@united-workforce/eval 0.1.0
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/LICENSE +21 -0
- package/dist/cli.d.ts +3 -0
- package/dist/cli.d.ts.map +1 -0
- package/dist/cli.js +14 -0
- package/dist/cli.js.map +1 -0
- package/dist/commands/diff.d.ts +3 -0
- package/dist/commands/diff.d.ts.map +1 -0
- package/dist/commands/diff.js +36 -0
- package/dist/commands/diff.js.map +1 -0
- package/dist/commands/format.d.ts +11 -0
- package/dist/commands/format.d.ts.map +1 -0
- package/dist/commands/format.js +114 -0
- package/dist/commands/format.js.map +1 -0
- package/dist/commands/index.d.ts +8 -0
- package/dist/commands/index.d.ts.map +1 -0
- package/dist/commands/index.js +7 -0
- package/dist/commands/index.js.map +1 -0
- package/dist/commands/list.d.ts +3 -0
- package/dist/commands/list.d.ts.map +1 -0
- package/dist/commands/list.js +35 -0
- package/dist/commands/list.js.map +1 -0
- package/dist/commands/read.d.ts +10 -0
- package/dist/commands/read.d.ts.map +1 -0
- package/dist/commands/read.js +37 -0
- package/dist/commands/read.js.map +1 -0
- package/dist/commands/report.d.ts +3 -0
- package/dist/commands/report.d.ts.map +1 -0
- package/dist/commands/report.js +30 -0
- package/dist/commands/report.js.map +1 -0
- package/dist/commands/run.d.ts +3 -0
- package/dist/commands/run.d.ts.map +1 -0
- package/dist/commands/run.js +64 -0
- package/dist/commands/run.js.map +1 -0
- package/dist/commands/types.d.ts +9 -0
- package/dist/commands/types.d.ts.map +1 -0
- package/dist/commands/types.js +2 -0
- package/dist/commands/types.js.map +1 -0
- package/dist/index.d.ts +8 -0
- package/dist/index.d.ts.map +1 -0
- package/dist/index.js +6 -0
- package/dist/index.js.map +1 -0
- package/dist/judge/builtin/frontmatter.d.ts +8 -0
- package/dist/judge/builtin/frontmatter.d.ts.map +1 -0
- package/dist/judge/builtin/frontmatter.js +75 -0
- package/dist/judge/builtin/frontmatter.js.map +1 -0
- package/dist/judge/builtin/hallucination.d.ts +10 -0
- package/dist/judge/builtin/hallucination.d.ts.map +1 -0
- package/dist/judge/builtin/hallucination.js +16 -0
- package/dist/judge/builtin/hallucination.js.map +1 -0
- package/dist/judge/builtin/index.d.ts +7 -0
- package/dist/judge/builtin/index.d.ts.map +1 -0
- package/dist/judge/builtin/index.js +6 -0
- package/dist/judge/builtin/index.js.map +1 -0
- package/dist/judge/builtin/read-steps.d.ts +4 -0
- package/dist/judge/builtin/read-steps.d.ts.map +1 -0
- package/dist/judge/builtin/read-steps.js +12 -0
- package/dist/judge/builtin/read-steps.js.map +1 -0
- package/dist/judge/builtin/token-stats.d.ts +8 -0
- package/dist/judge/builtin/token-stats.d.ts.map +1 -0
- package/dist/judge/builtin/token-stats.js +35 -0
- package/dist/judge/builtin/token-stats.js.map +1 -0
- package/dist/judge/builtin/types.d.ts +15 -0
- package/dist/judge/builtin/types.d.ts.map +1 -0
- package/dist/judge/builtin/types.js +2 -0
- package/dist/judge/builtin/types.js.map +1 -0
- package/dist/judge/builtin/upstream.d.ts +10 -0
- package/dist/judge/builtin/upstream.d.ts.map +1 -0
- package/dist/judge/builtin/upstream.js +16 -0
- package/dist/judge/builtin/upstream.js.map +1 -0
- package/dist/judge/index.d.ts +3 -0
- package/dist/judge/index.d.ts.map +1 -0
- package/dist/judge/index.js +2 -0
- package/dist/judge/index.js.map +1 -0
- package/dist/judge/types.d.ts +15 -0
- package/dist/judge/types.d.ts.map +1 -0
- package/dist/judge/types.js +2 -0
- package/dist/judge/types.js.map +1 -0
- package/dist/runner/collect.d.ts +16 -0
- package/dist/runner/collect.d.ts.map +1 -0
- package/dist/runner/collect.js +129 -0
- package/dist/runner/collect.js.map +1 -0
- package/dist/runner/execute.d.ts +9 -0
- package/dist/runner/execute.d.ts.map +1 -0
- package/dist/runner/execute.js +72 -0
- package/dist/runner/execute.js.map +1 -0
- package/dist/runner/index.d.ts +5 -0
