@vibe-agent-toolkit/agent-skills 0.1.42 → 0.2.0-rc.10

package/dist/skill-test/grading-schema.d.ts

@@ -31,20 +31,51 @@ export declare const GradedExpectationSchema: z.ZodObject<{
     text: z.ZodString;
     passed: z.ZodBoolean;
     evidence: z.ZodOptional<z.ZodString>;
+    /**
+     * Which eval this expectation was graded under. vat stamps it when it
+     * assembles the aggregate from per-eval grader fragments, so that a
+     * `--baseline` run's two artifacts can be lined up PER EVAL rather than only
+     * in aggregate — without it, a reader can compare two totals but cannot say
+     * which eval moved. Optional because a grading.json produced by external
+     * tooling (skill-creator's own) carries no per-eval attribution; the schema
+     * documents the field vat now emits rather than tightening the contract.
+     */
+    evalId: z.ZodOptional<z.ZodString>;
 }, "passthrough", z.ZodTypeAny, z.objectOutputType<{
     text: z.ZodString;
     passed: z.ZodBoolean;
     evidence: z.ZodOptional<z.ZodString>;
+    /**
+     * Which eval this expectation was graded under. vat stamps it when it
+     * assembles the aggregate from per-eval grader fragments, so that a
+     * `--baseline` run's two artifacts can be lined up PER EVAL rather than only
+     * in aggregate — without it, a reader can compare two totals but cannot say
+     * which eval moved. Optional because a grading.json produced by external
+     * tooling (skill-creator's own) carries no per-eval attribution; the schema
+     * documents the field vat now emits rather than tightening the contract.
+     */
+    evalId: z.ZodOptional<z.ZodString>;
 }, z.ZodTypeAny, "passthrough">, z.objectInputType<{
     text: z.ZodString;
     passed: z.ZodBoolean;
     evidence: z.ZodOptional<z.ZodString>;
+    /**
+     * Which eval this expectation was graded under. vat stamps it when it
+     * assembles the aggregate from per-eval grader fragments, so that a
+     * `--baseline` run's two artifacts can be lined up PER EVAL rather than only
+     * in aggregate — without it, a reader can compare two totals but cannot say
+     * which eval moved. Optional because a grading.json produced by external
+     * tooling (skill-creator's own) carries no per-eval attribution; the schema
+     * documents the field vat now emits rather than tightening the contract.
+     */
+    evalId: z.ZodOptional<z.ZodString>;
 }, z.ZodTypeAny, "passthrough">>;
 export type GradedExpectation = z.infer<typeof GradedExpectationSchema>;
 /**
  * Aggregate pass/fail counts. `failed`/`pass_rate` are documented but optional.
  * Counts are non-negative integers (a float or negative count is a grader bug),
- * and `passed` can never exceed `total`.
+ * and `passed` can never exceed `total` — see {@link SUMMARY_SCHEMA_DESCRIPTION}
+ * for why that last one has to be stated twice.
  */
 export declare const GradingSummarySchema: z.ZodEffects<z.ZodObject<{
     passed: z.ZodNumber;
@@ -83,14 +114,44 @@ export declare const GradingReportSchema: z.ZodObject<{
         text: z.ZodString;
         passed: z.ZodBoolean;
         evidence: z.ZodOptional<z.ZodString>;
+        /**
+         * Which eval this expectation was graded under. vat stamps it when it
+         * assembles the aggregate from per-eval grader fragments, so that a
+         * `--baseline` run's two artifacts can be lined up PER EVAL rather than only
+         * in aggregate — without it, a reader can compare two totals but cannot say
+         * which eval moved. Optional because a grading.json produced by external
+         * tooling (skill-creator's own) carries no per-eval attribution; the schema
+         * documents the field vat now emits rather than tightening the contract.
+         */
+        evalId: z.ZodOptional<z.ZodString>;
     }, "passthrough", z.ZodTypeAny, z.objectOutputType<{
         text: z.ZodString;
         passed: z.ZodBoolean;
         evidence: z.ZodOptional<z.ZodString>;
+        /**
+         * Which eval this expectation was graded under. vat stamps it when it
+         * assembles the aggregate from per-eval grader fragments, so that a
+         * `--baseline` run's two artifacts can be lined up PER EVAL rather than only
+         * in aggregate — without it, a reader can compare two totals but cannot say
+         * which eval moved. Optional because a grading.json produced by external
+         * tooling (skill-creator's own) carries no per-eval attribution; the schema
+         * documents the field vat now emits rather than tightening the contract.
+         */
+        evalId: z.ZodOptional<z.ZodString>;
     }, z.ZodTypeAny, "passthrough">, z.objectInputType<{
         text: z.ZodString;
         passed: z.ZodBoolean;
         evidence: z.ZodOptional<z.ZodString>;
+        /**
+         * Which eval this expectation was graded under. vat stamps it when it
+         * assembles the aggregate from per-eval grader fragments, so that a
+         * `--baseline` run's two artifacts can be lined up PER EVAL rather than only
+         * in aggregate — without it, a reader can compare two totals but cannot say
+         * which eval moved. Optional because a grading.json produced by external
+         * tooling (skill-creator's own) carries no per-eval attribution; the schema
+         * documents the field vat now emits rather than tightening the contract.
+         */
+        evalId: z.ZodOptional<z.ZodString>;
     }, z.ZodTypeAny, "passthrough">>, "many">;
     summary: z.ZodEffects<z.ZodObject<{
         passed: z.ZodNumber;
@@ -127,19 +188,58 @@ export declare const GradingReportSchema: z.ZodObject<{
      * by untrusted skill code is rejected.
      */
     runNonce: z.ZodOptional<z.ZodString>;
+    /**
+     * Which `--baseline` arm produced this report: `'with'` (the skill declared)
+     * or `'without'` (the control arm run with the skill withheld). vat writes
+     * the two arms to two files of the SAME shape — `grading.json` and
+     * `baseline.json` — so without this field a reader holding one of them
+     * cannot tell which arm it is looking at. Optional for the same reason
+     * `runNonce` is: an externally produced grading.json has no arm to declare.
+     */
+    arm: z.ZodOptional<z.ZodEnum<["with", "without"]>>;
 }, "passthrough", z.ZodTypeAny, z.objectOutputType<{
     expectations: z.ZodArray<z.ZodObject<{
         text: z.ZodString;
         passed: z.ZodBoolean;
         evidence: z.ZodOptional<z.ZodString>;
+        /**
+         * Which eval this expectation was graded under. vat stamps it when it
+         * assembles the aggregate from per-eval grader fragments, so that a
+         * `--baseline` run's two artifacts can be lined up PER EVAL rather than only
+         * in aggregate — without it, a reader can compare two totals but cannot say
+         * which eval moved. Optional because a grading.json produced by external
+         * tooling (skill-creator's own) carries no per-eval attribution; the schema
+         * documents the field vat now emits rather than tightening the contract.
+         */
+        evalId: z.ZodOptional<z.ZodString>;
     }, "passthrough", z.ZodTypeAny, z.objectOutputType<{
         text: z.ZodString;
         passed: z.ZodBoolean;
         evidence: z.ZodOptional<z.ZodString>;
+        /**
+         * Which eval this expectation was graded under. vat stamps it when it
+         * assembles the aggregate from per-eval grader fragments, so that a
+         * `--baseline` run's two artifacts can be lined up PER EVAL rather than only
+         * in aggregate — without it, a reader can compare two totals but cannot say
+         * which eval moved. Optional because a grading.json produced by external
+         * tooling (skill-creator's own) carries no per-eval attribution; the schema
+         * documents the field vat now emits rather than tightening the contract.
+         */
+        evalId: z.ZodOptional<z.ZodString>;
     }, z.ZodTypeAny, "passthrough">, z.objectInputType<{
         text: z.ZodString;
         passed: z.ZodBoolean;
         evidence: z.ZodOptional<z.ZodString>;
+        /**
+         * Which eval this expectation was graded under. vat stamps it when it
+         * assembles the aggregate from per-eval grader fragments, so that a
+         * `--baseline` run's two artifacts can be lined up PER EVAL rather than only
+         * in aggregate — without it, a reader can compare two totals but cannot say
+         * which eval moved. Optional because a grading.json produced by external
+         * tooling (skill-creator's own) carries no per-eval attribution; the schema
+         * documents the field vat now emits rather than tightening the contract.
+         */
+        evalId: z.ZodOptional<z.ZodString>;
     }, z.ZodTypeAny, "passthrough">>, "many">;
     summary: z.ZodEffects<z.ZodObject<{
         passed: z.ZodNumber;
@@ -176,19 +276,58 @@ export declare const GradingReportSchema: z.ZodObject<{
      * by untrusted skill code is rejected.
      */
     runNonce: z.ZodOptional<z.ZodString>;
+    /**
+     * Which `--baseline` arm produced this report: `'with'` (the skill declared)
+     * or `'without'` (the control arm run with the skill withheld). vat writes
+     * the two arms to two files of the SAME shape — `grading.json` and
+     * `baseline.json` — so without this field a reader holding one of them
+     * cannot tell which arm it is looking at. Optional for the same reason
+     * `runNonce` is: an externally produced grading.json has no arm to declare.
+     */
+    arm: z.ZodOptional<z.ZodEnum<["with", "without"]>>;
 }, z.ZodTypeAny, "passthrough">, z.objectInputType<{
     expectations: z.ZodArray<z.ZodObject<{
         text: z.ZodString;
         passed: z.ZodBoolean;
         evidence: z.ZodOptional<z.ZodString>;
+        /**
+         * Which eval this expectation was graded under. vat stamps it when it
+         * assembles the aggregate from per-eval grader fragments, so that a
+         * `--baseline` run's two artifacts can be lined up PER EVAL rather than only
+         * in aggregate — without it, a reader can compare two totals but cannot say
+         * which eval moved. Optional because a grading.json produced by external
+         * tooling (skill-creator's own) carries no per-eval attribution; the schema
+         * documents the field vat now emits rather than tightening the contract.
+         */
+        evalId: z.ZodOptional<z.ZodString>;
     }, "passthrough", z.ZodTypeAny, z.objectOutputType<{
         text: z.ZodString;
         passed: z.ZodBoolean;
         evidence: z.ZodOptional<z.ZodString>;
+        /**
+         * Which eval this expectation was graded under. vat stamps it when it
+         * assembles the aggregate from per-eval grader fragments, so that a
+         * `--baseline` run's two artifacts can be lined up PER EVAL rather than only
+         * in aggregate — without it, a reader can compare two totals but cannot say
+         * which eval moved. Optional because a grading.json produced by external
+         * tooling (skill-creator's own) carries no per-eval attribution; the schema
+         * documents the field vat now emits rather than tightening the contract.
+         */
+        evalId: z.ZodOptional<z.ZodString>;
     }, z.ZodTypeAny, "passthrough">, z.objectInputType<{
         text: z.ZodString;
         passed: z.ZodBoolean;
         evidence: z.ZodOptional<z.ZodString>;
+        /**
+         * Which eval this expectation was graded under. vat stamps it when it
+         * assembles the aggregate from per-eval grader fragments, so that a
+         * `--baseline` run's two artifacts can be lined up PER EVAL rather than only
+         * in aggregate — without it, a reader can compare two totals but cannot say
+         * which eval moved. Optional because a grading.json produced by external
+         * tooling (skill-creator's own) carries no per-eval attribution; the schema
+         * documents the field vat now emits rather than tightening the contract.
+         */
+        evalId: z.ZodOptional<z.ZodString>;
     }, z.ZodTypeAny, "passthrough">>, "many">;
     summary: z.ZodEffects<z.ZodObject<{
         passed: z.ZodNumber;
@@ -225,13 +364,29 @@ export declare const GradingReportSchema: z.ZodObject<{
      * by untrusted skill code is rejected.
      */
     runNonce: z.ZodOptional<z.ZodString>;
+    /**
+     * Which `--baseline` arm produced this report: `'with'` (the skill declared)
+     * or `'without'` (the control arm run with the skill withheld). vat writes
+     * the two arms to two files of the SAME shape — `grading.json` and
+     * `baseline.json` — so without this field a reader holding one of them
+     * cannot tell which arm it is looking at. Optional for the same reason
+     * `runNonce` is: an externally produced grading.json has no arm to declare.
+     */
+    arm: z.ZodOptional<z.ZodEnum<["with", "without"]>>;
 }, z.ZodTypeAny, "passthrough">>;
 export type GradingReport = z.infer<typeof GradingReportSchema>;
 /**
- * Published JSON Schema for grading.json — generated from {@link GradingReportSchema}
- * so the two never drift. Importable by external tooling that wants to validate a
- * grading.json without depending on Zod. Documented in
- * docs/skill-test-grading-schema.md.
+ * Published JSON Schema for grading.json — generated from {@link GradingReportSchema},
+ * so its SHAPE (fields, types, required-ness, additionalProperties) cannot drift from
+ * the Zod schema. Importable by external tooling that wants to validate a grading.json
+ * without depending on Zod. Documented in docs/skill-test-grading-schema.md.
+ *
+ * It is NOT equivalent to the Zod schema, and never can be. `zodToJsonSchema` drops
+ * every `.refine()`, and the two refinements in this file are cross-field relations
+ * JSON Schema has no way to express — so the generated document is strictly weaker
+ * than the validator that generated it. That gap is stated, in the document itself,
+ * on the field it applies to (see {@link SUMMARY_SCHEMA_DESCRIPTION}); a consumer
+ * that needs the invariant enforced must re-implement it or use the Zod schema.
  */
 export declare const GradingReportJsonSchema: import("zod-to-json-schema").JsonSchema7Type & {
     $schema?: string | undefined;

