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

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

@@ -1 +1 @@
-{"version":3,"file":"lock.d.ts","sourceRoot":"","sources":["../../src/skill-test/lock.ts"],"names":[],"mappings":"AAIA,qBAAa,oBAAqB,SAAQ,KAAK;gBACjC,QAAQ,EAAE,MAAM;CAI7B;AAED,MAAM,WAAW,WAAW;IAC1B,OAAO,IAAI,IAAI,CAAC;CACjB;AAED;;;;;GAKG;AACH,wBAAgB,kBAAkB,CAAC,WAAW,EAAE,MAAM,EAAE,IAAI,GAAE;IAAE,IAAI,CAAC,EAAE,OAAO,CAAA;CAAO,GAAG,WAAW,CAwBlG;AAWD,MAAM,WAAW,2BAA2B;IAC1C;;;;OAIG;IACH,QAAQ,EAAE,MAAM,IAAI,CAAC;IACrB;;;OAGG;IACH,IAAI,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,IAAI,CAAC;CAC/B;AAED;;;;;;;;GAQG;AACH,wBAAgB,oBAAoB,CAAC,IAAI,EAAE,2BAA2B,GAAG,MAAM,IAAI,CAgBlF"}
+{"version":3,"file":"lock.d.ts","sourceRoot":"","sources":["../../src/skill-test/lock.ts"],"names":[],"mappings":"AAEA,OAAO,EAAY,QAAQ,EAAE,MAAM,2BAA2B,CAAC;AAE/D;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,qBAAa,oBAAqB,SAAQ,QAAQ;aAEpB,QAAQ,EAAE,MAAM;IAD5C,QAAQ,CAAC,MAAM,EAAG,WAAW,CAAU;gBACX,QAAQ,EAAE,MAAM;CAS7C;AAED,MAAM,WAAW,WAAW;IAC1B;;;OAGG;IACH,OAAO,IAAI,IAAI,CAAC;CACjB;AAED;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAiCG;AACH,wBAAgB,kBAAkB,CAAC,WAAW,EAAE,MAAM,EAAE,IAAI,GAAE;IAAE,IAAI,CAAC,EAAE,OAAO,CAAA;CAAO,GAAG,WAAW,CAyClG;AAWD,MAAM,WAAW,2BAA2B;IAC1C;;;;OAIG;IACH,QAAQ,EAAE,MAAM,IAAI,CAAC;IACrB;;;OAGG;IACH,IAAI,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,IAAI,CAAC;CAC/B;AAED;;;;;;;;GAQG;AACH,wBAAgB,oBAAoB,CAAC,IAAI,EAAE,2BAA2B,GAAG,MAAM,IAAI,CAgBlF"}

package/dist/skill-test/lock.js

@@ -1,9 +1,36 @@
 import { closeSync, openSync, rmSync } from 'node:fs';
-import { safePath } from '@vibe-agent-toolkit/utils';
-export class HarnessLockBusyError extends Error {
+import { safePath, 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 class HarnessLockBusyError extends VatError {
+    lockPath;
+    reason = 'preflight';
     constructor(lockPath) {
-        super(`Another vat skill test run holds the harness lock: ${lockPath}. Wait for it to finish or use a different subject set.`);
-        this.name = 'HarnessLockBusyError';
+        super('HARNESS_LOCK_BUSY', `Another vat skill test run holds the harness lock: ${lockPath}. ` +
+            'Wait for it to finish, or use a different subject set. ' +
+            'If no other run is in progress the lock is stale (a previous run was killed, ' +
+            `crashed, or ran out of memory) — delete ${lockPath} and re-run.`);
+        this.lockPath = lockPath;
     }
 }
 /**
@@ -11,6 +38,34 @@ export class HarnessLockBusyError extends Error {
  * 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 function acquireHarnessLock(harnessRoot, opts = {}) {
     const lockPath = safePath.joinUnderRoot(harnessRoot, '.vat-skill-test.lock');
@@ -19,7 +74,6 @@ export function acquireHarnessLock(harnessRoot, opts = {}) {
     let fd;
     try {
         // 'wx' = O_CREAT | O_EXCL — fails with EEXIST if the lockfile already exists.
-        // eslint-disable-next-line security/detect-non-literal-fs-filename -- our own derived harness root
         fd = openSync(lockPath, 'wx');
     }
     catch (err) {
@@ -35,7 +89,25 @@ export function acquireHarnessLock(harnessRoot, opts = {}) {
             if (released)
                 return;
             released = true;
-            rmSync(lockPath, { force: true });
+            try {
+                rmSync(lockPath, { force: true });
+            }
+            catch (err) {
+                try {
+                    process.stderr.write(`warning: could not remove the harness lockfile ${lockPath} (the run's result stands): ` +
+                        `${err instanceof Error ? err.message : String(err)}. ` +
+                        'The next run of this skill will report the lock as busy until it is deleted.\n');
+                    // eslint-disable-next-line local/no-blind-catch -- the REPORTING channel (fd 2) is what threw; there is no surface left to carry the error to, and a rethrow from a `finally` would destroy a good harness result
+                }
+                catch {
+                    // The REPORTING channel is what failed — an EBADF/EPIPE on a file- or
+                    // TTY-backed fd 2 throws synchronously out of `write`. There is nowhere
+                    // left to report it to, and rethrowing here would do exactly what the
+                    // outer catch exists to prevent: destroy a good result from the harness
+                    // `finally`. Swallowed deliberately, and it is the only swallow in this
+                    // module that has no alternative surface.
+                }
+            }
         },
     };
 }

