
39
54
  const taskDir = resolve(task);
40
55
  const agent = opts.agent;
@@ -49,7 +64,7 @@ export function registerRunCommand(program) {
49
64
  try {
50
65
  const results = [];
51
66
  for (let i = 0; i < count; i++) {
52
- results.push(await runOnce(taskDir, agent, model, engineVersion));
67
+ results.push(await runOnce(taskDir, agent, model, engineVersion, opts.keepWorkdir));
53
68
  }
54
69
  const output = count === 1 ? results[0] : results;
55
70
  process.stdout.write(`${JSON.stringify(output)}\n`);
@@ -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,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"}
1
+ {"version":3,"file":"run.js","sourceRoot":"","sources":["../../src/commands/run.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,EAAE,EAAE,MAAM,kBAAkB,CAAC;AACtC,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,YAAY,EAAE,MAAM,wBAAwB,CAAC;AAGtD,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,gBAAgB,EAAE,OAAO,EAAE,MAAM,oBAAoB,CAAC;AAEjF,OAAO,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAC;AAEtD,MAAM,GAAG,GAAG,YAAY,CAAC,EAAE,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE,CAAC,CAAC;AACvD,MAAM,WAAW,GAAG,UAAU,CAAC;AAS/B,KAAK,UAAU,OAAO,CACpB,OAAe,EACf,KAAa,EACb,KAAa,EACb,aAAqB,EACrB,WAAoB;IAEpB,MAAM,QAAQ,GAAG,MAAM,OAAO,CAAC,OAAO,CAAC,CAAC;IACxC,MAAM,EAAE,QAAQ,EAAE,OAAO,EAAE,GAAG,QAAQ,CAAC;IAEvC,IAAI,CAAC;QACH,MAAM,EAAE,QAAQ,EAAE,GAAG,MAAM,OAAO,CAAC;YACjC,OAAO;YACP,QAAQ,EAAE,QAAQ,CAAC,QAAQ;YAC3B,MAAM,EAAE,QAAQ,CAAC,MAAM;YACvB,KAAK;YACL,QAAQ,EAAE,QAAQ,CAAC,MAAM,CAAC,QAAQ;SACnC,CAAC,CAAC;QAEH,MAAM,SAAS,GAAG,MAAM,eAAe,EAAE,CAAC;QAC1C,MAAM,MAAM,GAAkB,EAAE,KAAK,EAAE,KAAK,EAAE,aAAa,EAAE,CAAC;QAC9D,MAAM,SAAS,GAAG,MAAM,OAAO,CAAC;YAC9B,SAAS;YACT,OAAO,EAAE,QAAQ,CAAC,OAAO;YACzB,OAAO;YACP,QAAQ;YACR,QAAQ;YACR,MAAM;SACP,CAAC,CAAC;QAEH,OAAO;YACL,OAAO,EAAE,SAAS,CAAC,OAAO;YAC1B,OAAO,EAAE,SAAS,CAAC,OAAO;YAC1B,IAAI,EAAE,QAAQ,CAAC,IAAI;YACnB,MAAM,EAAE,SAAS,CAAC,MAAM;SACzB,CAAC;IACJ,CAAC;YAAS,CAAC;QACT,IAAI,WAAW,EAAE,CAAC;YAChB,GAAG,CAAC,WAAW,EAAE,oBAAoB,OAAO,EAAE,CAAC,CAAC;QAClD,CAAC;aAAM,CAAC;YACN,MAAM,EAAE,CAAC,OAAO,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;QACtD,CAAC;IACH,CAAC;AACH,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,aAAa,CAAC;SAC/D,MAAM,CAAC,iBAAiB,EAAE,gBAAgB,CAAC;SAC3C,MAAM,CAAC,aAAa,EAAE,qBAAqB,EAAE,GAAG,CAAC;SACjD,MAAM,CAAC,gBAAgB,EAAE,sDAAsD,EAAE,KAAK,CAAC;SACvF,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,EAAE,IAAI,CAAC,WAAW,CAAC,CAAC,CAAC;YACtF,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": "0.1.7",
3
+ "version": "0.2.1",
4
4
  "private": false,
5
5
  "files": [
6
6
  "src",
@@ -22,8 +22,8 @@
22
22
  "@ocas/fs": "^0.4.1",
23
23
  "commander": "^14.0.3",
24
24
  "yaml": "^2.9.0",
25
- "@united-workforce/protocol": "^0.4.0",
26
- "@united-workforce/util": "^0.2.1"
25
+ "@united-workforce/protocol": "^0.5.0",
26
+ "@united-workforce/util": "^0.3.0"
27
27
  },
28
28
  "devDependencies": {
29
29
  "typescript": "^5.8.3"
@@ -1,15 +1,20 @@
1
+ import { rm } from "node:fs/promises";
1
2
  import { resolve } from "node:path";
2
-
3
+ import { createLogger } from "@united-workforce/util";
3
4
  import type { Command } from "commander";
4
5
  import type { RunResult } from "../runner/index.js";
5
6
  import { collect, execute, getEngineVersion, prepare } from "../runner/index.js";
6
7
  import type { EvalRunConfig } from "../storage/index.js";
7
8
  import { createEvalStore } from "../storage/index.js";
8
9
 