package/dist/skill-test/grading-schema.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"grading-schema.d.ts","sourceRoot":"","sources":["../../src/skill-test/grading-schema.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAGxB;;;;;;;;;;;;;;;;;;;;;;;;;;GA0BG;AAEH,sFAAsF;AACtF,eAAO,MAAM,uBAAuB;;;;;;;;;;;;gCAMpB,CAAC;AAEjB,MAAM,MAAM,iBAAiB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,uBAAuB,CAAC,CAAC;AAExE;;;;GAIG;AACH,eAAO,MAAM,oBAAoB;;;;;;;;;;;;;;;;;;;;;;;;;gCAU7B,CAAC;AAEL,MAAM,MAAM,cAAc,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,oBAAoB,CAAC,CAAC;AAElE;;;;GAIG;AACH,eAAO,MAAM,mBAAmB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;IAI5B;;;;;;;OAOG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;IAPH;;;;;;;OAOG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;IAPH;;;;;;;OAOG;;gCAGS,CAAC;AAEjB,MAAM,MAAM,aAAa,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,mBAAmB,CAAC,CAAC;AAEhE;;;;;GAKG;AACH,eAAO,MAAM,uBAAuB;;;;;CAAyD,CAAC"}
+{"version":3,"file":"grading-schema.d.ts","sourceRoot":"","sources":["../../src/skill-test/grading-schema.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAGxB;;;;;;;;;;;;;;;;;;;;;;;;;;GA0BG;AAEH,sFAAsF;AACtF,eAAO,MAAM,uBAAuB;;;;IAKhC;;;;;;;;OAQG;;;;;;IARH;;;;;;;;OAQG;;;;;;IARH;;;;;;;;OAQG;;gCAGS,CAAC;AAEjB,MAAM,MAAM,iBAAiB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,uBAAuB,CAAC,CAAC;AAiCxE;;;;;GAKG;AACH,eAAO,MAAM,oBAAoB;;;;;;;;;;;;;;;;;;;;;;;;;gCAcM,CAAC;AAExC,MAAM,MAAM,cAAc,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,oBAAoB,CAAC,CAAC;AAElE;;;;GAIG;AACH,eAAO,MAAM,mBAAmB;;;;;QA3E5B;;;;;;;;WAQG;;;;;;QARH;;;;;;;;WAQG;;;;;;QARH;;;;;;;;WAQG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;IAuEH;;;;;;;OAOG;;IAEH;;;;;;;OAOG;;;;;;;QA/FH;;;;;;;;WAQG;;;;;;QARH;;;;;;;;WAQG;;;;;;QARH;;;;;;;;WAQG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;IAuEH;;;;;;;OAOG;;IAEH;;;;;;;OAOG;;;;;;;QA/FH;;;;;;;;WAQG;;;;;;QARH;;;;;;;;WAQG;;;;;;QARH;;;;;;;;WAQG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;IAuEH;;;;;;;OAOG;;IAEH;;;;;;;OAOG;;gCAGS,CAAC;AAEjB,MAAM,MAAM,aAAa,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,mBAAmB,CAAC,CAAC;AAEhE;;;;;;;;;;;;GAYG;AACH,eAAO,MAAM,uBAAuB;;;;;CAAyD,CAAC"}

package/dist/skill-test/grading-schema.js

@@ -33,12 +33,51 @@ export const GradedExpectationSchema = z
     text: z.string(),
     passed: z.boolean(),
     evidence: z.string().optional(),
+    /**
+     * Which eval this expectation was graded under. vat stamps it when it
+     * assembles the aggregate from per-eval grader fragments, so that a
+     * `--baseline` run's two artifacts can be lined up PER EVAL rather than only
+     * in aggregate — without it, a reader can compare two totals but cannot say
+     * which eval moved. Optional because a grading.json produced by external
+     * tooling (skill-creator's own) carries no per-eval attribution; the schema
+     * documents the field vat now emits rather than tightening the contract.
+     */
+    evalId: z.string().optional(),
 })
     .passthrough();
+/**
+ * The cross-field invariant, worded once and carried into BOTH surfaces that can
+ * state it: the Zod refine's failure message, and the `description` on `summary`
+ * in the published JSON Schema.
+ *
+ * It has to be carried explicitly because `zodToJsonSchema` DISCARDS `.refine()`.
+ * Without this, the published document said `{passed: integer≥0, total: integer≥0}`
+ * and nothing else — so external tooling validating against it accepted
+ * `{passed: 9, total: 3}`, the exact shape the Zod validator rejects and the exact
+ * shape that produced a confident `delta: -8`. The invariant was not enforced OR
+ * mentioned, which is worse than unenforced: a consumer reading the schema had no
+ * way to learn the check existed.
+ *
+ * JSON Schema cannot close the gap. No draft can compare one property against
+ * another (`maximum` takes a number, not a pointer; `if`/`then` can only test a
+ * value, not a relation), and the Ajv `$data` extension that could is not portable
+ * — a plain Ajv without `$data: true` FAILS TO COMPILE a schema containing it,
+ * which would break every conforming consumer to serve one. So the published
+ * schema records the constraint and says plainly that it is on the consumer to
+ * check, rather than pretending to enforce it.
+ */
+const SUMMARY_INVARIANT_MESSAGE = 'summary.passed must not exceed summary.total';
+const SUMMARY_SCHEMA_DESCRIPTION = 'Aggregate pass/fail counts. INVARIANT: ' +
+    `${SUMMARY_INVARIANT_MESSAGE}. It is enforced by the Zod schema this document is generated from ` +
+    'and by vat when it merges grader fragments, but NOT by this document: JSON Schema cannot express ' +
+    'a comparison between two properties in any draft, so a validator running this schema alone will ' +
+    'accept {"passed": 9, "total": 3}. Check it yourself before trusting any figure derived from these ' +
+    'counts (a baseline delta subtracts them).';
 /**
  * Aggregate pass/fail counts. `failed`/`pass_rate` are documented but optional.
  * Counts are non-negative integers (a float or negative count is a grader bug),
- * and `passed` can never exceed `total`.
+ * and `passed` can never exceed `total` — see {@link SUMMARY_SCHEMA_DESCRIPTION}
+ * for why that last one has to be stated twice.
  */
 export const GradingSummarySchema = z
     .object({
@@ -49,8 +88,12 @@ export const GradingSummarySchema = z
 })
     .passthrough()
     .refine(s => s.passed <= s.total, {
-    message: 'summary.passed must not exceed summary.total',
-});
+    message: SUMMARY_INVARIANT_MESSAGE,
+})
+    // `.describe()` AFTER `.refine()`: zodToJsonSchema reads the description off the
+    // outermost node, and this is the only channel through which a Zod refinement's
+    // meaning can reach the emitted document at all.
+    .describe(SUMMARY_SCHEMA_DESCRIPTION);
 /**
  * The full grading.json contract. Required: top-level `expectations[]` and
  * `summary`. Everything else passes through untouched (forward-compatible with
@@ -69,13 +112,29 @@ export const GradingReportSchema = z
      * by untrusted skill code is rejected.
      */
     runNonce: z.string().optional(),
+    /**
+     * Which `--baseline` arm produced this report: `'with'` (the skill declared)
+     * or `'without'` (the control arm run with the skill withheld). vat writes
+     * the two arms to two files of the SAME shape — `grading.json` and
+     * `baseline.json` — so without this field a reader holding one of them
+     * cannot tell which arm it is looking at. Optional for the same reason
+     * `runNonce` is: an externally produced grading.json has no arm to declare.
+     */
+    arm: z.enum(['with', 'without']).optional(),
 })
     .passthrough();
 /**
- * Published JSON Schema for grading.json — generated from {@link GradingReportSchema}
- * so the two never drift. Importable by external tooling that wants to validate a
- * grading.json without depending on Zod. Documented in
- * docs/skill-test-grading-schema.md.
+ * Published JSON Schema for grading.json — generated from {@link GradingReportSchema},
+ * so its SHAPE (fields, types, required-ness, additionalProperties) cannot drift from
+ * the Zod schema. Importable by external tooling that wants to validate a grading.json
+ * without depending on Zod. Documented in docs/skill-test-grading-schema.md.
+ *
+ * It is NOT equivalent to the Zod schema, and never can be. `zodToJsonSchema` drops
+ * every `.refine()`, and the two refinements in this file are cross-field relations
+ * JSON Schema has no way to express — so the generated document is strictly weaker
+ * than the validator that generated it. That gap is stated, in the document itself,
+ * on the field it applies to (see {@link SUMMARY_SCHEMA_DESCRIPTION}); a consumer
+ * that needs the invariant enforced must re-implement it or use the Zod schema.
  */
 export const GradingReportJsonSchema = zodToJsonSchema(GradingReportSchema, 'grading-report');
 //# sourceMappingURL=grading-schema.js.map