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

@@ -1 +1 @@
-{"version":3,"file":"lock.js","sourceRoot":"","sources":["../../src/skill-test/lock.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,SAAS,EAAE,QAAQ,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AAEtD,OAAO,EAAE,QAAQ,EAAE,MAAM,2BAA2B,CAAC;AAErD,MAAM,OAAO,oBAAqB,SAAQ,KAAK;IAC7C,YAAY,QAAgB;QAC1B,KAAK,CAAC,sDAAsD,QAAQ,yDAAyD,CAAC,CAAC;QAC/H,IAAI,CAAC,IAAI,GAAG,sBAAsB,CAAC;IACrC,CAAC;CACF;AAMD;;;;;GAKG;AACH,MAAM,UAAU,kBAAkB,CAAC,WAAmB,EAAE,OAA2B,EAAE;IACnF,MAAM,QAAQ,GAAG,QAAQ,CAAC,aAAa,CAAC,WAAW,EAAE,sBAAsB,CAAC,CAAC;IAC7E,wGAAwG;IACxG,KAAK,IAAI,CAAC,IAAI,CAAC;IACf,IAAI,EAAU,CAAC;IACf,IAAI,CAAC;QACH,8EAA8E;QAC9E,mGAAmG;QACnG,EAAE,GAAG,QAAQ,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;IAChC,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,IAAK,GAA6B,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;YACrD,MAAM,IAAI,oBAAoB,CAAC,QAAQ,CAAC,CAAC;QAC3C,CAAC;QACD,MAAM,GAAG,CAAC;IACZ,CAAC;IACD,SAAS,CAAC,EAAE,CAAC,CAAC;IACd,IAAI,QAAQ,GAAG,KAAK,CAAC;IACrB,OAAO;QACL,OAAO;YACL,IAAI,QAAQ;gBAAE,OAAO;YACrB,QAAQ,GAAG,IAAI,CAAC;YAChB,MAAM,CAAC,QAAQ,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;QACpC,CAAC;KACF,CAAC;AACJ,CAAC;AAED,8EAA8E;AAC9E,MAAM,gBAAgB,GAAG,GAAG,CAAC;AAE7B,mFAAmF;AACnF,MAAM,cAAc,GAAqD;IACvE,CAAC,QAAQ,EAAE,CAAC,CAAC;IACb,CAAC,SAAS,EAAE,EAAE,CAAC;CAChB,CAAC;AAgBF;;;;;;;;GAQG;AACH,MAAM,UAAU,oBAAoB,CAAC,IAAiC;IACpE,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,IAAI,CAAC,CAAC,IAAY,EAAQ,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAC5E,MAAM,UAAU,GAAiD,EAAE,CAAC;IACpE,MAAM,MAAM,GAAG,GAAS,EAAE;QACxB,KAAK,MAAM,CAAC,GAAG,EAAE,OAAO,CAAC,IAAI,UAAU;YAAE,OAAO,CAAC,GAAG,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC;IACrE,CAAC,CAAC;IACF,KAAK,MAAM,CAAC,GAAG,EAAE,GAAG,CAAC,IAAI,cAAc,EAAE,CAAC;QACxC,MAAM,OAAO,GAAG,GAAS,EAAE;YACzB,MAAM,EAAE,CAAC;YACT,IAAI,CAAC,QAAQ,EAAE,CAAC;YAChB,IAAI,CAAC,gBAAgB,GAAG,GAAG,CAAC,CAAC;QAC/B,CAAC,CAAC;QACF,OAAO,CAAC,EAAE,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC;QACzB,UAAU,CAAC,IAAI,CAAC,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC,CAAC;IAClC,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC"}
+{"version":3,"file":"lock.js","sourceRoot":"","sources":["../../src/skill-test/lock.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,SAAS,EAAE,QAAQ,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AAEtD,OAAO,EAAE,QAAQ,EAAE,QAAQ,EAAE,MAAM,2BAA2B,CAAC;AAE/D;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,MAAM,OAAO,oBAAqB,SAAQ,QAAQ;IAEpB;IADnB,MAAM,GAAG,WAAoB,CAAC;IACvC,YAA4B,QAAgB;QAC1C,KAAK,CACH,mBAAmB,EACnB,sDAAsD,QAAQ,IAAI;YAChE,yDAAyD;YACzD,+EAA+E;YAC/E,2CAA2C,QAAQ,cAAc,CACpE,CAAC;QAPwB,aAAQ,GAAR,QAAQ,CAAQ;IAQ5C,CAAC;CACF;AAUD;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAiCG;AACH,MAAM,UAAU,kBAAkB,CAAC,WAAmB,EAAE,OAA2B,EAAE;IACnF,MAAM,QAAQ,GAAG,QAAQ,CAAC,aAAa,CAAC,WAAW,EAAE,sBAAsB,CAAC,CAAC;IAC7E,wGAAwG;IACxG,KAAK,IAAI,CAAC,IAAI,CAAC;IACf,IAAI,EAAU,CAAC;IACf,IAAI,CAAC;QACH,8EAA8E;QAC9E,EAAE,GAAG,QAAQ,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;IAChC,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,IAAK,GAA6B,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;YACrD,MAAM,IAAI,oBAAoB,CAAC,QAAQ,CAAC,CAAC;QAC3C,CAAC;QACD,MAAM,GAAG,CAAC;IACZ,CAAC;IACD,SAAS,CAAC,EAAE,CAAC,CAAC;IACd,IAAI,QAAQ,GAAG,KAAK,CAAC;IACrB,OAAO;QACL,OAAO;YACL,IAAI,QAAQ;gBAAE,OAAO;YACrB,QAAQ,GAAG,IAAI,CAAC;YAChB,IAAI,CAAC;gBACH,MAAM,CAAC,QAAQ,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;YACpC,CAAC;YAAC,OAAO,GAAG,EAAE,CAAC;gBACb,IAAI,CAAC;oBACH,OAAO,CAAC,MAAM,CAAC,KAAK,CAClB,kDAAkD,QAAQ,8BAA8B;wBACtF,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI;wBACvD,gFAAgF,CACnF,CAAC;oBACF,kNAAkN;gBACpN,CAAC;gBAAC,MAAM,CAAC;oBACP,sEAAsE;oBACtE,wEAAwE;oBACxE,sEAAsE;oBACtE,wEAAwE;oBACxE,wEAAwE;oBACxE,0CAA0C;gBAC5C,CAAC;YACH,CAAC;QACH,CAAC;KACF,CAAC;AACJ,CAAC;AAED,8EAA8E;AAC9E,MAAM,gBAAgB,GAAG,GAAG,CAAC;AAE7B,mFAAmF;AACnF,MAAM,cAAc,GAAqD;IACvE,CAAC,QAAQ,EAAE,CAAC,CAAC;IACb,CAAC,SAAS,EAAE,EAAE,CAAC;CAChB,CAAC;AAgBF;;;;;;;;GAQG;AACH,MAAM,UAAU,oBAAoB,CAAC,IAAiC;IACpE,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,IAAI,CAAC,CAAC,IAAY,EAAQ,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAC5E,MAAM,UAAU,GAAiD,EAAE,CAAC;IACpE,MAAM,MAAM,GAAG,GAAS,EAAE;QACxB,KAAK,MAAM,CAAC,GAAG,EAAE,OAAO,CAAC,IAAI,UAAU;YAAE,OAAO,CAAC,GAAG,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC;IACrE,CAAC,CAAC;IACF,KAAK,MAAM,CAAC,GAAG,EAAE,GAAG,CAAC,IAAI,cAAc,EAAE,CAAC;QACxC,MAAM,OAAO,GAAG,GAAS,EAAE;YACzB,MAAM,EAAE,CAAC;YACT,IAAI,CAAC,QAAQ,EAAE,CAAC;YAChB,IAAI,CAAC,gBAAgB,GAAG,GAAG,CAAC,CAAC;QAC/B,CAAC,CAAC;QACF,OAAO,CAAC,EAAE,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC;QACzB,UAAU,CAAC,IAAI,CAAC,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC,CAAC;IAClC,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC"}

