@united-workforce/eval 0.1.0 → 0.1.2
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/commands/run.js +1 -1
- package/dist/commands/run.js.map +1 -1
- package/dist/judge/builtin/frontmatter.d.ts.map +1 -1
- package/dist/judge/builtin/frontmatter.js +9 -0
- package/dist/judge/builtin/frontmatter.js.map +1 -1
- package/package.json +1 -1
- package/src/commands/run.ts +1 -1
- package/src/judge/builtin/frontmatter.ts +10 -0
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"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"frontmatter.d.ts","sourceRoot":"","sources":["../../../src/judge/builtin/frontmatter.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,YAAY,CAAC;
|
|
1
|
+
{"version":3,"file":"frontmatter.d.ts","sourceRoot":"","sources":["../../../src/judge/builtin/frontmatter.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,YAAY,CAAC;AAuErD;;;;GAIG;AACH,wBAAsB,mBAAmB,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAAC,kBAAkB,CAAC,CAuBvF"}
|
|
@@ -26,6 +26,15 @@ function extractFrontmatterYaml(output) {
|
|
|
26
26
|
}
|
|
27
27
|
/** Validate a single step's frontmatter, returning a list of errors (empty = valid). */
|
|
28
28
|
function validateStepFrontmatter(output) {
|
|
29
|
+
// CAS stores the extracted output as a JSON object after the extract pipeline.
|
|
30
|
+
// Accept both: parsed object (from step.output) or raw markdown string.
|
|
31
|
+
if (typeof output === "object" && output !== null && !Array.isArray(output)) {
|
|
32
|
+
const status = output.$status;
|
|
33
|
+
if (typeof status !== "string" || status.trim() === "") {
|
|
34
|
+
return ["$status field is missing or not a non-empty string"];
|
|
35
|
+
}
|
|
36
|
+
return [];
|
|
37
|
+
}
|
|
29
38
|
const yaml = extractFrontmatterYaml(output);
|
|
30
39
|
if (yaml === null) {
|
|
31
40
|
return ["output does not begin with a valid '---' frontmatter block"];
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"frontmatter.js","sourceRoot":"","sources":["../../../src/judge/builtin/frontmatter.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,MAAM,wBAAwB,CAAC;AACtD,OAAO,EAAE,KAAK,IAAI,SAAS,EAAE,MAAM,MAAM,CAAC;AAE1C,OAAO,EAAE,6BAA6B,EAAE,MAAM,wBAAwB,CAAC;AACvE,OAAO,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAGlD,MAAM,GAAG,GAAG,YAAY,CAAC,EAAE,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE,CAAC,CAAC;AAEvD,MAAM,UAAU,GAAG,UAAU,CAAC;AAE9B,MAAM,KAAK,GAAG,KAAK,CAAC;AAQpB;;;;GAIG;AACH,SAAS,sBAAsB,CAAC,MAAe;IAC7C,IAAI,OAAO,MAAM,KAAK,QAAQ,EAAE,CAAC;QAC/B,OAAO,IAAI,CAAC;IACd,CAAC;IACD,IAAI,CAAC,MAAM,CAAC,UAAU,CAAC,GAAG,KAAK,IAAI,CAAC,EAAE,CAAC;QACrC,OAAO,IAAI,CAAC;IACd,CAAC;IACD,MAAM,IAAI,GAAG,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IAC5C,MAAM,UAAU,GAAG,IAAI,CAAC,OAAO,CAAC,KAAK,KAAK,EAAE,CAAC,CAAC;IAC9C,IAAI,UAAU,KAAK,CAAC,CAAC,EAAE,CAAC;QACtB,OAAO,IAAI,CAAC;IACd,CAAC;IACD,OAAO,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,UAAU,CAAC,CAAC;AACnC,CAAC;AAED,wFAAwF;AACxF,SAAS,uBAAuB,CAAC,MAAe;IAC9C,MAAM,IAAI,GAAG,sBAAsB,CAAC,MAAM,CAAC,CAAC;IAC5C,IAAI,IAAI,KAAK,IAAI,EAAE,CAAC;QAClB,OAAO,CAAC,4DAA4D,CAAC,CAAC;IACxE,CAAC;IAED,IAAI,MAAe,CAAC;IACpB,IAAI,CAAC;QACH,MAAM,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC;IAC3B,CAAC;IAAC,OAAO,CAAC,EAAE,CAAC;QACX,MAAM,OAAO,GAAG,CAAC,YAAY,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;QAC3D,OAAO,CAAC,qCAAqC,OAAO,EAAE,CAAC,CAAC;IAC1D,CAAC;IAED,IAAI,OAAO,MAAM,KAAK,QAAQ,IAAI,MAAM,KAAK,IAAI,IAAI,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC;QAC3E,OAAO,CAAC,mCAAmC,CAAC,CAAC;IAC/C,CAAC;IAED,MAAM,MAAM,GAAI,MAAkC,CAAC,OAAO,CAAC;IAC3D,IAAI,OAAO,MAAM,KAAK,QAAQ,IAAI,MAAM,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC;QACvD,OAAO,CAAC,oDAAoD,CAAC,CAAC;IAChE,CAAC;IAED,OAAO,EAAE,CAAC;AACZ,CAAC;AAED;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,mBAAmB,CAAC,QAAgB;IACxD,MAAM,KAAK,GAAG,eAAe,CAAC,QAAQ,CAAC,CAAC;IAExC,MAAM,YAAY,GAAkB,EAAE,CAAC;IACvC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACtC,MAAM,IAAI,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC;QACtB,MAAM,MAAM,GAAG,uBAAuB,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACpD,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACtB,YAAY,CAAC,IAAI,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE,MAAM,EAAE,CAAC,CAAC;QAC/D,CAAC;IACH,CAAC;IAED,MAAM,UAAU,GAAG,KAAK,CAAC,MAAM,CAAC;IAChC,MAAM,UAAU,GAAG,UAAU,GAAG,YAAY,CAAC,MAAM,CAAC;IACpD,MAAM,KAAK,GAAG,UAAU,GAAG,CAAC,CAAC,CAAC,CAAC,UAAU,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC;IAE3D,GAAG,CAAC,UAAU,EAAE,sBAAsB,QAAQ,UAAU,UAAU,IAAI,UAAU,EAAE,CAAC,CAAC;IAEpF,OAAO;QACL,KAAK;QACL,IAAI,EAAE,EAAE,UAAU,EAAE,UAAU,EAAE,YAAY,EAAE;QAC9C,MAAM,EAAE,6BAA6B;KACtC,CAAC;AACJ,CAAC"}
|
|
1
|
+
{"version":3,"file":"frontmatter.js","sourceRoot":"","sources":["../../../src/judge/builtin/frontmatter.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,MAAM,wBAAwB,CAAC;AACtD,OAAO,EAAE,KAAK,IAAI,SAAS,EAAE,MAAM,MAAM,CAAC;AAE1C,OAAO,EAAE,6BAA6B,EAAE,MAAM,wBAAwB,CAAC;AACvE,OAAO,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAGlD,MAAM,GAAG,GAAG,YAAY,CAAC,EAAE,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE,CAAC,CAAC;AAEvD,MAAM,UAAU,GAAG,UAAU,CAAC;AAE9B,MAAM,KAAK,GAAG,KAAK,CAAC;AAQpB;;;;GAIG;AACH,SAAS,sBAAsB,CAAC,MAAe;IAC7C,IAAI,OAAO,MAAM,KAAK,QAAQ,EAAE,CAAC;QAC/B,OAAO,IAAI,CAAC;IACd,CAAC;IACD,IAAI,CAAC,MAAM,CAAC,UAAU,CAAC,GAAG,KAAK,IAAI,CAAC,EAAE,CAAC;QACrC,OAAO,IAAI,CAAC;IACd,CAAC;IACD,MAAM,IAAI,GAAG,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IAC5C,MAAM,UAAU,GAAG,IAAI,CAAC,OAAO,CAAC,KAAK,KAAK,EAAE,CAAC,CAAC;IAC9C,IAAI,UAAU,KAAK,CAAC,CAAC,EAAE,CAAC;QACtB,OAAO,IAAI,CAAC;IACd,CAAC;IACD,OAAO,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,UAAU,CAAC,CAAC;AACnC,CAAC;AAED,wFAAwF;AACxF,SAAS,uBAAuB,CAAC,MAAe;IAC9C,+EAA+E;IAC/E,wEAAwE;IACxE,IAAI,OAAO,MAAM,KAAK,QAAQ,IAAI,MAAM,KAAK,IAAI,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC;QAC5E,MAAM,MAAM,GAAI,MAAkC,CAAC,OAAO,CAAC;