package/dist/skill-test/grading-schema.js.map

@@ -1 +1 @@
-{"version":3,"file":"grading-schema.js","sourceRoot":"","sources":["../../src/skill-test/grading-schema.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,eAAe,EAAE,MAAM,oBAAoB,CAAC;AAErD;;;;;;;;;;;;;;;;;;;;;;;;;;GA0BG;AAEH,sFAAsF;AACtF,MAAM,CAAC,MAAM,uBAAuB,GAAG,CAAC;KACrC,MAAM,CAAC;IACN,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE;IAChB,MAAM,EAAE,CAAC,CAAC,OAAO,EAAE;IACnB,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;CAChC,CAAC;KACD,WAAW,EAAE,CAAC;AAIjB;;;;GAIG;AACH,MAAM,CAAC,MAAM,oBAAoB,GAAG,CAAC;KAClC,MAAM,CAAC;IACN,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,WAAW,EAAE;IACtC,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,WAAW,EAAE;IACrC,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,WAAW,EAAE,CAAC,QAAQ,EAAE;IACjD,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;CACjC,CAAC;KACD,WAAW,EAAE;KACb,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,IAAI,CAAC,CAAC,KAAK,EAAE;IAChC,OAAO,EAAE,8CAA8C;CACxD,CAAC,CAAC;AAIL;;;;GAIG;AACH,MAAM,CAAC,MAAM,mBAAmB,GAAG,CAAC;KACjC,MAAM,CAAC;IACN,YAAY,EAAE,CAAC,CAAC,KAAK,CAAC,uBAAuB,CAAC;IAC9C,OAAO,EAAE,oBAAoB;IAC7B;;;;;;;OAOG;IACH,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;CAChC,CAAC;KACD,WAAW,EAAE,CAAC;AAIjB;;;;;GAKG;AACH,MAAM,CAAC,MAAM,uBAAuB,GAAG,eAAe,CAAC,mBAAmB,EAAE,gBAAgB,CAAC,CAAC"}
+{"version":3,"file":"grading-schema.js","sourceRoot":"","sources":["../../src/skill-test/grading-schema.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,eAAe,EAAE,MAAM,oBAAoB,CAAC;AAErD;;;;;;;;;;;;;;;;;;;;;;;;;;GA0BG;AAEH,sFAAsF;AACtF,MAAM,CAAC,MAAM,uBAAuB,GAAG,CAAC;KACrC,MAAM,CAAC;IACN,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE;IAChB,MAAM,EAAE,CAAC,CAAC,OAAO,EAAE;IACnB,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC/B;;;;;;;;OAQG;IACH,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;CAC9B,CAAC;KACD,WAAW,EAAE,CAAC;AAIjB;;;;;;;;;;;;;;;;;;;;GAoBG;AACH,MAAM,yBAAyB,GAAG,8CAA8C,CAAC;AAEjF,MAAM,0BAA0B,GAC9B,yCAAyC;IACzC,GAAG,yBAAyB,qEAAqE;IACjG,mGAAmG;IACnG,kGAAkG;IAClG,oGAAoG;IACpG,2CAA2C,CAAC;AAE9C;;;;;GAKG;AACH,MAAM,CAAC,MAAM,oBAAoB,GAAG,CAAC;KAClC,MAAM,CAAC;IACN,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,WAAW,EAAE;IACtC,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,WAAW,EAAE;IACrC,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,WAAW,EAAE,CAAC,QAAQ,EAAE;IACjD,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;CACjC,CAAC;KACD,WAAW,EAAE;KACb,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,IAAI,CAAC,CAAC,KAAK,EAAE;IAChC,OAAO,EAAE,yBAAyB;CACnC,CAAC;IACF,iFAAiF;IACjF,gFAAgF;IAChF,iDAAiD;KAChD,QAAQ,CAAC,0BAA0B,CAAC,CAAC;AAIxC;;;;GAIG;AACH,MAAM,CAAC,MAAM,mBAAmB,GAAG,CAAC;KACjC,MAAM,CAAC;IACN,YAAY,EAAE,CAAC,CAAC,KAAK,CAAC,uBAAuB,CAAC;IAC9C,OAAO,EAAE,oBAAoB;IAC7B;;;;;;;OAOG;IACH,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC/B;;;;;;;OAOG;IACH,GAAG,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC,CAAC,QAAQ,EAAE;CAC5C,CAAC;KACD,WAAW,EAAE,CAAC;AAIjB;;;;;;;;;;;;GAYG;AACH,MAAM,CAAC,MAAM,uBAAuB,GAAG,eAAe,CAAC,mBAAmB,EAAE,gBAAgB,CAAC,CAAC"}

package/dist/skill-test/harness-location.d.ts

@@ -1,6 +1,7 @@
+import { VatError } from '@vibe-agent-toolkit/utils';
 /** Harness-location failure — maps to exit code 2. */
-export declare class HarnessLocationError extends Error {
-    readonly exitCode: 2;
+export declare class HarnessLocationError extends VatError {
+    readonly reason: "preflight";
     constructor(message: string);
 }
 /**

package/dist/skill-test/harness-location.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"harness-location.d.ts","sourceRoot":"","sources":["../../src/skill-test/harness-location.ts"],"names":[],"mappings":"AAKA,sDAAsD;AACtD,qBAAa,oBAAqB,SAAQ,KAAK;IAC7C,QAAQ,CAAC,QAAQ,EAAG,CAAC,CAAU;gBACnB,OAAO,EAAE,MAAM;CAI5B;AAOD;;;;GAIG;AACH,wBAAgB,gBAAgB,CAAC,UAAU,EAAE,MAAM,EAAE,GAAG,MAAM,CAQ7D;AAED,mEAAmE;AACnE,wBAAgB,kBAAkB,CAAC,UAAU,EAAE,MAAM,EAAE,EAAE,OAAO,CAAC,EAAE,MAAM,GAAG,MAAM,CAGjF;AAED;;;;;;;;;;;;GAYG;AACH,wBAAgB,iBAAiB,CAAC,GAAG,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,MAAM,GAAG,IAAI,CAgBpE;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,kBAAkB,CAAC,GAAG,EAAE,MAAM,GAAG,IAAI,CAkBpD;AAkDD;;;;;;;;;;;;;;;GAeG;AACH,wBAAgB,qBAAqB,CACnC,GAAG,EAAE,MAAM,EACX,UAAU,EAAE,MAAM,EAClB,WAAW,GAAE,MAA2B,GACvC,IAAI,CAkBN"}
+{"version":3,"file":"harness-location.d.ts","sourceRoot":"","sources":["../../src/skill-test/harness-location.ts"],"names":[],"mappings":"AAGA,OAAO,EAAmD,QAAQ,EAAE,MAAM,2BAA2B,CAAC;AAEtG,sDAAsD;AACtD,qBAAa,oBAAqB,SAAQ,QAAQ;IAChD,QAAQ,CAAC,MAAM,EAAG,WAAW,CAAU;gBAC3B,OAAO,EAAE,MAAM;CAG5B;AAOD;;;;GAIG;AACH,wBAAgB,gBAAgB,CAAC,UAAU,EAAE,MAAM,EAAE,GAAG,MAAM,CAQ7D;AAED,mEAAmE;AACnE,wBAAgB,kBAAkB,CAAC,UAAU,EAAE,MAAM,EAAE,EAAE,OAAO,CAAC,EAAE,MAAM,GAAG,MAAM,CAGjF;AAED;;;;;;;;;;;;GAYG;AACH,wBAAgB,iBAAiB,CAAC,GAAG,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,MAAM,GAAG,IAAI,CAcpE;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,kBAAkB,CAAC,GAAG,EAAE,MAAM,GAAG,IAAI,CAcpD;AAgDD;;;;;;;;;;;;;;;GAeG;AACH,wBAAgB,qBAAqB,CACnC,GAAG,EAAE,MAAM,EACX,UAAU,EAAE,MAAM,EAClB,WAAW,GAAE,MAA2B,GACvC,IAAI,CAgBN"}