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

@@ -1,3 +1,4 @@
+import { VatError } from '@vibe-agent-toolkit/utils';
 /**
  * Generic, pure, unit-testable bounded-parallel pipeline runner.
  *
@@ -12,7 +13,8 @@
  * tests (no real timers, no `Math.random`).
  */
 /** Thrown by a worker to back off (sleep) and retry the current item. */
-export declare class RateLimitSignal extends Error {
+export declare class RateLimitSignal extends VatError {
+    constructor(message: string);
 }
 export interface RunPipelineOptions<T, R> {
     readonly items: readonly T[];
@@ -24,6 +26,22 @@ export interface RunPipelineOptions<T, R> {
     readonly onRateLimit?: (attempt: number) => number;
     /** Injectable for tests; defaults to a real `setTimeout`-based sleep. */
     readonly sleep?: (ms: number) => Promise<void>;
+    /**
+     * What to do when an item's retry budget is SPENT, instead of failing the run.
+     *
+     * The pipeline is the only layer that knows retries are exhausted: a worker sees a
+     * `RateLimitSignal` and cannot tell whether it is the first of five or the last.
+     * That is why a control-arm rate limit still destroyed a `--baseline` run — the
+     * worker correctly rethrew the signal (it is the retry protocol, not a failure),
+     * `runItemWithRetry` rethrew the IDENTICAL class once the budget was gone, and
+     * nothing downstream could distinguish the two.
+     *
+     * Returning a result records the exhaustion as an outcome and lets the run
+     * continue; the handler may still throw for items whose failure IS fatal. Absent →
+     * rethrow, which is the historical behaviour and stays the default so a caller
+     * that has not thought about it does not silently swallow a dead item.
+     */
+    readonly onRetriesExhausted?: (item: T, index: number, error: RateLimitSignal) => Promise<R> | R;
 }
 export declare function runPipeline<T, R>(o: RunPipelineOptions<T, R>): Promise<R[]>;
 //# sourceMappingURL=pipeline.d.ts.map

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

@@ -1 +1 @@
-{"version":3,"file":"pipeline.d.ts","sourceRoot":"","sources":["../../src/skill-test/pipeline.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,yEAAyE;AACzE,qBAAa,eAAgB,SAAQ,KAAK;CAAG;AAE7C,MAAM,WAAW,kBAAkB,CAAC,CAAC,EAAE,CAAC;IACtC,QAAQ,CAAC,KAAK,EAAE,SAAS,CAAC,EAAE,CAAC;IAC7B,6DAA6D;IAC7D,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,EAAE,KAAK,EAAE,MAAM,KAAK,OAAO,CAAC,CAAC,CAAC,CAAC;IACxD;2EACuE;IACvE,QAAQ,CAAC,WAAW,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,MAAM,CAAC;IACnD,yEAAyE;IACzE,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC,EAAE,EAAE,MAAM,KAAK,OAAO,CAAC,IAAI,CAAC,CAAC;CAChD;AA8CD,wBAAsB,WAAW,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,kBAAkB,CAAC,CAAC,EAAE,CAAC,CAAC,GAAG,OAAO,CAAC,CAAC,EAAE,CAAC,CA8BjF"}
+{"version":3,"file":"pipeline.d.ts","sourceRoot":"","sources":["../../src/skill-test/pipeline.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,2BAA2B,CAAC;AAErD;;;;;;;;;;;;GAYG;AAEH,yEAAyE;AACzE,qBAAa,eAAgB,SAAQ,QAAQ;gBAC/B,OAAO,EAAE,MAAM;CAG5B;AAED,MAAM,WAAW,kBAAkB,CAAC,CAAC,EAAE,CAAC;IACtC,QAAQ,CAAC,KAAK,EAAE,SAAS,CAAC,EAAE,CAAC;IAC7B,6DAA6D;IAC7D,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,EAAE,KAAK,EAAE,MAAM,KAAK,OAAO,CAAC,CAAC,CAAC,CAAC;IACxD;2EACuE;IACvE,QAAQ,CAAC,WAAW,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,MAAM,CAAC;IACnD,yEAAyE;IACzE,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC,EAAE,EAAE,MAAM,KAAK,OAAO,CAAC,IAAI,CAAC,CAAC;IAC/C;;;;;;;;;;;;;;OAcG;IACH,QAAQ,CAAC,kBAAkB,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,eAAe,KAAK,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;CAClG;AAmDD,wBAAsB,WAAW,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,kBAAkB,CAAC,CAAC,EAAE,CAAC,CAAC,GAAG,OAAO,CAAC,CAAC,EAAE,CAAC,CAoCjF"}

package/dist/skill-test/pipeline.js

@@ -1,3 +1,4 @@
+import { VatError } from '@vibe-agent-toolkit/utils';
 /**
  * Generic, pure, unit-testable bounded-parallel pipeline runner.
  *
@@ -12,7 +13,10 @@
  * tests (no real timers, no `Math.random`).
  */
 /** Thrown by a worker to back off (sleep) and retry the current item. */
-export class RateLimitSignal extends Error {
+export class RateLimitSignal extends VatError {
+    constructor(message) {
+        super('RATE_LIMIT', message);
+    }
 }
 /** Bounded retries on RateLimitSignal before giving up and rethrowing. */
 const MAX_RATE_LIMIT_RETRIES = 5;
@@ -26,15 +30,11 @@ function defaultSleep(ms) {
         setTimeout(resolve, ms);
     });
 }
-/**
- * Runs a single item to completion, retrying on {@link RateLimitSignal} up to
- * {@link MAX_RATE_LIMIT_RETRIES} times using the provided backoff + sleep.
- */
-async function runItemWithRetry(item, index, worker, onRateLimit, sleep) {
+async function runItemWithRetry(item, index, deps) {
     let attempt = 0;
     for (;;) {
         try {
-            return await worker(item, index);
+            return await deps.worker(item, index);
         }
         catch (error) {
             if (!(error instanceof RateLimitSignal)) {
@@ -42,9 +42,14 @@ async function runItemWithRetry(item, index, worker, onRateLimit, sleep) {
             }
             attempt += 1;
             if (attempt > MAX_RATE_LIMIT_RETRIES) {
-                throw error;
+                // Budget spent. This is the ONE place that can tell an exhausted signal from
+                // a retryable one, so it is the only place that can offer the caller the
+                // choice — see {@link RunPipelineOptions.onRetriesExhausted}.
+                if (deps.onRetriesExhausted === undefined)
+                    throw error;
+                return await deps.onRetriesExhausted(item, index, error);
             }
-            await sleep(onRateLimit(attempt));
+            await deps.sleep(deps.onRateLimit(attempt));
         }
     }
 }
@@ -56,6 +61,12 @@ export async function runPipeline(o) {
         throw new Error(`runPipeline: concurrency must be an integer >= 1 (got ${concurrency})`);
     }
     const results = new Array(items.length);
+    const deps = {
+        worker,
+        onRateLimit,
+        sleep,
+        ...(o.onRetriesExhausted === undefined ? {} : { onRetriesExhausted: o.onRetriesExhausted }),
+    };
     let nextIndex = 0;
     async function runWorkerLoop() {
         for (;;) {
@@ -66,7 +77,7 @@ export async function runPipeline(o) {
             }
             // Non-null: index < items.length was just checked above.
             const item = items[index];
-            results[index] = await runItemWithRetry(item, index, worker, onRateLimit, sleep);
+            results[index] = await runItemWithRetry(item, index, deps);
         }
     }
     const poolSize = Math.min(concurrency, items.length);

