@vibe-agent-toolkit/agent-skills 0.1.40 → 0.1.41-rc.2

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,sBAAsB,EACtB,kBAAkB,EAClB,wBAAwB,EACxB,oBAAoB,EACpB,sBAAsB,GAGvB,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EACL,kBAAkB,EAClB,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,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,EACL,YAAY,GAIb,MAAM,gBAAgB,CAAC;AACxB,OAAO,EACL,6BAA6B,EAC7B,sBAAsB,EACtB,uBAAuB,EACvB,2BAA2B,EAC3B,oBAAoB,EACpB,WAAW,EACX,0BAA0B,GAE3B,MAAM,0BAA0B,CAAC;AAClC,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,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,uBAAuB,EACvB,oBAAoB,EACpB,oBAAoB,EACpB,cAAc,EACd,mBAAmB,EACnB,uBAAuB,EACvB,eAAe,GAKhB,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,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,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,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,oBAAoB,EACpB,cAAc,EACd,mBAAmB,EACnB,uBAAuB,EACvB,eAAe,GAIhB,MAAM,kBAAkB,CAAC"}

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

@@ -0,0 +1,29 @@
+/**
+ * Generic, pure, unit-testable bounded-parallel pipeline runner.
+ *
+ * Runs `worker` over `items` with at most `concurrency` in flight at once.
+ * Results are returned in input order regardless of completion order. A
+ * worker may throw {@link RateLimitSignal} to back off and retry: the affected
+ * item sleeps (per-item, via the injected `sleep`) then retries, bounded by a
+ * retry cap; other in-flight workers are NOT paused. Any other thrown error
+ * propagates and fails the whole run.
+ *
+ * Zero claude/network dependency: callers inject `sleep` for deterministic
+ * 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 interface RunPipelineOptions<T, R> {
+    readonly items: readonly T[];
+    /** Max number of workers in flight at once. Must be >= 1. */
+    readonly concurrency: number;
+    readonly worker: (item: T, index: number) => Promise<R>;
+    /** ms to wait before retrying, given the 1-based retry attempt number.
+     *  Default: exponential backoff, min(60_000, 1000 * 2^(attempt-1)). */
+    readonly onRateLimit?: (attempt: number) => number;
+    /** Injectable for tests; defaults to a real `setTimeout`-based sleep. */
+    readonly sleep?: (ms: number) => Promise<void>;
+}
+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

@@ -0,0 +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"}

package/dist/skill-test/pipeline.js

@@ -0,0 +1,77 @@
+/**
+ * Generic, pure, unit-testable bounded-parallel pipeline runner.
+ *
+ * Runs `worker` over `items` with at most `concurrency` in flight at once.
+ * Results are returned in input order regardless of completion order. A
+ * worker may throw {@link RateLimitSignal} to back off and retry: the affected
+ * item sleeps (per-item, via the injected `sleep`) then retries, bounded by a
+ * retry cap; other in-flight workers are NOT paused. Any other thrown error
+ * propagates and fails the whole run.
+ *
+ * Zero claude/network dependency: callers inject `sleep` for deterministic
+ * 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 {
+}
+/** Bounded retries on RateLimitSignal before giving up and rethrowing. */
+const MAX_RATE_LIMIT_RETRIES = 5;
+const DEFAULT_BACKOFF_BASE_MS = 1000;
+const DEFAULT_BACKOFF_CAP_MS = 60_000;
+function defaultOnRateLimit(attempt) {
+    return Math.min(DEFAULT_BACKOFF_CAP_MS, DEFAULT_BACKOFF_BASE_MS * 2 ** (attempt - 1));
+}
+function defaultSleep(ms) {
+    return new Promise((resolve) => {
+        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) {
+    let attempt = 0;
+    for (;;) {
+        try {
+            return await worker(item, index);
+        }
+        catch (error) {
+            if (!(error instanceof RateLimitSignal)) {
+                throw error;
+            }
+            attempt += 1;
+            if (attempt > MAX_RATE_LIMIT_RETRIES) {
+                throw error;
+            }
+            await sleep(onRateLimit(attempt));
+        }
+    }
+}
+export async function runPipeline(o) {
+    const { items, concurrency, worker } = o;
+    const onRateLimit = o.onRateLimit ?? defaultOnRateLimit;
+    const sleep = o.sleep ?? defaultSleep;
+    if (!Number.isInteger(concurrency) || concurrency < 1) {
+        throw new Error(`runPipeline: concurrency must be an integer >= 1 (got ${concurrency})`);
+    }
+    const results = new Array(items.length);
+    let nextIndex = 0;
+    async function runWorkerLoop() {
+        for (;;) {
+            const index = nextIndex;
+            nextIndex += 1;
+            if (index >= items.length) {
+                return;
+            }
+            // Non-null: index < items.length was just checked above.
+            const item = items[index];
+            results[index] = await runItemWithRetry(item, index, worker, onRateLimit, sleep);
+        }
+    }
+    const poolSize = Math.min(concurrency, items.length);
+    const pool = Array.from({ length: poolSize }, () => runWorkerLoop());
+    await Promise.all(pool);
+    return results;
+}
+//# sourceMappingURL=pipeline.js.map