10
+ const log = createLogger({ sink: { kind: "stderr" } });
11
+ const LOG_CLEANUP = "EVC1NP7Q";
12
+
9
13
  type RunCliOptions = {
10
14
  agent: string;
11
15
  model: string | undefined;
12
16
  count: string;
17
+ keepWorkdir: boolean;
13
18
  };
14
19
 
15
20
  async function runOnce(
@@ -17,44 +22,54 @@ async function runOnce(
17
22
  agent: string,
18
23
  model: string,
19
24
  engineVersion: string,
25
+ keepWorkdir: boolean,
20
26
  ): Promise<RunResult> {
21
27
  const prepared = await prepare(taskDir);
22
28
  const { manifest, workDir } = prepared;
23
29
 
24
- const { threadId } = await execute({
25
- workDir,
26
- workflow: manifest.workflow,
27
- prompt: manifest.prompt,
28
- agent,
29
- maxSteps: manifest.limits.maxSteps,
30
- });
30
+ try {
31
+ const { threadId } = await execute({
32
+ workDir,
33
+ workflow: manifest.workflow,
34
+ prompt: manifest.prompt,
35
+ agent,
36
+ maxSteps: manifest.limits.maxSteps,
37
+ });
31
38
 
32
- const evalStore = await createEvalStore();
33
- const config: EvalRunConfig = { agent, model, engineVersion };
34
- const collected = await collect({
35
- evalStore,
36
- taskDir: prepared.taskDir,
37
- workDir,
38
- threadId,
39
- manifest,
40
- config,
41
- });
39
+ const evalStore = await createEvalStore();
40
+ const config: EvalRunConfig = { agent, model, engineVersion };
41
+ const collected = await collect({
42
+ evalStore,
43
+ taskDir: prepared.taskDir,
44
+ workDir,
45
+ threadId,
46
+ manifest,
47
+ config,
48
+ });
42
49
 
43
- return {
44
- runHash: collected.runHash,
45
- overall: collected.overall,
46
- task: manifest.name,
47
- judges: collected.judges,
48
- };
50
+ return {
51
+ runHash: collected.runHash,
52
+ overall: collected.overall,
53
+ task: manifest.name,
54
+ judges: collected.judges,
55
+ };
56
+ } finally {
57
+ if (keepWorkdir) {
58
+ log(LOG_CLEANUP, `keeping workDir: ${workDir}`);
59
+ } else {
60
+ await rm(workDir, { recursive: true, force: true });
61
+ }
62
+ }
49
63
  }
50
64
 
51
65
  export function registerRunCommand(program: Command): void {
52
66
  program
53
67
  .command("run <task>")
54
68
  .description("Run eval on a task directory or tarball")
55
- .option("--agent <name>", "agent adapter to use", "uwf-hermes")
69
+ .option("--agent <name>", "agent adapter to use", "uwf-builtin")
56
70
  .option("--model <model>", "model override")
57
71
  .option("--count <n>", "number of eval runs", "1")
72
+ .option("--keep-workdir", "keep the temporary workDir after run (for debugging)", false)
58
73
  .action(async (task: string, opts: RunCliOptions) => {
59
74
  const taskDir = resolve(task);
60
75
  const agent = opts.agent;
@@ -71,7 +86,7 @@ export function registerRunCommand(program: Command): void {
71
86
  try {
72
87
  const results: RunResult[] = [];
73
88
  for (let i = 0; i < count; i++) {
74
- results.push(await runOnce(taskDir, agent, model, engineVersion));
89
+ results.push(await runOnce(taskDir, agent, model, engineVersion, opts.keepWorkdir));
75
90
  }
76
91
  const output = count === 1 ? results[0] : results;
77
92
  process.stdout.write(`${JSON.stringify(output)}\n`);