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

@@ -1 +1 @@
-{"version":3,"file":"pipeline.js","sourceRoot":"","sources":["../../src/skill-test/pipeline.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,yEAAyE;AACzE,MAAM,OAAO,eAAgB,SAAQ,KAAK;CAAG;AAc7C,0EAA0E;AAC1E,MAAM,sBAAsB,GAAG,CAAC,CAAC;AAEjC,MAAM,uBAAuB,GAAG,IAAI,CAAC;AACrC,MAAM,sBAAsB,GAAG,MAAM,CAAC;AAEtC,SAAS,kBAAkB,CAAC,OAAe;IACzC,OAAO,IAAI,CAAC,GAAG,CAAC,sBAAsB,EAAE,uBAAuB,GAAG,CAAC,IAAI,CAAC,OAAO,GAAG,CAAC,CAAC,CAAC,CAAC;AACxF,CAAC;AAED,SAAS,YAAY,CAAC,EAAU;IAC9B,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE;QAC7B,UAAU,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;IAC1B,CAAC,CAAC,CAAC;AACL,CAAC;AAED;;;GAGG;AACH,KAAK,UAAU,gBAAgB,CAC7B,IAAO,EACP,KAAa,EACb,MAA8C,EAC9C,WAAwC,EACxC,KAAoC;IAEpC,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,SAAS,CAAC;QACR,IAAI,CAAC;YACH,OAAO,MAAM,MAAM,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;QACnC,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,IAAI,CAAC,CAAC,KAAK,YAAY,eAAe,CAAC,EAAE,CAAC;gBACxC,MAAM,KAAK,CAAC;YACd,CAAC;YACD,OAAO,IAAI,CAAC,CAAC;YACb,IAAI,OAAO,GAAG,sBAAsB,EAAE,CAAC;gBACrC,MAAM,KAAK,CAAC;YACd,CAAC;YACD,MAAM,KAAK,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC;QACpC,CAAC;IACH,CAAC;AACH,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,WAAW,CAAO,CAA2B;IACjE,MAAM,EAAE,KAAK,EAAE,WAAW,EAAE,MAAM,EAAE,GAAG,CAAC,CAAC;IACzC,MAAM,WAAW,GAAG,CAAC,CAAC,WAAW,IAAI,kBAAkB,CAAC;IACxD,MAAM,KAAK,GAAG,CAAC,CAAC,KAAK,IAAI,YAAY,CAAC;IAEtC,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,WAAW,CAAC,IAAI,WAAW,GAAG,CAAC,EAAE,CAAC;QACtD,MAAM,IAAI,KAAK,CAAC,yDAAyD,WAAW,GAAG,CAAC,CAAC;IAC3F,CAAC;IAED,MAAM,OAAO,GAAQ,IAAI,KAAK,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;IAC7C,IAAI,SAAS,GAAG,CAAC,CAAC;IAElB,KAAK,UAAU,aAAa;QAC1B,SAAS,CAAC;YACR,MAAM,KAAK,GAAG,SAAS,CAAC;YACxB,SAAS,IAAI,CAAC,CAAC;YACf,IAAI,KAAK,IAAI,KAAK,CAAC,MAAM,EAAE,CAAC;gBAC1B,OAAO;YACT,CAAC;YACD,yDAAyD;YACzD,MAAM,IAAI,GAAG,KAAK,CAAC,KAAK,CAAM,CAAC;YAC/B,OAAO,CAAC,KAAK,CAAC,GAAG,MAAM,gBAAgB,CAAC,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,WAAW,EAAE,KAAK,CAAC,CAAC;QACnF,CAAC;IACH,CAAC;IAED,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,CAAC,WAAW,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;IACrD,MAAM,IAAI,GAAG,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,QAAQ,EAAE,EAAE,GAAG,EAAE,CAAC,aAAa,EAAE,CAAC,CAAC;IACrE,MAAM,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IAExB,OAAO,OAAO,CAAC;AACjB,CAAC"}
+{"version":3,"file":"pipeline.js","sourceRoot":"","sources":["../../src/skill-test/pipeline.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,2BAA2B,CAAC;AAErD;;;;;;;;;;;;GAYG;AAEH,yEAAyE;AACzE,MAAM,OAAO,eAAgB,SAAQ,QAAQ;IAC3C,YAAY,OAAe;QACzB,KAAK,CAAC,YAAY,EAAE,OAAO,CAAC,CAAC;IAC/B,CAAC;CACF;AA8BD,0EAA0E;AAC1E,MAAM,sBAAsB,GAAG,CAAC,CAAC;AAEjC,MAAM,uBAAuB,GAAG,IAAI,CAAC;AACrC,MAAM,sBAAsB,GAAG,MAAM,CAAC;AAEtC,SAAS,kBAAkB,CAAC,OAAe;IACzC,OAAO,IAAI,CAAC,GAAG,CAAC,sBAAsB,EAAE,uBAAuB,GAAG,CAAC,IAAI,CAAC,OAAO,GAAG,CAAC,CAAC,CAAC,CAAC;AACxF,CAAC;AAED,SAAS,YAAY,CAAC,EAAU;IAC9B,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE;QAC7B,UAAU,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;IAC1B,CAAC,CAAC,CAAC;AACL,CAAC;AAaD,KAAK,UAAU,gBAAgB,CAAO,IAAO,EAAE,KAAa,EAAE,IAAuB;IACnF,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,SAAS,CAAC;QACR,IAAI,CAAC;YACH,OAAO,MAAM,IAAI,CAAC,MAAM,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;QACxC,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,IAAI,CAAC,CAAC,KAAK,YAAY,eAAe,CAAC,EAAE,CAAC;gBACxC,MAAM,KAAK,CAAC;YACd,CAAC;YACD,OAAO,IAAI,CAAC,CAAC;YACb,IAAI,OAAO,GAAG,sBAAsB,EAAE,CAAC;gBACrC,6EAA6E;gBAC7E,yEAAyE;gBACzE,8DAA8D;gBAC9D,IAAI,IAAI,CAAC,kBAAkB,KAAK,SAAS;oBAAE,MAAM,KAAK,CAAC;gBACvD,OAAO,MAAM,IAAI,CAAC,kBAAkB,CAAC,IAAI,EAAE,KAAK,EAAE,KAAK,CAAC,CAAC;YAC3D,CAAC;YACD,MAAM,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC;QAC9C,CAAC;IACH,CAAC;AACH,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,WAAW,CAAO,CAA2B;IACjE,MAAM,EAAE,KAAK,EAAE,WAAW,EAAE,MAAM,EAAE,GAAG,CAAC,CAAC;IACzC,MAAM,WAAW,GAAG,CAAC,CAAC,WAAW,IAAI,kBAAkB,CAAC;IACxD,MAAM,KAAK,GAAG,CAAC,CAAC,KAAK,IAAI,YAAY,CAAC;IAEtC,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,WAAW,CAAC,IAAI,WAAW,GAAG,CAAC,EAAE,CAAC;QACtD,MAAM,IAAI,KAAK,CAAC,yDAAyD,WAAW,GAAG,CAAC,CAAC;IAC3F,CAAC;IAED,MAAM,OAAO,GAAQ,IAAI,KAAK,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;IAC7C,MAAM,IAAI,GAAsB;QAC9B,MAAM;QACN,WAAW;QACX,KAAK;QACL,GAAG,CAAC,CAAC,CAAC,kBAAkB,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,kBAAkB,EAAE,CAAC,CAAC,kBAAkB,EAAE,CAAC;KAC5F,CAAC;IACF,IAAI,SAAS,GAAG,CAAC,CAAC;IAElB,KAAK,UAAU,aAAa;QAC1B,SAAS,CAAC;YACR,MAAM,KAAK,GAAG,SAAS,CAAC;YACxB,SAAS,IAAI,CAAC,CAAC;YACf,IAAI,KAAK,IAAI,KAAK,CAAC,MAAM,EAAE,CAAC;gBAC1B,OAAO;YACT,CAAC;YACD,yDAAyD;YACzD,MAAM,IAAI,GAAG,KAAK,CAAC,KAAK,CAAM,CAAC;YAC/B,OAAO,CAAC,KAAK,CAAC,GAAG,MAAM,gBAAgB,CAAC,IAAI,EAAE,KAAK,EAAE,IAAI,CAAC,CAAC;QAC7D,CAAC;IACH,CAAC;IAED,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,CAAC,WAAW,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;IACrD,MAAM,IAAI,GAAG,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,QAAQ,EAAE,EAAE,GAAG,EAAE,CAAC,aAAa,EAAE,CAAC,CAAC;IACrE,MAAM,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IAExB,OAAO,OAAO,CAAC;AACjB,CAAC"}