package/dist/skill-test/harness-location.js

@@ -1,12 +1,11 @@
 import { createHash } from 'node:crypto';
 import { chmodSync, existsSync, lstatSync, statSync } from 'node:fs';
-import { isAbsolutePath, normalizedTmpdir, safePath } from '@vibe-agent-toolkit/utils';
+import { normalizedTmpdir, relativeEscapesRoot, safePath, VatError } from '@vibe-agent-toolkit/utils';
 /** Harness-location failure — maps to exit code 2. */
-export class HarnessLocationError extends Error {
-    exitCode = 2;
+export class HarnessLocationError extends VatError {
+    reason = 'preflight';
     constructor(message) {
-        super(message);
-        this.name = 'HarnessLocationError';
+        super('HARNESS_LOCATION', message);
     }
 }
 /** Sanitize a skill name to a path-safe token (no separators, no traversal). */
@@ -50,11 +49,9 @@ export function assertSafeWorkdir(dir, stopAt) {
     let current = safePath.resolve(dir);
     let previous = '';
     while (current !== previous && current !== boundary) {
-        // eslint-disable-next-line security/detect-non-literal-fs-filename -- ancestry walk over a user-supplied workdir
         if (existsSync(safePath.join(current, 'CLAUDE.md'))) {
             throw new HarnessLocationError(`--workdir is inside a project: CLAUDE.md found at ${current}. Use an OS-tmp location.`);
         }
-        // eslint-disable-next-line security/detect-non-literal-fs-filename -- ancestry walk over a user-supplied workdir
         if (existsSync(safePath.join(current, '.claude'))) {
             throw new HarnessLocationError(`--workdir is inside a project: .claude/ found at ${current}. Use an OS-tmp location.`);
         }
@@ -75,19 +72,15 @@ export function assertSafeWorkdir(dir, stopAt) {
  * assertSafeHarnessRoot's platform guard).
  */
 export function prepareHarnessRoot(dir) {
-    // eslint-disable-next-line security/detect-non-literal-fs-filename -- our own derived harness root
     if (!existsSync(dir))
         return;
-    // eslint-disable-next-line security/detect-non-literal-fs-filename -- our own derived harness root
     const ls = lstatSync(dir);
     if (ls.isSymbolicLink()) {
         throw new HarnessLocationError(`Refusing to use a symlinked harness root: ${dir}.`);
     }
     if (process.platform !== 'win32') {
-        // eslint-disable-next-line security/detect-non-literal-fs-filename -- our own derived harness root
         const mode = statSync(dir).mode & 0o777;
         if (mode !== 0o700) {
-            // eslint-disable-next-line security/detect-non-literal-fs-filename -- our own derived harness root
             chmodSync(dir, 0o700);
         }
     }
@@ -97,7 +90,7 @@ function isStrictlyUnder(root, child) {
     if (child === root)
         return false;
     const rel = safePath.relative(root, child);
-    return rel !== '' && rel !== '..' && !rel.startsWith('../') && !isAbsolutePath(rel);
+    return rel !== '' && !relativeEscapesRoot(rel);
 }
 /**
  * The harness path components to validate, leaf-first: the leaf and every
@@ -127,10 +120,8 @@ function harnessAncestry(leaf, trustedRoot) {
  * skipped (no real uids); the symlink refusal still applies.
  */
 function assertComponentSafe(component, currentUid) {
-    // eslint-disable-next-line security/detect-non-literal-fs-filename -- our own derived harness path component
     if (!existsSync(component))
         return;
-    // eslint-disable-next-line security/detect-non-literal-fs-filename -- our own derived harness path component
     const ls = lstatSync(component);
     if (ls.isSymbolicLink()) {
         throw new HarnessLocationError(`Refusing to use a symlinked harness path component: ${component}.`);
@@ -164,10 +155,8 @@ export function assertSafeHarnessRoot(dir, currentUid, trustedRoot = normalizedT
     }
     // Leaf-only 0700 check (intermediates are created 0700 by recursive mkdir and
     // need only the symlink/ownership guarantees above). Skipped on win32.
-    // eslint-disable-next-line security/detect-non-literal-fs-filename -- our own derived harness root
     if (!existsSync(resolved))
         return; // not yet created — caller creates it 0700
-    // eslint-disable-next-line security/detect-non-literal-fs-filename -- our own derived harness root
     const mode = statSync(resolved).mode & 0o777;
     if (mode !== 0o700 && process.platform !== 'win32') {
         throw new HarnessLocationError(`Harness root ${resolved} must be 0700 (found ${mode.toString(8)}).`);