- package/dist/runner/index.d.ts.map +1 -0
- package/dist/runner/index.js +4 -0
- package/dist/runner/index.js.map +1 -0
- package/dist/runner/prepare.d.ts +7 -0
- package/dist/runner/prepare.d.ts.map +1 -0
- package/dist/runner/prepare.js +38 -0
- package/dist/runner/prepare.js.map +1 -0
- package/dist/runner/types.d.ts +70 -0
- package/dist/runner/types.d.ts.map +1 -0
- package/dist/runner/types.js +2 -0
- package/dist/runner/types.js.map +1 -0
- package/dist/storage/index.d.ts +4 -0
- package/dist/storage/index.d.ts.map +1 -0
- package/dist/storage/index.js +3 -0
- package/dist/storage/index.js.map +1 -0
- package/dist/storage/schemas.d.ts +7 -0
- package/dist/storage/schemas.d.ts.map +1 -0
- package/dist/storage/schemas.js +118 -0
- package/dist/storage/schemas.js.map +1 -0
- package/dist/storage/store.d.ts +10 -0
- package/dist/storage/store.d.ts.map +1 -0
- package/dist/storage/store.js +36 -0
- package/dist/storage/store.js.map +1 -0
- package/dist/storage/types.d.ts +30 -0
- package/dist/storage/types.d.ts.map +1 -0
- package/dist/storage/types.js +2 -0
- package/dist/storage/types.js.map +1 -0
- package/dist/task/index.d.ts +3 -0
- package/dist/task/index.d.ts.map +1 -0
- package/dist/task/index.js +2 -0
- package/dist/task/index.js.map +1 -0
- package/dist/task/loader.d.ts +6 -0
- package/dist/task/loader.d.ts.map +1 -0
- package/dist/task/loader.js +69 -0
- package/dist/task/loader.js.map +1 -0
- package/dist/task/types.d.ts +27 -0
- package/dist/task/types.d.ts.map +1 -0
- package/dist/task/types.js +2 -0
- package/dist/task/types.js.map +1 -0
- package/package.json +45 -0
- package/src/cli.ts +22 -0
- package/src/commands/diff.ts +38 -0
- package/src/commands/format.ts +148 -0
- package/src/commands/index.ts +7 -0
- package/src/commands/list.ts +43 -0
- package/src/commands/read.ts +41 -0
- package/src/commands/report.ts +32 -0
- package/src/commands/run.ts +84 -0
- package/src/commands/types.ts +9 -0
- package/src/index.ts +34 -0
- package/src/judge/builtin/frontmatter.ts +95 -0
- package/src/judge/builtin/hallucination.ts +17 -0
- package/src/judge/builtin/index.ts +6 -0
- package/src/judge/builtin/read-steps.ts +14 -0
- package/src/judge/builtin/token-stats.ts +53 -0
- package/src/judge/builtin/types.ts +16 -0
- package/src/judge/builtin/upstream.ts +17 -0
- package/src/judge/index.ts +10 -0
- package/src/judge/types.ts +15 -0
- package/src/runner/collect.ts +172 -0
- package/src/runner/execute.ts +87 -0
- package/src/runner/index.ts +15 -0
- package/src/runner/prepare.ts +45 -0
- package/src/runner/types.ts +85 -0
- package/src/storage/index.ts +9 -0
- package/src/storage/schemas.ts +123 -0
- package/src/storage/store.ts +42 -0
- package/src/storage/types.ts +33 -0
- package/src/task/index.ts +2 -0
- package/src/task/loader.ts +74 -0
- package/src/task/types.ts +28 -0
|
@@ -0,0 +1,28 @@
|
|
|
1
|
+
/** Judge entry in task.yaml */
|
|
2
|
+
export type JudgeEntry = {
|
|
3
|
+
name: string;
|
|
4
|
+
weight: number;
|
|
5
|
+
builtin: boolean;
|
|
6
|
+
/** Path to judge entry script (relative to task root). Required for non-builtin judges. */
|
|
7
|
+
entry: string | null;
|
|
8
|
+
/** Path to OCAS schema JSON for judge data. Required for non-builtin judges. */
|
|
9
|
+
schema: string | null;
|
|
10
|
+
};
|
|
11
|
+
|
|
12
|
+
/** Limits for eval execution. */
|
|
13
|
+
export type TaskLimits = {
|
|
14
|
+
maxSteps: number;
|
|
15
|
+
timeoutMinutes: number;
|
|
16
|
+
};
|
|
17
|
+
|
|
18
|
+
/** Parsed task.yaml manifest. */
|
|
19
|
+
export type TaskManifest = {
|
|
20
|
+
name: string;
|
|
21
|
+
description: string;
|
|
22
|
+
/** Workflow name or relative path to .yaml file. */
|
|
23
|
+
workflow: string;
|
|
24
|
+
/** Initial prompt for thread start. */
|
|
25
|
+
prompt: string;
|
|
26
|
+
limits: TaskLimits;
|
|
27
|
+
judges: JudgeEntry[];
|
|
28
|
+
};
|