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

@@ -1,4 +1,4 @@
-import { type AuthMechanism, type AuthMode, type AuthStatusProbe, type ResolvedAuth } from '@vibe-agent-toolkit/utils';
+import { type AuthMechanism, type AuthMode, type AuthStatusProbe, type ResolvedAuth } from '@vibe-agent-toolkit/utils/skill-test';
 export interface PreflightCheck {
     name: string;
     passed: boolean;

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

@@ -1 +1 @@
-{"version":3,"file":"preflight.d.ts","sourceRoot":"","sources":["../../src/skill-test/preflight.ts"],"names":[],"mappings":"AAEA,OAAO,EAGL,KAAK,aAAa,EAClB,KAAK,QAAQ,EACb,KAAK,eAAe,EACpB,KAAK,YAAY,EAClB,MAAM,2BAA2B,CAAC;AAEnC,MAAM,WAAW,cAAc;IAC7B,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,OAAO,CAAC;IAChB,OAAO,EAAE,MAAM,CAAC;IAChB,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,cAAc;IAC7B,kBAAkB,EAAE,MAAM,MAAM,GAAG,IAAI,CAAC;IACxC,cAAc,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,OAAO,CAAC;IAC1C,SAAS,EAAE,eAAe,CAAC;IAC3B,cAAc,EAAE,MAAM,EAAE,CAAC;IACzB,eAAe,EAAE,MAAM,EAAE,CAAC;IAC1B,WAAW,EAAE,MAAM,OAAO,CAAC;IAC3B,YAAY,EAAE;QAAE,SAAS,EAAE,MAAM,CAAC;QAAC,cAAc,EAAE,MAAM,CAAC;QAAC,YAAY,EAAE,MAAM,CAAC;QAAC,YAAY,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;IACzG,QAAQ,EAAE,QAAQ,CAAC;IACnB,WAAW,CAAC,EAAE,aAAa,CAAC;IAC5B,SAAS,EAAE,MAAM,CAAC,UAAU,CAAC;CAC9B;AAED,MAAM,WAAW,eAAe;IAC9B,MAAM,EAAE,cAAc,EAAE,CAAC;IACzB,MAAM,EAAE,OAAO,CAAC;IAChB,YAAY,EAAE,YAAY,GAAG,IAAI,CAAC;CACnC;AAeD,wBAAgB,YAAY,CAAC,KAAK,EAAE,cAAc,GAAG,eAAe,CA4DnE"}
+{"version":3,"file":"preflight.d.ts","sourceRoot":"","sources":["../../src/skill-test/preflight.ts"],"names":[],"mappings":"AAEA,OAAO,EAGL,KAAK,aAAa,EAClB,KAAK,QAAQ,EACb,KAAK,eAAe,EACpB,KAAK,YAAY,EAClB,MAAM,sCAAsC,CAAC;AAE9C,MAAM,WAAW,cAAc;IAC7B,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,OAAO,CAAC;IAChB,OAAO,EAAE,MAAM,CAAC;IAChB,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,cAAc;IAC7B,kBAAkB,EAAE,MAAM,MAAM,GAAG,IAAI,CAAC;IACxC,cAAc,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,OAAO,CAAC;IAC1C,SAAS,EAAE,eAAe,CAAC;IAC3B,cAAc,EAAE,MAAM,EAAE,CAAC;IACzB,eAAe,EAAE,MAAM,EAAE,CAAC;IAC1B,WAAW,EAAE,MAAM,OAAO,CAAC;IAC3B,YAAY,EAAE;QAAE,SAAS,EAAE,MAAM,CAAC;QAAC,cAAc,EAAE,MAAM,CAAC;QAAC,YAAY,EAAE,MAAM,CAAC;QAAC,YAAY,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;IACzG,QAAQ,EAAE,QAAQ,CAAC;IACnB,WAAW,CAAC,EAAE,aAAa,CAAC;IAC5B,SAAS,EAAE,MAAM,CAAC,UAAU,CAAC;CAC9B;AAED,MAAM,WAAW,eAAe;IAC9B,MAAM,EAAE,cAAc,EAAE,CAAC;IACzB,MAAM,EAAE,OAAO,CAAC;IAChB,YAAY,EAAE,YAAY,GAAG,IAAI,CAAC;CACnC;AAyED,wBAAgB,YAAY,CAAC,KAAK,EAAE,cAAc,GAAG,eAAe,CAgDnE"}