package/dist/skill-test/harness-location.js.map

@@ -1 +1 @@
-{"version":3,"file":"harness-location.js","sourceRoot":"","sources":["../../src/skill-test/harness-location.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,SAAS,EAAE,UAAU,EAAE,SAAS,EAAE,QAAQ,EAAE,MAAM,SAAS,CAAC;AAErE,OAAO,EAAE,cAAc,EAAE,gBAAgB,EAAE,QAAQ,EAAE,MAAM,2BAA2B,CAAC;AAEvF,sDAAsD;AACtD,MAAM,OAAO,oBAAqB,SAAQ,KAAK;IACpC,QAAQ,GAAG,CAAU,CAAC;IAC/B,YAAY,OAAe;QACzB,KAAK,CAAC,OAAO,CAAC,CAAC;QACf,IAAI,CAAC,IAAI,GAAG,sBAAsB,CAAC;IACrC,CAAC;CACF;AAED,gFAAgF;AAChF,SAAS,QAAQ,CAAC,IAAY;IAC5B,OAAO,IAAI,CAAC,UAAU,CAAC,iBAAiB,EAAE,GAAG,CAAC,CAAC;AACjD,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,gBAAgB,CAAC,UAAoB;IACnD,IAAI,UAAU,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC5B,MAAM,IAAI,oBAAoB,CAAC,sDAAsD,CAAC,CAAC;IACzF,CAAC;IACD,MAAM,MAAM,GAAG,CAAC,GAAG,UAAU,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,CAAC;IAClE,MAAM,MAAM,GAAG,MAAM,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IAC9C,MAAM,IAAI,GAAG,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IACtF,OAAO,GAAG,MAAM,IAAI,IAAI,EAAE,CAAC;AAC7B,CAAC;AAED,mEAAmE;AACnE,MAAM,UAAU,kBAAkB,CAAC,UAAoB,EAAE,OAAgB;IACvE,MAAM,IAAI,GAAG,OAAO,IAAI,gBAAgB,EAAE,CAAC;IAC3C,OAAO,QAAQ,CAAC,IAAI,CAAC,IAAI,EAAE,gBAAgB,EAAE,gBAAgB,CAAC,UAAU,CAAC,CAAC,CAAC;AAC7E,CAAC;AAED;;;;;;;;;;;;GAYG;AACH,MAAM,UAAU,iBAAiB,CAAC,GAAW,EAAE,MAAe;IAC5D,MAAM,QAAQ,GAAG,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,QAAQ,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;IAC7E,IAAI,OAAO,GAAG,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;IACpC,IAAI,QAAQ,GAAG,EAAE,CAAC;IAClB,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,KAAK,QAAQ,EAAE,CAAC;QACpD,iHAAiH;QACjH,IAAI,UAAU,CAAC,QAAQ,CAAC,IAAI,CAAC,OAAO,EAAE,WAAW,CAAC,CAAC,EAAE,CAAC;YACpD,MAAM,IAAI,oBAAoB,CAAC,qDAAqD,OAAO,2BAA2B,CAAC,CAAC;QAC1H,CAAC;QACD,iHAAiH;QACjH,IAAI,UAAU,CAAC,QAAQ,CAAC,IAAI,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC,EAAE,CAAC;YAClD,MAAM,IAAI,oBAAoB,CAAC,oDAAoD,OAAO,2BAA2B,CAAC,CAAC;QACzH,CAAC;QACD,QAAQ,GAAG,OAAO,CAAC;QACnB,OAAO,GAAG,QAAQ,CAAC,IAAI,CAAC,OAAO,EAAE,IAAI,CAAC,CAAC;IACzC,CAAC;AACH,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,kBAAkB,CAAC,GAAW;IAC5C,mGAAmG;IACnG,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC;QAAE,OAAO;IAE7B,mGAAmG;IACnG,MAAM,EAAE,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC;IAC1B,IAAI,EAAE,CAAC,cAAc,EAAE,EAAE,CAAC;QACxB,MAAM,IAAI,oBAAoB,CAAC,6CAA6C,GAAG,GAAG,CAAC,CAAC;IACtF,CAAC;IAED,IAAI,OAAO,CAAC,QAAQ,KAAK,OAAO,EAAE,CAAC;QACjC,mGAAmG;QACnG,MAAM,IAAI,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,IAAI,GAAG,KAAK,CAAC;QACxC,IAAI,IAAI,KAAK,KAAK,EAAE,CAAC;YACnB,mGAAmG;YACnG,SAAS,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;QACxB,CAAC;IACH,CAAC;AACH,CAAC;AAED,uFAAuF;AACvF,SAAS,eAAe,CAAC,IAAY,EAAE,KAAa;IAClD,IAAI,KAAK,KAAK,IAAI;QAAE,OAAO,KAAK,CAAC;IACjC,MAAM,GAAG,GAAG,QAAQ,CAAC,QAAQ,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;IAC3C,OAAO,GAAG,KAAK,EAAE,IAAI,GAAG,KAAK,IAAI,IAAI,CAAC,GAAG,CAAC,UAAU,CAAC,KAAK,CAAC,IAAI,CAAC,cAAc,CAAC,GAAG,CAAC,CAAC;AACtF,CAAC;AAED;;;;;;GAMG;AACH,SAAS,eAAe,CAAC,IAAY,EAAE,WAAmB;IACxD,IAAI,CAAC,eAAe,CAAC,WAAW,EAAE,IAAI,CAAC;QAAE,OAAO,CAAC,IAAI,CAAC,CAAC;IACvD,MAAM,UAAU,GAAa,EAAE,CAAC;IAChC,IAAI,OAAO,GAAG,IAAI,CAAC;IACnB,OAAO,OAAO,KAAK,WAAW,EAAE,CAAC;QAC/B,MAAM,MAAM,GAAG,QAAQ,CAAC,IAAI,CAAC,OAAO,EAAE,IAAI,CAAC,CAAC;QAC5C,IAAI,MAAM,KAAK,OAAO;YAAE,MAAM,CAAC,0CAA0C;QACzE,UAAU,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QACzB,OAAO,GAAG,MAAM,CAAC;IACnB,CAAC;IACD,OAAO,UAAU,CAAC;AACpB,CAAC;AAED;;;;;GAKG;AACH,SAAS,mBAAmB,CAAC,SAAiB,EAAE,UAAkB;IAChE,6GAA6G;IAC7G,IAAI,CAAC,UAAU,CAAC,SAAS,CAAC;QAAE,OAAO;IACnC,6GAA6G;IAC7G,MAAM,EAAE,GAAG,SAAS,CAAC,SAAS,CAAC,CAAC;IAChC,IAAI,EAAE,CAAC,cAAc,EAAE,EAAE,CAAC;QACxB,MAAM,IAAI,oBAAoB,CAAC,uDAAuD,SAAS,GAAG,CAAC,CAAC;IACtG,CAAC;IACD,IAAI,OAAO,CAAC,QAAQ,KAAK,OAAO,IAAI,OAAO,EAAE,CAAC,GAAG,KAAK,QAAQ,IAAI,EAAE,CAAC,GAAG,KAAK,UAAU,EAAE,CAAC;QACxF,MAAM,IAAI,oBAAoB,CAC5B,0BAA0B,SAAS,0CAA0C,EAAE,CAAC,GAAG,OAAO,UAAU,IAAI,CACzG,CAAC;IACJ,CAAC;AACH,CAAC;AAED;;;;;;;;;;;;;;;GAeG;AACH,MAAM,UAAU,qBAAqB,CACnC,GAAW,EACX,UAAkB,EAClB,cAAsB,gBAAgB,EAAE;IAExC,MAAM,QAAQ,GAAG,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;IACvC,8NAA8N;IAC9N,MAAM,IAAI,GAAG,QAAQ,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC;IAE3C,KAAK,MAAM,SAAS,IAAI,eAAe,CAAC,QAAQ,EAAE,IAAI,CAAC,EAAE,CAAC;QACxD,mBAAmB,CAAC,SAAS,EAAE,UAAU,CAAC,CAAC;IAC7C,CAAC;IAED,8EAA8E;IAC9E,uEAAuE;IACvE,mGAAmG;IACnG,IAAI,CAAC,UAAU,CAAC,QAAQ,CAAC;QAAE,OAAO,CAAC,2CAA2C;IAC9E,mGAAmG;IACnG,MAAM,IAAI,GAAG,QAAQ,CAAC,QAAQ,CAAC,CAAC,IAAI,GAAG,KAAK,CAAC;IAC7C,IAAI,IAAI,KAAK,KAAK,IAAI,OAAO,CAAC,QAAQ,KAAK,OAAO,EAAE,CAAC;QACnD,MAAM,IAAI,oBAAoB,CAAC,gBAAgB,QAAQ,wBAAwB,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;IACvG,CAAC;AACH,CAAC"}
+{"version":3,"file":"harness-location.js","sourceRoot":"","sources":["../../src/skill-test/harness-location.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,SAAS,EAAE,UAAU,EAAE,SAAS,EAAE,QAAQ,EAAE,MAAM,SAAS,CAAC;AAErE,OAAO,EAAE,gBAAgB,EAAE,mBAAmB,EAAE,QAAQ,EAAE,QAAQ,EAAE,MAAM,2BAA2B,CAAC;AAEtG,sDAAsD;AACtD,MAAM,OAAO,oBAAqB,SAAQ,QAAQ;IACvC,MAAM,GAAG,WAAoB,CAAC;IACvC,YAAY,OAAe;QACzB,KAAK,CAAC,kBAAkB,EAAE,OAAO,CAAC,CAAC;IACrC,CAAC;CACF;AAED,gFAAgF;AAChF,SAAS,QAAQ,CAAC,IAAY;IAC5B,OAAO,IAAI,CAAC,UAAU,CAAC,iBAAiB,EAAE,GAAG,CAAC,CAAC;AACjD,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,gBAAgB,CAAC,UAAoB;IACnD,IAAI,UAAU,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC5B,MAAM,IAAI,oBAAoB,CAAC,sDAAsD,CAAC,CAAC;IACzF,CAAC;IACD,MAAM,MAAM,GAAG,CAAC,GAAG,UAAU,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,CAAC;IAClE,MAAM,MAAM,GAAG,MAAM,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IAC9C,MAAM,IAAI,GAAG,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IACtF,OAAO,GAAG,MAAM,IAAI,IAAI,EAAE,CAAC;AAC7B,CAAC;AAED,mEAAmE;AACnE,MAAM,UAAU,kBAAkB,CAAC,UAAoB,EAAE,OAAgB;IACvE,MAAM,IAAI,GAAG,OAAO,IAAI,gBAAgB,EAAE,CAAC;IAC3C,OAAO,QAAQ,CAAC,IAAI,CAAC,IAAI,EAAE,gBAAgB,EAAE,gBAAgB,CAAC,UAAU,CAAC,CAAC,CAAC;AAC7E,CAAC;AAED;;;;;;;;;;;;GAYG;AACH,MAAM,UAAU,iBAAiB,CAAC,GAAW,EAAE,MAAe;IAC5D,MAAM,QAAQ,GAAG,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,QAAQ,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;IAC7E,IAAI,OAAO,GAAG,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;IACpC,IAAI,QAAQ,GAAG,EAAE,CAAC;IAClB,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,KAAK,QAAQ,EAAE,CAAC;QACpD,IAAI,UAAU,CAAC,QAAQ,CAAC,IAAI,CAAC,OAAO,EAAE,WAAW,CAAC,CAAC,EAAE,CAAC;YACpD,MAAM,IAAI,oBAAoB,CAAC,qDAAqD,OAAO,2BAA2B,CAAC,CAAC;QAC1H,CAAC;QACD,IAAI,UAAU,CAAC,QAAQ,CAAC,IAAI,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC,EAAE,CAAC;YAClD,MAAM,IAAI,oBAAoB,CAAC,oDAAoD,OAAO,2BAA2B,CAAC,CAAC;QACzH,CAAC;QACD,QAAQ,GAAG,OAAO,CAAC;QACnB,OAAO,GAAG,QAAQ,CAAC,IAAI,CAAC,OAAO,EAAE,IAAI,CAAC,CAAC;IACzC,CAAC;AACH,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,kBAAkB,CAAC,GAAW;IAC5C,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC;QAAE,OAAO;IAE7B,MAAM,EAAE,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC;IAC1B,IAAI,EAAE,CAAC,cAAc,EAAE,EAAE,CAAC;QACxB,MAAM,IAAI,oBAAoB,CAAC,6CAA6C,GAAG,GAAG,CAAC,CAAC;IACtF,CAAC;IAED,IAAI,OAAO,CAAC,QAAQ,KAAK,OAAO,EAAE,CAAC;QACjC,MAAM,IAAI,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,IAAI,GAAG,KAAK,CAAC;QACxC,IAAI,IAAI,KAAK,KAAK,EAAE,CAAC;YACnB,SAAS,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;QACxB,CAAC;IACH,CAAC;AACH,CAAC;AAED,uFAAuF;AACvF,SAAS,eAAe,CAAC,IAAY,EAAE,KAAa;IAClD,IAAI,KAAK,KAAK,IAAI;QAAE,OAAO,KAAK,CAAC;IACjC,MAAM,GAAG,GAAG,QAAQ,CAAC,QAAQ,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;IAC3C,OAAO,GAAG,KAAK,EAAE,IAAI,CAAC,mBAAmB,CAAC,GAAG,CAAC,CAAC;AACjD,CAAC;AAED;;;;;;GAMG;AACH,SAAS,eAAe,CAAC,IAAY,EAAE,WAAmB;IACxD,IAAI,CAAC,eAAe,CAAC,WAAW,EAAE,IAAI,CAAC;QAAE,OAAO,CAAC,IAAI,CAAC,CAAC;IACvD,MAAM,UAAU,GAAa,EAAE,CAAC;IAChC,IAAI,OAAO,GAAG,IAAI,CAAC;IACnB,OAAO,OAAO,KAAK,WAAW,EAAE,CAAC;QAC/B,MAAM,MAAM,GAAG,QAAQ,CAAC,IAAI,CAAC,OAAO,EAAE,IAAI,CAAC,CAAC;QAC5C,IAAI,MAAM,KAAK,OAAO;YAAE,MAAM,CAAC,0CAA0C;QACzE,UAAU,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QACzB,OAAO,GAAG,MAAM,CAAC;IACnB,CAAC;IACD,OAAO,UAAU,CAAC;AACpB,CAAC;AAED;;;;;GAKG;AACH,SAAS,mBAAmB,CAAC,SAAiB,EAAE,UAAkB;IAChE,IAAI,CAAC,UAAU,CAAC,SAAS,CAAC;QAAE,OAAO;IACnC,MAAM,EAAE,GAAG,SAAS,CAAC,SAAS,CAAC,CAAC;IAChC,IAAI,EAAE,CAAC,cAAc,EAAE,EAAE,CAAC;QACxB,MAAM,IAAI,oBAAoB,CAAC,uDAAuD,SAAS,GAAG,CAAC,CAAC;IACtG,CAAC;IACD,IAAI,OAAO,CAAC,QAAQ,KAAK,OAAO,IAAI,OAAO,EAAE,CAAC,GAAG,KAAK,QAAQ,IAAI,EAAE,CAAC,GAAG,KAAK,UAAU,EAAE,CAAC;QACxF,MAAM,IAAI,oBAAoB,CAC5B,0BAA0B,SAAS,0CAA0C,EAAE,CAAC,GAAG,OAAO,UAAU,IAAI,CACzG,CAAC;IACJ,CAAC;AACH,CAAC;AAED;;;;;;;;;;;;;;;GAeG;AACH,MAAM,UAAU,qBAAqB,CACnC,GAAW,EACX,UAAkB,EAClB,cAAsB,gBAAgB,EAAE;IAExC,MAAM,QAAQ,GAAG,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;IACvC,8NAA8N;IAC9N,MAAM,IAAI,GAAG,QAAQ,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC;IAE3C,KAAK,MAAM,SAAS,IAAI,eAAe,CAAC,QAAQ,EAAE,IAAI,CAAC,EAAE,CAAC;QACxD,mBAAmB,CAAC,SAAS,EAAE,UAAU,CAAC,CAAC;IAC7C,CAAC;IAED,8EAA8E;IAC9E,uEAAuE;IACvE,IAAI,CAAC,UAAU,CAAC,QAAQ,CAAC;QAAE,OAAO,CAAC,2CAA2C;IAC9E,MAAM,IAAI,GAAG,QAAQ,CAAC,QAAQ,CAAC,CAAC,IAAI,GAAG,KAAK,CAAC;IAC7C,IAAI,IAAI,KAAK,KAAK,IAAI,OAAO,CAAC,QAAQ,KAAK,OAAO,EAAE,CAAC;QACnD,MAAM,IAAI,oBAAoB,CAAC,gBAAgB,QAAQ,wBAAwB,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;IACvG,CAAC;AACH,CAAC"}