QAC3D,IAAI,OAAO,MAAM,KAAK,QAAQ,IAAI,MAAM,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC;YACvD,OAAO,CAAC,oDAAoD,CAAC,CAAC;QAChE,CAAC;QACD,OAAO,EAAE,CAAC;IACZ,CAAC;IAED,MAAM,IAAI,GAAG,sBAAsB,CAAC,MAAM,CAAC,CAAC;IAC5C,IAAI,IAAI,KAAK,IAAI,EAAE,CAAC;QAClB,OAAO,CAAC,4DAA4D,CAAC,CAAC;IACxE,CAAC;IAED,IAAI,MAAe,CAAC;IACpB,IAAI,CAAC;QACH,MAAM,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC;IAC3B,CAAC;IAAC,OAAO,CAAC,EAAE,CAAC;QACX,MAAM,OAAO,GAAG,CAAC,YAAY,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;QAC3D,OAAO,CAAC,qCAAqC,OAAO,EAAE,CAAC,CAAC;IAC1D,CAAC;IAED,IAAI,OAAO,MAAM,KAAK,QAAQ,IAAI,MAAM,KAAK,IAAI,IAAI,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC;QAC3E,OAAO,CAAC,mCAAmC,CAAC,CAAC;IAC/C,CAAC;IAED,MAAM,MAAM,GAAI,MAAkC,CAAC,OAAO,CAAC;IAC3D,IAAI,OAAO,MAAM,KAAK,QAAQ,IAAI,MAAM,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC;QACvD,OAAO,CAAC,oDAAoD,CAAC,CAAC;IAChE,CAAC;IAED,OAAO,EAAE,CAAC;AACZ,CAAC;AAED;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,mBAAmB,CAAC,QAAgB;IACxD,MAAM,KAAK,GAAG,eAAe,CAAC,QAAQ,CAAC,CAAC;IAExC,MAAM,YAAY,GAAkB,EAAE,CAAC;IACvC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACtC,MAAM,IAAI,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC;QACtB,MAAM,MAAM,GAAG,uBAAuB,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACpD,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACtB,YAAY,CAAC,IAAI,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE,MAAM,EAAE,CAAC,CAAC;QAC/D,CAAC;IACH,CAAC;IAED,MAAM,UAAU,GAAG,KAAK,CAAC,MAAM,CAAC;IAChC,MAAM,UAAU,GAAG,UAAU,GAAG,YAAY,CAAC,MAAM,CAAC;IACpD,MAAM,KAAK,GAAG,UAAU,GAAG,CAAC,CAAC,CAAC,CAAC,UAAU,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC;IAE3D,GAAG,CAAC,UAAU,EAAE,sBAAsB,QAAQ,UAAU,UAAU,IAAI,UAAU,EAAE,CAAC,CAAC;IAEpF,OAAO;QACL,KAAK;QACL,IAAI,EAAE,EAAE,UAAU,EAAE,UAAU,EAAE,YAAY,EAAE;QAC9C,MAAM,EAAE,6BAA6B;KACtC,CAAC;AACJ,CAAC"}
|
package/package.json
CHANGED
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) => {
|
|
@@ -39,6 +39,16 @@ function extractFrontmatterYaml(output: unknown): string | null {
|
|
|
39
39
|
|
|
40
40
|
/** Validate a single step's frontmatter, returning a list of errors (empty = valid). */
|
|
41
41
|
function validateStepFrontmatter(output: unknown): string[] {
|
|
42
|
+
// CAS stores the extracted output as a JSON object after the extract pipeline.
|
|
43
|
+
// Accept both: parsed object (from step.output) or raw markdown string.
|
|
44
|
+
if (typeof output === "object" && output !== null && !Array.isArray(output)) {
|
|
45
|
+
const status = (output as Record<string, unknown>).$status;
|
|
46
|
+
if (typeof status !== "string" || status.trim() === "") {
|
|
47
|
+
return ["$status field is missing or not a non-empty string"];
|
|
48
|
+
}
|
|
49
|
+
return [];
|
|
50
|
+
}
|
|
51
|
+
|
|
42
52
|
const yaml = extractFrontmatterYaml(output);
|
|
43
53
|
if (yaml === null) {
|
|
44
54
|
return ["output does not begin with a valid '---' frontmatter block"];
|