package/dist/skill-test/preflight.js

@@ -1,34 +1,77 @@
 import { existsSync } from 'node:fs';
-import { AuthPreflightError, resolveAuth, } from '@vibe-agent-toolkit/utils';
+import { AuthPreflightError, resolveAuth, } from '@vibe-agent-toolkit/utils/skill-test';
+/**
+ * Flags vat's spawn argv depends on that `claude --help` DOES document, so their
+ * absence is a real, detectable incompatibility and a hard preflight failure.
+ *
+ * `--no-session-persistence` is here for integrity, not convenience: without it
+ * Claude Code writes each headless session — grading nonce, answer key, the whole
+ * transcript — to `$CLAUDE_CONFIG_DIR/projects/`, which vat hands the untrusted
+ * child. See {@link assembleClaudeArgs}. A `claude` that cannot suppress that
+ * cannot run these evals honestly, so it fails closed rather than silently
+ * persisting.
+ */
 const REQUIRED_FLAGS = [
     '--plugin-dir', '--setting-sources', '--output-format',
-    '--permission-mode', '--max-turns', '--max-budget-usd',
+    '--permission-mode', '--max-budget-usd', '--no-session-persistence',
 ];
+/**
+ * Flags vat passes that are functional but NOT listed in `claude --help`.
+ *
+ * `--max-turns` works and its absence would be load-bearing, but it is
+ * undocumented in 2.x — so a help-text probe reports it missing on a claude that
+ * supports it perfectly. Reporting these as verified would be a lie in the other
+ * direction, so they are surfaced as unverifiable and never fail the run.
+ *
+ * Keep this list SHORT and re-check it when the vendor documents a flag: an entry
+ * here is a check vat is choosing not to make.
+ */
+const UNVERIFIABLE_FLAGS = ['--max-turns'];
 function checkExists(label, paths) {
-    // eslint-disable-next-line security/detect-non-literal-fs-filename -- our own resolved absolute paths
     const missing = paths.filter(p => !existsSync(p));
     return missing.length === 0
         ? { name: label, passed: true, message: `all ${paths.length} present` }
         : { name: label, passed: false, message: `missing: ${missing.join(', ')}`, suggestion: 'These are declared but absent (exit 2).' };
 }
+/**
+ * Why a missing flag stops the run. `--no-session-persistence` gets its own text
+ * because its absence is not a compatibility inconvenience — it puts the grading
+ * nonce and the eval answer key on a disk the skill under test can read.
+ */
+function upgradeSuggestionFor(flag) {
+    return flag === '--no-session-persistence'
+        ? 'Upgrade Claude Code. Without --no-session-persistence the harness cannot keep the grading nonce and answer key off a disk the skill under test can read, so it refuses to run.'
+        : 'Upgrade Claude Code — this flag is part of the harness spawn and has no fallback.';
+}
+/**
+ * One check per flag the spawn argv carries: gated for the ones `claude --help`
+ * documents, informational for the ones it does not (see {@link UNVERIFIABLE_FLAGS}).
+ * Both are REPORTED — the probe this replaced said "supported" about every flag
+ * without being able to tell, so silence about a flag is exactly the failure mode.
+ */
+function flagChecks(probe) {
+    const gated = REQUIRED_FLAGS.map((flag) => {
+        const ok = probe(flag);
+        return {
+            name: `flag ${flag}`,
+            passed: ok,
+            message: ok ? 'supported' : 'NOT supported by this claude',
+            ...(ok ? {} : { suggestion: upgradeSuggestionFor(flag) }),
+        };
+    });
+    const informational = UNVERIFIABLE_FLAGS.map((flag) => ({
+        name: `flag ${flag}`,
+        passed: true,
+        message: 'passed but undocumented — support not verifiable from --help',
+    }));
+    return [...gated, ...informational];
+}
 export function runPreflight(input) {
     const checks = [];
     const version = input.claudeVersionProbe();
     checks.push(version
         ? { name: 'claude binary', passed: true, message: version }
-        : { name: 'claude binary', passed: false, message: 'not reachable', suggestion: 'Install Claude Code CLI.' });
-    for (const flag of REQUIRED_FLAGS) {
-        const ok = input.flagParseProbe(flag);
-        checks.push({
-            name: `flag ${flag}`,
-            passed: ok,
-            message: ok ? 'supported' : 'NOT supported by this claude',
-            ...(flag === '--max-turns' && !ok
-                ? { suggestion: '--max-turns is functional-but-undocumented in 2.x; its absence is load-bearing.' }
-                : {}),
-        });
-    }
-    checks.push(input.integrityOk()
+        : { name: 'claude binary', passed: false, message: 'not reachable', suggestion: 'Install Claude Code CLI.' }, ...flagChecks(input.flagParseProbe), input.integrityOk()
         ? { name: 'vendored skill-creator integrity', passed: true, message: 'manifest verified' }
         : { name: 'vendored skill-creator integrity', passed: false, message: 'hash manifest mismatch', suggestion: 'Re-sync the vendored copy.' }, checkExists('eval input files', input.evalInputPaths), checkExists('declared dependencies', input.declaredDepDirs));
     let resolvedAuth = null;