package/dist/skill-test/index.d.ts

@@ -4,12 +4,12 @@ export { lintEvalExpectations, lintToolExpectationExecutables, type EvalLintWarn
 export { FrictionCategorySchema, FrictionItemSchema, FrictionReportJsonSchema, FrictionReportSchema, FrictionSeveritySchema, type FrictionItem, type FrictionReport, } from './friction-schema.js';
 export { mergeFragmentsToFriction, mergeFragmentsToGrading, mergeFragmentsToToolEval, } from './fragment-merge.js';
 export { ToolEvalReportJsonSchema, ToolEvalReportSchema, ToolVerdictBodySchema, ToolVerdictSchema, type ToolEvalReport, type ToolVerdict, type ToolVerdictBody, } from './tool-eval-schema.js';
-export { GradingNonceError, GradingSkewError, parseGradingJson, reconcileGrading, type GradingVerdict, type NormalizedGrading, } from './grading-adapter.js';
+export { GradingArmError, GradingNonceError, GradingSkewError, parseGradingJson, reconcileGrading, type GradingVerdict, type NormalizedGrading, } from './grading-adapter.js';
 export { GradedExpectationSchema, GradingReportJsonSchema, GradingReportSchema, GradingSummarySchema, type GradedExpectation, type GradingReport, type GradingSummary, } from './grading-schema.js';
 export { assertSafeHarnessRoot, assertSafeWorkdir, deriveHarnessKey, HarnessLocationError, resolveHarnessRoot, } from './harness-location.js';
 export { computeReconcilePlan, StagedEntrySchema, StagedManifestSchema, type ReconcilePlan, type StagedEntry, type StagedManifest, } from './manifest.js';
 export { assembleChildEnv, computeEnvTokens, interpolateEnvValue, resolveInjectEnv, UnknownEnvTokenError, type AssembleChildEnvInput, type AssembledChildEnv, type EnvInterpolationTokens, type EnvTokenInputs, } from './declared-env.js';
-export { BootstrapNeededError, DuplicateStagedSkillError, InternalHarnessError, mapErrorToExitCode, SecurityAckError, SkillBuildError, SkillTestExitCode, type SkillTestExitCodeValue, } from './exit-codes.js';
+export { BootstrapNeededError, DuplicateStagedSkillError, InternalHarnessError, SecurityAckError, SkillBuildError, type SkillTestFailureReason, skillTestFailureReason, } from './failure-reason.js';
 export { BuildHookError, runPreStageBuild } from './build-hook.js';
 export { DEFAULT_CONCURRENCY, DEFAULT_GRADER_MODEL } from './grader-model.js';
 export { RateLimitSignal, runPipeline, type RunPipelineOptions } from './pipeline.js';

package/dist/skill-test/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/skill-test/index.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,eAAe,EACf,cAAc,EACd,eAAe,EACf,cAAc,EACd,mBAAmB,EACnB,KAAK,SAAS,EACd,KAAK,SAAS,EACd,KAAK,wBAAwB,GAC9B,MAAM,kBAAkB,CAAC;AAC1B,OAAO,EACL,iBAAiB,EACjB,6BAA6B,EAC7B,kBAAkB,EAClB,iBAAiB,EACjB,KAAK,YAAY,EACjB,KAAK,uBAAuB,GAC7B,MAAM,oBAAoB,CAAC;AAC5B,OAAO,EAAE,oBAAoB,EAAE,8BAA8B,EAAE,KAAK,eAAe,EAAE,MAAM,gBAAgB,CAAC;AAC5G,OAAO,EACL,sBAAsB,EACtB,kBAAkB,EAClB,wBAAwB,EACxB,oBAAoB,EACpB,sBAAsB,EACtB,KAAK,YAAY,EACjB,KAAK,cAAc,GACpB,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EACL,wBAAwB,EACxB,uBAAuB,EACvB,wBAAwB,GACzB,MAAM,qBAAqB,CAAC;AAC7B,OAAO,EACL,wBAAwB,EACxB,oBAAoB,EACpB,qBAAqB,EACrB,iBAAiB,EACjB,KAAK,cAAc,EACnB,KAAK,WAAW,EAChB,KAAK,eAAe,GACrB,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EACL,iBAAiB,EACjB,gBAAgB,EAChB,gBAAgB,EAChB,gBAAgB,EAChB,KAAK,cAAc,EACnB,KAAK,iBAAiB,GACvB,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EACL,uBAAuB,EACvB,uBAAuB,EACvB,mBAAmB,EACnB,oBAAoB,EACpB,KAAK,iBAAiB,EACtB,KAAK,aAAa,EAClB,KAAK,cAAc,GACpB,MAAM,qBAAqB,CAAC;AAC7B,OAAO,EACL,qBAAqB,EACrB,iBAAiB,EACjB,gBAAgB,EAChB,oBAAoB,EACpB,kBAAkB,GACnB,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EACL,oBAAoB,EACpB,iBAAiB,EACjB,oBAAoB,EACpB,KAAK,aAAa,EAClB,KAAK,WAAW,EAChB,KAAK,cAAc,GACpB,MAAM,eAAe,CAAC;AACvB,OAAO,EACL,gBAAgB,EAChB,gBAAgB,EAChB,mBAAmB,EACnB,gBAAgB,EAChB,oBAAoB,EACpB,KAAK,qBAAqB,EAC1B,KAAK,iBAAiB,EACtB,KAAK,sBAAsB,EAC3B,KAAK,cAAc,GACpB,MAAM,mBAAmB,CAAC;AAC3B,OAAO,EACL,oBAAoB,EACpB,yBAAyB,EACzB,oBAAoB,EACpB,kBAAkB,EAClB,gBAAgB,EAChB,eAAe,EACf,iBAAiB,EACjB,KAAK,sBAAsB,GAC5B,MAAM,iBAAiB,CAAC;AACzB,OAAO,EAAE,cAAc,EAAE,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AACnE,OAAO,EAAE,mBAAmB,EAAE,oBAAoB,EAAE,MAAM,mBAAmB,CAAC;AAC9E,OAAO,EAAE,eAAe,EAAE,WAAW,EAAE,KAAK,kBAAkB,EAAE,MAAM,eAAe,CAAC;AACtF,OAAO,EACL,YAAY,EACZ,KAAK,cAAc,EACnB,KAAK,cAAc,EACnB,KAAK,eAAe,GACrB,MAAM,gBAAgB,CAAC;AACxB,OAAO,EACL,kBAAkB,EAClB,KAAK,eAAe,EACpB,KAAK,gBAAgB,GACtB,MAAM,oBAAoB,CAAC;AAC5B,OAAO,EACL,gBAAgB,EAChB,KAAK,cAAc,GACpB,MAAM,kBAAkB,CAAC;AAC1B,OAAO,EACL,8BAA8B,EAC9B,mBAAmB,EACnB,KAAK,0BAA0B,GAChC,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EACL,4BAA4B,EAC5B,iBAAiB,EACjB,KAAK,wBAAwB,GAC9B,MAAM,oBAAoB,CAAC;AAC5B,OAAO,EAAE,6BAA6B,EAAE,oBAAoB,EAAE,MAAM,wBAAwB,CAAC;AAC7F,OAAO,EAAE,kBAAkB,EAAE,oBAAoB,EAAE,KAAK,WAAW,EAAE,MAAM,WAAW,CAAC;AACvF,OAAO,EACL,qBAAqB,EACrB,kBAAkB,EAClB,aAAa,EACb,YAAY,EACZ,KAAK,mBAAmB,EACxB,KAAK,kBAAkB,EACvB,KAAK,SAAS,GACf,MAAM,cAAc,CAAC;AACtB,OAAO,EACL,iBAAiB,EACjB,gBAAgB,EAChB,KAAK,qBAAqB,GAC3B,MAAM,2BAA2B,CAAC;AACnC,OAAO,EAAE,0BAA0B,EAAE,sBAAsB,EAAE,sBAAsB,EAAE,MAAM,sBAAsB,CAAC;AAClH,OAAO,EAAE,gBAAgB,EAAE,MAAM,uBAAuB,CAAC;AACzD,OAAO,EAAE,kBAAkB,EAAE,kBAAkB,EAAE,MAAM,qBAAqB,CAAC;AAC7E,OAAO,EACL,kBAAkB,EAClB,0BAA0B,EAC1B,oBAAoB,EACpB,cAAc,EACd,mBAAmB,EACnB,uBAAuB,EACvB,eAAe,EACf,KAAK,kBAAkB,EACvB,KAAK,iBAAiB,EACtB,KAAK,gBAAgB,GACtB,MAAM,kBAAkB,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/skill-test/index.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,eAAe,EACf,cAAc,EACd,eAAe,EACf,cAAc,EACd,mBAAmB,EACnB,KAAK,SAAS,EACd,KAAK,SAAS,EACd,KAAK,wBAAwB,GAC9B,MAAM,kBAAkB,CAAC;AAC1B,OAAO,EACL,iBAAiB,EACjB,6BAA6B,EAC7B,kBAAkB,EAClB,iBAAiB,EACjB,KAAK,YAAY,EACjB,KAAK,uBAAuB,GAC7B,MAAM,oBAAoB,CAAC;AAC5B,OAAO,EAAE,oBAAoB,EAAE,8BAA8B,EAAE,KAAK,eAAe,EAAE,MAAM,gBAAgB,CAAC;AAC5G,OAAO,EACL,sBAAsB,EACtB,kBAAkB,EAClB,wBAAwB,EACxB,oBAAoB,EACpB,sBAAsB,EACtB,KAAK,YAAY,EACjB,KAAK,cAAc,GACpB,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EACL,wBAAwB,EACxB,uBAAuB,EACvB,wBAAwB,GACzB,MAAM,qBAAqB,CAAC;AAC7B,OAAO,EACL,wBAAwB,EACxB,oBAAoB,EACpB,qBAAqB,EACrB,iBAAiB,EACjB,KAAK,cAAc,EACnB,KAAK,WAAW,EAChB,KAAK,eAAe,GACrB,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EACL,eAAe,EACf,iBAAiB,EACjB,gBAAgB,EAChB,gBAAgB,EAChB,gBAAgB,EAChB,KAAK,cAAc,EACnB,KAAK,iBAAiB,GACvB,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EACL,uBAAuB,EACvB,uBAAuB,EACvB,mBAAmB,EACnB,oBAAoB,EACpB,KAAK,iBAAiB,EACtB,KAAK,aAAa,EAClB,KAAK,cAAc,GACpB,MAAM,qBAAqB,CAAC;AAC7B,OAAO,EACL,qBAAqB,EACrB,iBAAiB,EACjB,gBAAgB,EAChB,oBAAoB,EACpB,kBAAkB,GACnB,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EACL,oBAAoB,EACpB,iBAAiB,EACjB,oBAAoB,EACpB,KAAK,aAAa,EAClB,KAAK,WAAW,EAChB,KAAK,cAAc,GACpB,MAAM,eAAe,CAAC;AACvB,OAAO,EACL,gBAAgB,EAChB,gBAAgB,EAChB,mBAAmB,EACnB,gBAAgB,EAChB,oBAAoB,EACpB,KAAK,qBAAqB,EAC1B,KAAK,iBAAiB,EACtB,KAAK,sBAAsB,EAC3B,KAAK,cAAc,GACpB,MAAM,mBAAmB,CAAC;AAC3B,OAAO,EACL,oBAAoB,EACpB,yBAAyB,EACzB,oBAAoB,EACpB,gBAAgB,EAChB,eAAe,EACf,KAAK,sBAAsB,EAC3B,sBAAsB,GACvB,MAAM,qBAAqB,CAAC;AAC7B,OAAO,EAAE,cAAc,EAAE,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AACnE,OAAO,EAAE,mBAAmB,EAAE,oBAAoB,EAAE,MAAM,mBAAmB,CAAC;AAC9E,OAAO,EAAE,eAAe,EAAE,WAAW,EAAE,KAAK,kBAAkB,EAAE,MAAM,eAAe,CAAC;AACtF,OAAO,EACL,YAAY,EACZ,KAAK,cAAc,EACnB,KAAK,cAAc,EACnB,KAAK,eAAe,GACrB,MAAM,gBAAgB,CAAC;AACxB,OAAO,EACL,kBAAkB,EAClB,KAAK,eAAe,EACpB,KAAK,gBAAgB,GACtB,MAAM,oBAAoB,CAAC;AAC5B,OAAO,EACL,gBAAgB,EAChB,KAAK,cAAc,GACpB,MAAM,kBAAkB,CAAC;AAC1B,OAAO,EACL,8BAA8B,EAC9B,mBAAmB,EACnB,KAAK,0BAA0B,GAChC,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EACL,4BAA4B,EAC5B,iBAAiB,EACjB,KAAK,wBAAwB,GAC9B,MAAM,oBAAoB,CAAC;AAC5B,OAAO,EAAE,6BAA6B,EAAE,oBAAoB,EAAE,MAAM,wBAAwB,CAAC;AAC7F,OAAO,EAAE,kBAAkB,EAAE,oBAAoB,EAAE,KAAK,WAAW,EAAE,MAAM,WAAW,CAAC;AACvF,OAAO,EACL,qBAAqB,EACrB,kBAAkB,EAClB,aAAa,EACb,YAAY,EACZ,KAAK,mBAAmB,EACxB,KAAK,kBAAkB,EACvB,KAAK,SAAS,GACf,MAAM,cAAc,CAAC;AACtB,OAAO,EACL,iBAAiB,EACjB,gBAAgB,EAChB,KAAK,qBAAqB,GAC3B,MAAM,2BAA2B,CAAC;AACnC,OAAO,EAAE,0BAA0B,EAAE,sBAAsB,EAAE,sBAAsB,EAAE,MAAM,sBAAsB,CAAC;AAClH,OAAO,EAAE,gBAAgB,EAAE,MAAM,uBAAuB,CAAC;AACzD,OAAO,EAAE,kBAAkB,EAAE,kBAAkB,EAAE,MAAM,qBAAqB,CAAC;AAC7E,OAAO,EACL,kBAAkB,EAClB,0BAA0B,EAC1B,oBAAoB,EACpB,cAAc,EACd,mBAAmB,EACnB,uBAAuB,EACvB,eAAe,EACf,KAAK,kBAAkB,EACvB,KAAK,iBAAiB,EACtB,KAAK,gBAAgB,GACtB,MAAM,kBAAkB,CAAC"}

package/dist/skill-test/index.js

@@ -4,12 +4,12 @@ export { lintEvalExpectations, lintToolExpectationExecutables } from './eval-lin
 export { FrictionCategorySchema, FrictionItemSchema, FrictionReportJsonSchema, FrictionReportSchema, FrictionSeveritySchema, } from './friction-schema.js';
 export { mergeFragmentsToFriction, mergeFragmentsToGrading, mergeFragmentsToToolEval, } from './fragment-merge.js';
 export { ToolEvalReportJsonSchema, ToolEvalReportSchema, ToolVerdictBodySchema, ToolVerdictSchema, } from './tool-eval-schema.js';
-export { GradingNonceError, GradingSkewError, parseGradingJson, reconcileGrading, } from './grading-adapter.js';
+export { GradingArmError, GradingNonceError, GradingSkewError, parseGradingJson, reconcileGrading, } from './grading-adapter.js';
 export { GradedExpectationSchema, GradingReportJsonSchema, GradingReportSchema, GradingSummarySchema, } from './grading-schema.js';
 export { assertSafeHarnessRoot, assertSafeWorkdir, deriveHarnessKey, HarnessLocationError, resolveHarnessRoot, } from './harness-location.js';
 export { computeReconcilePlan, StagedEntrySchema, StagedManifestSchema, } from './manifest.js';
 export { assembleChildEnv, computeEnvTokens, interpolateEnvValue, resolveInjectEnv, UnknownEnvTokenError, } from './declared-env.js';
-export { BootstrapNeededError, DuplicateStagedSkillError, InternalHarnessError, mapErrorToExitCode, SecurityAckError, SkillBuildError, SkillTestExitCode, } from './exit-codes.js';
+export { BootstrapNeededError, DuplicateStagedSkillError, InternalHarnessError, SecurityAckError, SkillBuildError, skillTestFailureReason, } from './failure-reason.js';
 export { BuildHookError, runPreStageBuild } from './build-hook.js';
 export { DEFAULT_CONCURRENCY, DEFAULT_GRADER_MODEL } from './grader-model.js';
 export { RateLimitSignal, runPipeline } from './pipeline.js';