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

@@ -1 +1 @@
-{"version":3,"file":"preflight.js","sourceRoot":"","sources":["../../src/skill-test/preflight.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AAErC,OAAO,EACL,kBAAkB,EAClB,WAAW,GAKZ,MAAM,2BAA2B,CAAC;AA4BnC,MAAM,cAAc,GAAG;IACrB,cAAc,EAAE,mBAAmB,EAAE,iBAAiB;IACtD,mBAAmB,EAAE,aAAa,EAAE,kBAAkB;CAC9C,CAAC;AAEX,SAAS,WAAW,CAAC,KAAa,EAAE,KAAe;IACjD,sGAAsG;IACtG,MAAM,OAAO,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC;IAClD,OAAO,OAAO,CAAC,MAAM,KAAK,CAAC;QACzB,CAAC,CAAC,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,KAAK,CAAC,MAAM,UAAU,EAAE;QACvE,CAAC,CAAC,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,YAAY,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,EAAE,UAAU,EAAE,yCAAyC,EAAE,CAAC;AACvI,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,KAAqB;IAChD,MAAM,MAAM,GAAqB,EAAE,CAAC;IAEpC,MAAM,OAAO,GAAG,KAAK,CAAC,kBAAkB,EAAE,CAAC;IAC3C,MAAM,CAAC,IAAI,CAAC,OAAO;QACjB,CAAC,CAAC,EAAE,IAAI,EAAE,eAAe,EAAE,MAAM,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE;QAC3D,CAAC,CAAC,EAAE,IAAI,EAAE,eAAe,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,eAAe,EAAE,UAAU,EAAE,0BAA0B,EAAE,CAAC,CAAC;IAEhH,KAAK,MAAM,IAAI,IAAI,cAAc,EAAE,CAAC;QAClC,MAAM,EAAE,GAAG,KAAK,CAAC,cAAc,CAAC,IAAI,CAAC,CAAC;QACtC,MAAM,CAAC,IAAI,CAAC;YACV,IAAI,EAAE,QAAQ,IAAI,EAAE;YACpB,MAAM,EAAE,EAAE;YACV,OAAO,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,8BAA8B;YAC1D,GAAG,CAAC,IAAI,KAAK,aAAa,IAAI,CAAC,EAAE;gBAC/B,CAAC,CAAC,EAAE,UAAU,EAAE,iFAAiF,EAAE;gBACnG,CAAC,CAAC,EAAE,CAAC;SACR,CAAC,CAAC;IACL,CAAC;IAED,MAAM,CAAC,IAAI,CACT,KAAK,CAAC,WAAW,EAAE;QACjB,CAAC,CAAC,EAAE,IAAI,EAAE,kCAAkC,EAAE,MAAM,EAAE,IAAI,EAAE,OAAO,EAAE,mBAAmB,EAAE;QAC1F,CAAC,CAAC,EAAE,IAAI,EAAE,kCAAkC,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,wBAAwB,EAAE,UAAU,EAAE,4BAA4B,EAAE,EAC5I,WAAW,CAAC,kBAAkB,EAAE,KAAK,CAAC,cAAc,CAAC,EACrD,WAAW,CAAC,uBAAuB,EAAE,KAAK,CAAC,eAAe,CAAC,CAC5D,CAAC;IAEF,IAAI,YAAY,GAAwB,IAAI,CAAC;IAC7C,IAAI,CAAC;QACH,YAAY,GAAG,WAAW,CAAC;YACzB,IAAI,EAAE,KAAK,CAAC,QAAQ;YACpB,GAAG,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,KAAK,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YAChE,SAAS,EAAE,KAAK,CAAC,SAAS;YAC1B,KAAK,EAAE,KAAK,CAAC,SAAS;SACvB,CAAC,CAAC;QACH,MAAM,UAAU,GAAG,YAAY,CAAC,YAAY,CAAC,CAAC,CAAC,kBAAkB,YAAY,CAAC,YAAY,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC;QACnG,MAAM,cAAc,GAAG,YAAY,CAAC,UAAU,CAAC,CAAC,CAAC,eAAe,YAAY,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC/F,MAAM,WAAW,GAAG,aAAa,YAAY,CAAC,kBAAkB,GAAG,UAAU,GAAG,cAAc,EAAE,CAAC;QACjG,MAAM,CAAC,IAAI,CAAC;YACV,IAAI,EAAE,MAAM;YACZ,MAAM,EAAE,IAAI;YACZ,OAAO,EAAE,WAAW;SACrB,CAAC,CAAC;IACL,CAAC;IAAC,OAAO,CAAC,EAAE,CAAC;QACX,MAAM,GAAG,GAAG,CAAC,YAAY,kBAAkB,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;QACpE,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,GAAG,EAAE,UAAU,EAAE,qDAAqD,EAAE,CAAC,CAAC;IAChI,CAAC;IAED,MAAM,EAAE,SAAS,EAAE,cAAc,EAAE,YAAY,EAAE,YAAY,EAAE,GAAG,KAAK,CAAC,YAAY,CAAC;IACrF,MAAM,eAAe,GAAG,SAAS,GAAG,cAAc,GAAG,YAAY,CAAC;IAClE,MAAM,UAAU,GAAG,YAAY,CAAC,CAAC,CAAC,aAAa,YAAY,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;IACnE,MAAM,WAAW,GAAG,GAAG,SAAS,YAAY,cAAc,aAAa,YAAY,WAAW,eAAe,cAAc,UAAU,EAAE,CAAC;IACxI,MAAM,CAAC,IAAI,CAAC;QACV,IAAI,EAAE,eAAe;QACrB,MAAM,EAAE,IAAI;QACZ,OAAO,EAAE,WAAW;KACrB,CAAC,CAAC;IAEH,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,EAAE,YAAY,EAAE,CAAC;AACvE,CAAC"}
+{"version":3,"file":"preflight.js","sourceRoot":"","sources":["../../src/skill-test/preflight.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AAErC,OAAO,EACL,kBAAkB,EAClB,WAAW,GAKZ,MAAM,sCAAsC,CAAC;AA4B9C;;;;;;;;;;GAUG;AACH,MAAM,cAAc,GAAG;IACrB,cAAc,EAAE,mBAAmB,EAAE,iBAAiB;IACtD,mBAAmB,EAAE,kBAAkB,EAAE,0BAA0B;CAC3D,CAAC;AAEX;;;;;;;;;;GAUG;AACH,MAAM,kBAAkB,GAAG,CAAC,aAAa,CAAU,CAAC;AAEpD,SAAS,WAAW,CAAC,KAAa,EAAE,KAAe;IACjD,MAAM,OAAO,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC;IAClD,OAAO,OAAO,CAAC,MAAM,KAAK,CAAC;QACzB,CAAC,CAAC,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,KAAK,CAAC,MAAM,UAAU,EAAE;QACvE,CAAC,CAAC,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,YAAY,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,EAAE,UAAU,EAAE,yCAAyC,EAAE,CAAC;AACvI,CAAC;AAED;;;;GAIG;AACH,SAAS,oBAAoB,CAAC,IAAY;IACxC,OAAO,IAAI,KAAK,0BAA0B;QACxC,CAAC,CAAC,gLAAgL;QAClL,CAAC,CAAC,mFAAmF,CAAC;AAC1F,CAAC;AAED;;;;;GAKG;AACH,SAAS,UAAU,CAAC,KAAgC;IAClD,MAAM,KAAK,GAAG,cAAc,CAAC,GAAG,CAAC,CAAC,IAAI,EAAkB,EAAE;QACxD,MAAM,EAAE,GAAG,KAAK,CAAC,IAAI,CAAC,CAAC;QACvB,OAAO;YACL,IAAI,EAAE,QAAQ,IAAI,EAAE;YACpB,MAAM,EAAE,EAAE;YACV,OAAO,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,8BAA8B;YAC1D,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,oBAAoB,CAAC,IAAI,CAAC,EAAE,CAAC;SAC1D,CAAC;IACJ,CAAC,CAAC,CAAC;IACH,MAAM,aAAa,GAAG,kBAAkB,CAAC,GAAG,CAAC,CAAC,IAAI,EAAkB,EAAE,CAAC,CAAC;QACtE,IAAI,EAAE,QAAQ,IAAI,EAAE;QACpB,MAAM,EAAE,IAAI;QACZ,OAAO,EAAE,8DAA8D;KACxE,CAAC,CAAC,CAAC;IACJ,OAAO,CAAC,GAAG,KAAK,EAAE,GAAG,aAAa,CAAC,CAAC;AACtC,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,KAAqB;IAChD,MAAM,MAAM,GAAqB,EAAE,CAAC;IAEpC,MAAM,OAAO,GAAG,KAAK,CAAC,kBAAkB,EAAE,CAAC;IAC3C,MAAM,CAAC,IAAI,CACT,OAAO;QACL,CAAC,CAAC,EAAE,IAAI,EAAE,eAAe,EAAE,MAAM,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE;QAC3D,CAAC,CAAC,EAAE,IAAI,EAAE,eAAe,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,eAAe,EAAE,UAAU,EAAE,0BAA0B,EAAE,EAC9G,GAAG,UAAU,CAAC,KAAK,CAAC,cAAc,CAAC,EACnC,KAAK,CAAC,WAAW,EAAE;QACjB,CAAC,CAAC,EAAE,IAAI,EAAE,kCAAkC,EAAE,MAAM,EAAE,IAAI,EAAE,OAAO,EAAE,mBAAmB,EAAE;QAC1F,CAAC,CAAC,EAAE,IAAI,EAAE,kCAAkC,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,wBAAwB,EAAE,UAAU,EAAE,4BAA4B,EAAE,EAC5I,WAAW,CAAC,kBAAkB,EAAE,KAAK,CAAC,cAAc,CAAC,EACrD,WAAW,CAAC,uBAAuB,EAAE,KAAK,CAAC,eAAe,CAAC,CAC5D,CAAC;IAEF,IAAI,YAAY,GAAwB,IAAI,CAAC;IAC7C,IAAI,CAAC;QACH,YAAY,GAAG,WAAW,CAAC;YACzB,IAAI,EAAE,KAAK,CAAC,QAAQ;YACpB,GAAG,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,KAAK,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YAChE,SAAS,EAAE,KAAK,CAAC,SAAS;YAC1B,KAAK,EAAE,KAAK,CAAC,SAAS;SACvB,CAAC,CAAC;QACH,MAAM,UAAU,GAAG,YAAY,CAAC,YAAY,CAAC,CAAC,CAAC,kBAAkB,YAAY,CAAC,YAAY,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC;QACnG,MAAM,cAAc,GAAG,YAAY,CAAC,UAAU,CAAC,CAAC,CAAC,eAAe,YAAY,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC/F,MAAM,WAAW,GAAG,aAAa,YAAY,CAAC,kBAAkB,GAAG,UAAU,GAAG,cAAc,EAAE,CAAC;QACjG,MAAM,CAAC,IAAI,CAAC;YACV,IAAI,EAAE,MAAM;YACZ,MAAM,EAAE,IAAI;YACZ,OAAO,EAAE,WAAW;SACrB,CAAC,CAAC;IACL,CAAC;IAAC,OAAO,CAAC,EAAE,CAAC;QACX,MAAM,GAAG,GAAG,CAAC,YAAY,kBAAkB,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;QACpE,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,GAAG,EAAE,UAAU,EAAE,qDAAqD,EAAE,CAAC,CAAC;IAChI,CAAC;IAED,MAAM,EAAE,SAAS,EAAE,cAAc,EAAE,YAAY,EAAE,YAAY,EAAE,GAAG,KAAK,CAAC,YAAY,CAAC;IACrF,MAAM,eAAe,GAAG,SAAS,GAAG,cAAc,GAAG,YAAY,CAAC;IAClE,MAAM,UAAU,GAAG,YAAY,CAAC,CAAC,CAAC,aAAa,YAAY,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;IACnE,MAAM,WAAW,GAAG,GAAG,SAAS,YAAY,cAAc,aAAa,YAAY,WAAW,eAAe,cAAc,UAAU,EAAE,CAAC;IACxI,MAAM,CAAC,IAAI,CAAC;QACV,IAAI,EAAE,eAAe;QACrB,MAAM,EAAE,IAAI;QACZ,OAAO,EAAE,WAAW;KACrB,CAAC,CAAC;IAEH,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,EAAE,YAAY,EAAE,CAAC;AACvE,CAAC"}