package/dist/skill-test/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/skill-test/index.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,eAAe,EACf,cAAc,EACd,eAAe,EACf,cAAc,EACd,mBAAmB,GAIpB,MAAM,kBAAkB,CAAC;AAC1B,OAAO,EACL,iBAAiB,EACjB,6BAA6B,EAC7B,kBAAkB,EAClB,iBAAiB,GAGlB,MAAM,oBAAoB,CAAC;AAC5B,OAAO,EAAE,oBAAoB,EAAE,8BAA8B,EAAwB,MAAM,gBAAgB,CAAC;AAC5G,OAAO,EACL,sBAAsB,EACtB,kBAAkB,EAClB,wBAAwB,EACxB,oBAAoB,EACpB,sBAAsB,GAGvB,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EACL,wBAAwB,EACxB,uBAAuB,EACvB,wBAAwB,GACzB,MAAM,qBAAqB,CAAC;AAC7B,OAAO,EACL,wBAAwB,EACxB,oBAAoB,EACpB,qBAAqB,EACrB,iBAAiB,GAIlB,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EACL,iBAAiB,EACjB,gBAAgB,EAChB,gBAAgB,EAChB,gBAAgB,GAGjB,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EACL,uBAAuB,EACvB,uBAAuB,EACvB,mBAAmB,EACnB,oBAAoB,GAIrB,MAAM,qBAAqB,CAAC;AAC7B,OAAO,EACL,qBAAqB,EACrB,iBAAiB,EACjB,gBAAgB,EAChB,oBAAoB,EACpB,kBAAkB,GACnB,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EACL,oBAAoB,EACpB,iBAAiB,EACjB,oBAAoB,GAIrB,MAAM,eAAe,CAAC;AACvB,OAAO,EACL,gBAAgB,EAChB,gBAAgB,EAChB,mBAAmB,EACnB,gBAAgB,EAChB,oBAAoB,GAKrB,MAAM,mBAAmB,CAAC;AAC3B,OAAO,EACL,oBAAoB,EACpB,yBAAyB,EACzB,oBAAoB,EACpB,kBAAkB,EAClB,gBAAgB,EAChB,eAAe,EACf,iBAAiB,GAElB,MAAM,iBAAiB,CAAC;AACzB,OAAO,EAAE,cAAc,EAAE,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AACnE,OAAO,EAAE,mBAAmB,EAAE,oBAAoB,EAAE,MAAM,mBAAmB,CAAC;AAC9E,OAAO,EAAE,eAAe,EAAE,WAAW,EAA2B,MAAM,eAAe,CAAC;AACtF,OAAO,EACL,YAAY,GAIb,MAAM,gBAAgB,CAAC;AACxB,OAAO,EACL,kBAAkB,GAGnB,MAAM,oBAAoB,CAAC;AAC5B,OAAO,EACL,gBAAgB,GAEjB,MAAM,kBAAkB,CAAC;AAC1B,OAAO,EACL,8BAA8B,EAC9B,mBAAmB,GAEpB,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EACL,4BAA4B,EAC5B,iBAAiB,GAElB,MAAM,oBAAoB,CAAC;AAC5B,OAAO,EAAE,6BAA6B,EAAE,oBAAoB,EAAE,MAAM,wBAAwB,CAAC;AAC7F,OAAO,EAAE,kBAAkB,EAAE,oBAAoB,EAAoB,MAAM,WAAW,CAAC;AACvF,OAAO,EACL,qBAAqB,EACrB,kBAAkB,EAClB,aAAa,EACb,YAAY,GAIb,MAAM,cAAc,CAAC;AACtB,OAAO,EACL,iBAAiB,EACjB,gBAAgB,GAEjB,MAAM,2BAA2B,CAAC;AACnC,OAAO,EAAE,0BAA0B,EAAE,sBAAsB,EAAE,sBAAsB,EAAE,MAAM,sBAAsB,CAAC;AAClH,OAAO,EAAE,gBAAgB,EAAE,MAAM,uBAAuB,CAAC;AACzD,OAAO,EAAE,kBAAkB,EAAE,kBAAkB,EAAE,MAAM,qBAAqB,CAAC;AAC7E,OAAO,EACL,kBAAkB,EAClB,0BAA0B,EAC1B,oBAAoB,EACpB,cAAc,EACd,mBAAmB,EACnB,uBAAuB,EACvB,eAAe,GAIhB,MAAM,kBAAkB,CAAC"}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/skill-test/index.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,eAAe,EACf,cAAc,EACd,eAAe,EACf,cAAc,EACd,mBAAmB,GAIpB,MAAM,kBAAkB,CAAC;AAC1B,OAAO,EACL,iBAAiB,EACjB,6BAA6B,EAC7B,kBAAkB,EAClB,iBAAiB,GAGlB,MAAM,oBAAoB,CAAC;AAC5B,OAAO,EAAE,oBAAoB,EAAE,8BAA8B,EAAwB,MAAM,gBAAgB,CAAC;AAC5G,OAAO,EACL,sBAAsB,EACtB,kBAAkB,EAClB,wBAAwB,EACxB,oBAAoB,EACpB,sBAAsB,GAGvB,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EACL,wBAAwB,EACxB,uBAAuB,EACvB,wBAAwB,GACzB,MAAM,qBAAqB,CAAC;AAC7B,OAAO,EACL,wBAAwB,EACxB,oBAAoB,EACpB,qBAAqB,EACrB,iBAAiB,GAIlB,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EACL,eAAe,EACf,iBAAiB,EACjB,gBAAgB,EAChB,gBAAgB,EAChB,gBAAgB,GAGjB,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EACL,uBAAuB,EACvB,uBAAuB,EACvB,mBAAmB,EACnB,oBAAoB,GAIrB,MAAM,qBAAqB,CAAC;AAC7B,OAAO,EACL,qBAAqB,EACrB,iBAAiB,EACjB,gBAAgB,EAChB,oBAAoB,EACpB,kBAAkB,GACnB,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EACL,oBAAoB,EACpB,iBAAiB,EACjB,oBAAoB,GAIrB,MAAM,eAAe,CAAC;AACvB,OAAO,EACL,gBAAgB,EAChB,gBAAgB,EAChB,mBAAmB,EACnB,gBAAgB,EAChB,oBAAoB,GAKrB,MAAM,mBAAmB,CAAC;AAC3B,OAAO,EACL,oBAAoB,EACpB,yBAAyB,EACzB,oBAAoB,EACpB,gBAAgB,EAChB,eAAe,EAEf,sBAAsB,GACvB,MAAM,qBAAqB,CAAC;AAC7B,OAAO,EAAE,cAAc,EAAE,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AACnE,OAAO,EAAE,mBAAmB,EAAE,oBAAoB,EAAE,MAAM,mBAAmB,CAAC;AAC9E,OAAO,EAAE,eAAe,EAAE,WAAW,EAA2B,MAAM,eAAe,CAAC;AACtF,OAAO,EACL,YAAY,GAIb,MAAM,gBAAgB,CAAC;AACxB,OAAO,EACL,kBAAkB,GAGnB,MAAM,oBAAoB,CAAC;AAC5B,OAAO,EACL,gBAAgB,GAEjB,MAAM,kBAAkB,CAAC;AAC1B,OAAO,EACL,8BAA8B,EAC9B,mBAAmB,GAEpB,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EACL,4BAA4B,EAC5B,iBAAiB,GAElB,MAAM,oBAAoB,CAAC;AAC5B,OAAO,EAAE,6BAA6B,EAAE,oBAAoB,EAAE,MAAM,wBAAwB,CAAC;AAC7F,OAAO,EAAE,kBAAkB,EAAE,oBAAoB,EAAoB,MAAM,WAAW,CAAC;AACvF,OAAO,EACL,qBAAqB,EACrB,kBAAkB,EAClB,aAAa,EACb,YAAY,GAIb,MAAM,cAAc,CAAC;AACtB,OAAO,EACL,iBAAiB,EACjB,gBAAgB,GAEjB,MAAM,2BAA2B,CAAC;AACnC,OAAO,EAAE,0BAA0B,EAAE,sBAAsB,EAAE,sBAAsB,EAAE,MAAM,sBAAsB,CAAC;AAClH,OAAO,EAAE,gBAAgB,EAAE,MAAM,uBAAuB,CAAC;AACzD,OAAO,EAAE,kBAAkB,EAAE,kBAAkB,EAAE,MAAM,qBAAqB,CAAC;AAC7E,OAAO,EACL,kBAAkB,EAClB,0BAA0B,EAC1B,oBAAoB,EACpB,cAAc,EACd,mBAAmB,EACnB,uBAAuB,EACvB,eAAe,GAIhB,MAAM,kBAAkB,CAAC"}

package/dist/skill-test/lock.d.ts

@@ -1,7 +1,36 @@
-export declare class HarnessLockBusyError extends Error {
+import { VatError } from '@vibe-agent-toolkit/utils';
+/**
+ * The harness root for a subject set is already locked.
+ *
+ * Reason `preflight`, not `internal`. This is the most user-correctable condition in
+ * the command — the remedy is "wait, or delete one file" — and it was the ONE error
+ * class in this feature with no declared reason, so it fell through
+ * `skillTestFailureReason`'s default to `internal`. The published CI recipe reads
+ * `internal` as "the harness broke, fail the build", which is exactly the wrong
+ * verdict for a lock held by the operator's own second terminal.
+ *
+ * The field below is what `skillTestFailureReason` READS — it is not decoration
+ * mirroring an `instanceof` row, which is what it was when it was added. Deleting it
+ * turns this back into `internal`.
+ *
+ * The message names `lockPath` because the lockfile is created `O_EXCL` in the
+ * DETERMINISTIC harness root and released only on normal exit or SIGINT/SIGTERM —
+ * a SIGKILL, an OOM, or a crash leaves it behind and every later run of that skill
+ * fails here, with no `--force` to escape. Until the staleness protocol (pid +
+ * timestamp + liveness probe + `--force`) lands in its own lane, naming the path
+ * and saying to delete it IS the escape hatch, so it belongs in the message rather
+ * than in a doc the operator is not currently reading.
+ */
+export declare class HarnessLockBusyError extends VatError {
+    readonly lockPath: string;
+    readonly reason: "preflight";
     constructor(lockPath: string);
 }
 export interface HarnessLock {
+    /**
+     * Release the lock. NEVER THROWS — see {@link acquireHarnessLock}. Callers run
+     * this from a cleanup path and must be able to call it bare.
+     */
     release(): void;
 }
 /**
@@ -9,6 +38,34 @@ export interface HarnessLock {
  * vat-vs-vat staging/result races for the same subject set. `wait: false`
  * fails fast; the default `wait: true` is reserved for the CLI to poll (v1
  * keeps the simple fail-fast — the CLI surfaces the busy message).
+ *
+ * ACQUIRING can throw (that is the whole point: EEXIST → {@link HarnessLockBusyError}).
+ * RELEASING cannot, and that asymmetry is deliberate. `rmSync(..., {force: true})`
+ * swallows only ENOENT, so an EPERM/EACCES/EROFS on the lockfile — a read-only
+ * volume, a `root`-owned temp dir, a Windows handle still open — threw out of the
+ * harness `finally` and REPLACED an already-good result (verdict computed, artifacts
+ * written, summary composed) with exit 1 and no summary, every artifact sitting
+ * unread on disk.
+ *
+ * The guard lives HERE rather than at the call site, where it originally went. A
+ * call-site wrapper is invisible to the compiler: reverting it to a bare
+ * `lock.release()` left the whole unit suite AND the integration suite green, so
+ * nothing but review stood between the fix and its own silent removal. Owning the
+ * contract in the definition — `release()` never throws — means there is no unwrapped
+ * call site left to lose. Note what that does NOT mean: `release(): void` encodes
+ * nothing about throwing, because TypeScript has no throws clause. The contract is
+ * held up by the two nested `try`s below and by the tests that pin each one, not by
+ * the signature.
+ *
+ * BOTH statements of the catch are guarded, and the second one is why: the warning
+ * write is itself an fd-level operation, and a synchronous EBADF/EPIPE on a file- or
+ * TTY-backed fd 2 threw straight back out — reinstating the exact failure the outer
+ * catch was added to prevent, on a path only reachable once something had already
+ * gone wrong. A test that mocks `process.stderr.write` to RETURN cannot see it.
+ *
+ * The failure is not silent: a lockfile that could not be removed will fail the NEXT
+ * run of this skill with a "busy" error, and an operator who saw nothing here has no
+ * way to connect the two.
  */
 export declare function acquireHarnessLock(harnessRoot: string, opts?: {
     wait?: boolean;