package/dist/skill-test/prompt-invariants.d.ts

@@ -1,14 +1,16 @@
+import { VatError } from '@vibe-agent-toolkit/utils';
 /**
  * Thrown when a vat-generated prompt (executor or grader) violates one of its
  * required invariants — a missing required directive, or the presence of a
  * forbidden phrase (e.g. an executor "blinding breaker").
  *
  * Shared across the `*-prompt.ts` modules in `skill-test/` so
- * `exit-codes.ts`'s `mapErrorToExitCode` maps all of them to the same
+ * `failure-reason.ts`'s `skillTestFailureReason` maps all of them to the same
  * user-correctable preflight exit code (2), regardless of which prompt
  * builder raised it.
  */
-export declare class PromptInvariantError extends Error {
+export declare class PromptInvariantError extends VatError {
+    readonly reason: "preflight";
     constructor(message: string);
 }
 /**

package/dist/skill-test/prompt-invariants.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"prompt-invariants.d.ts","sourceRoot":"","sources":["../../src/skill-test/prompt-invariants.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AACH,qBAAa,oBAAqB,SAAQ,KAAK;gBACjC,OAAO,EAAE,MAAM;CAI5B;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,6BAA6B,CAAC,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,GAAG,MAAM,CAQnF"}
+{"version":3,"file":"prompt-invariants.d.ts","sourceRoot":"","sources":["../../src/skill-test/prompt-invariants.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,2BAA2B,CAAC;AAErD;;;;;;;;;GASG;AACH,qBAAa,oBAAqB,SAAQ,QAAQ;IAChD,QAAQ,CAAC,MAAM,EAAG,WAAW,CAAU;gBAC3B,OAAO,EAAE,MAAM;CAG5B;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,6BAA6B,CAAC,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,GAAG,MAAM,CAQnF"}

package/dist/skill-test/prompt-invariants.js

@@ -1,17 +1,18 @@
+import { VatError } from '@vibe-agent-toolkit/utils';
 /**
  * Thrown when a vat-generated prompt (executor or grader) violates one of its
  * required invariants — a missing required directive, or the presence of a
  * forbidden phrase (e.g. an executor "blinding breaker").
  *
  * Shared across the `*-prompt.ts` modules in `skill-test/` so
- * `exit-codes.ts`'s `mapErrorToExitCode` maps all of them to the same
+ * `failure-reason.ts`'s `skillTestFailureReason` maps all of them to the same
  * user-correctable preflight exit code (2), regardless of which prompt
  * builder raised it.
  */
-export class PromptInvariantError extends Error {
+export class PromptInvariantError extends VatError {
+    reason = 'preflight';
     constructor(message) {
-        super(`Prompt invariant violated: ${message}`);
-        this.name = 'PromptInvariantError';
+        super('PROMPT_INVARIANT', `Prompt invariant violated: ${message}`);
     }
 }
 /**

package/dist/skill-test/prompt-invariants.js.map

@@ -1 +1 @@
-{"version":3,"file":"prompt-invariants.js","sourceRoot":"","sources":["../../src/skill-test/prompt-invariants.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AACH,MAAM,OAAO,oBAAqB,SAAQ,KAAK;IAC7C,YAAY,OAAe;QACzB,KAAK,CAAC,8BAA8B,OAAO,EAAE,CAAC,CAAC;QAC/C,IAAI,CAAC,IAAI,GAAG,sBAAsB,CAAC;IACrC,CAAC;CACF;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,6BAA6B,CAAC,MAAc,EAAE,KAAa;IACzE,OAAO;QACL,MAAM;QACN,EAAE;QACF,2EAA2E;QAC3E,6DAA6D,KAAK,EAAE;QACpE,+EAA+E;KAChF,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACf,CAAC"}
+{"version":3,"file":"prompt-invariants.js","sourceRoot":"","sources":["../../src/skill-test/prompt-invariants.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,2BAA2B,CAAC;AAErD;;;;;;;;;GASG;AACH,MAAM,OAAO,oBAAqB,SAAQ,QAAQ;IACvC,MAAM,GAAG,WAAoB,CAAC;IACvC,YAAY,OAAe;QACzB,KAAK,CAAC,kBAAkB,EAAE,8BAA8B,OAAO,EAAE,CAAC,CAAC;IACrE,CAAC;CACF;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,6BAA6B,CAAC,MAAc,EAAE,KAAa;IACzE,OAAO;QACL,MAAM;QACN,EAAE;QACF,2EAA2E;QAC3E,6DAA6D,KAAK,EAAE;QACpE,+EAA+E;KAChF,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACf,CAAC"}