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

@@ -0,0 +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"}

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

@@ -0,0 +1,27 @@
+/**
+ * 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
+ * user-correctable preflight exit code (2), regardless of which prompt
+ * builder raised it.
+ */
+export declare class PromptInvariantError extends Error {
+    constructor(message: string);
+}
+/**
+ * Append the per-run integrity-nonce directive to the grader prompt. The recipient must copy
+ * `nonce` verbatim into the artifact it writes, as a top-level `runNonce`
+ * field; the harness then rejects any artifact whose nonce is absent or wrong.
+ * This is defense-in-depth against untrusted skill code forging a passing
+ * result in the shared sandbox — without the secret nonce (delivered only via
+ * stdin, never written to disk) a forged result cannot be authenticated.
+ *
+ * Appended AFTER any prompt-invariant assertion and AFTER any user prompt
+ * override, so the nonce requirement is ALWAYS enforced and a committed
+ * config cannot opt out of it.
+ */
+export declare function appendIntegrityNonceDirective(prompt: string, nonce: string): string;
+//# sourceMappingURL=prompt-invariants.d.ts.map

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

@@ -0,0 +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"}

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

@@ -0,0 +1,38 @@
+/**
+ * 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
+ * user-correctable preflight exit code (2), regardless of which prompt
+ * builder raised it.
+ */
+export class PromptInvariantError extends Error {
+    constructor(message) {
+        super(`Prompt invariant violated: ${message}`);
+        this.name = 'PromptInvariantError';
+    }
+}
+/**
+ * Append the per-run integrity-nonce directive to the grader prompt. The recipient must copy
+ * `nonce` verbatim into the artifact it writes, as a top-level `runNonce`
+ * field; the harness then rejects any artifact whose nonce is absent or wrong.
+ * This is defense-in-depth against untrusted skill code forging a passing
+ * result in the shared sandbox — without the secret nonce (delivered only via
+ * stdin, never written to disk) a forged result cannot be authenticated.
+ *
+ * Appended AFTER any prompt-invariant assertion and AFTER any user prompt
+ * override, so the nonce requirement is ALWAYS enforced and a committed
+ * config cannot opt out of it.
+ */
+export function appendIntegrityNonceDirective(prompt, nonce) {
+    return [
+        prompt,
+        '',
+        'INTEGRITY (required, do this LAST): the artifact you write MUST include a',
+        `top-level string field "runNonce" whose value is EXACTLY: ${nonce}`,
+        'Copy it verbatim. vat rejects an artifact whose runNonce is missing or wrong.',
+    ].join('\n');
+}
+//# sourceMappingURL=prompt-invariants.js.map

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

@@ -0,0 +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"}

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

@@ -5,15 +5,23 @@
  *   lock → assert safe workdir + harness root → stage → resolve staged-subject
  *   eval path → bootstrap-check (scaffold template + exit 3 if absent)
  *   → preflight (return early exitCode 2 on failure) → ack enforcement
- *   → build effective prompt + assertPromptInvariants → dry-run short-circuit
- *   → spawnHeadlessClaude → parseGradingJson → release lock → return result
+ *   → dry-run short-circuit → build per-eval work items → run the vat-owned
+ *   executor→grader pipeline (bounded-parallel) → merge grader fragments →
+ *   write grading.json/friction.json (vat is SOLE writer) → reconcile verdict
+ *   → release lock → return result
  */
 import type { SkillSourceDescriptor } from '@vibe-agent-toolkit/resources';
+import { type spawnHeadlessClaude } from '@vibe-agent-toolkit/utils';
 import type { ResolveSkillSourceContext, SkillSource } from '../skill-source/types.js';
+import type { EvalFragment } from './eval-fragment.js';
+import { type EvalEntry, type EvalSuite } from './eval-inputs.js';
 import { type SkillTestExitCodeValue } from './exit-codes.js';
 import { type FrictionItem } from './friction-schema.js';
+import { type GradingVerdict } from './grading-adapter.js';
 import { type PreflightInput } from './preflight.js';
 import { type StageItem } from './staging.js';
+import { type SkippedEvalsSummary } from './tier-plan.js';
+import { type ToolEvalReport } from './tool-eval-schema.js';
 /** Auth modes supported by the harness. */
 export type HarnessAuthMode = 'inherit' | 'subscription' | 'api-key' | 'auto';
 /** Auth mechanism requirements. */
@@ -52,8 +60,22 @@ export interface RunHarnessOptions {
     auth?: HarnessAuthMode;
     /** Required auth mechanism. */
     requireAuth?: HarnessAuthMechanism;
-    /** Override the experimenter prompt. */
-    promptOverride?: string;
+    /**
+     * Pinned grader/judge model for the per-eval grader spawns. Defaults to
+     * {@link DEFAULT_GRADER_MODEL}. Deliberately distinct from `model` (which is
+     * the model under test in the executor spawns) so the judge stays comparable
+     * across runs regardless of the subject model.
+     */
+    graderModel?: string;
+    /** Bounded-parallel executor→grader pipeline width. Defaults to {@link DEFAULT_CONCURRENCY}. */
+    concurrency?: number;
+    /**
+     * Injectable spawn seam (tests only). When set, it replaces the real
+     * {@link spawnHeadlessClaude} in BOTH the executor and grader per-eval spawns,
+     * so a test can drive the full harness with a fake `claude` and no real
+     * install. Production callers leave it undefined (the real spawn is used).
+     */
+    spawn?: typeof spawnHeadlessClaude;
     /** Enable A/B baseline run (with/without skill). */
     baseline?: boolean;
     /** Allow unverified skill source (skip manifest check). */
@@ -103,8 +125,21 @@ export interface RunHarnessOptions {
     dryRunStagedExistingDist?: boolean;
     /** Feature B: explicit env var injections (interpolated at stage time). */
     env?: Record<string, string>;
-    /** Feature A: host env var names to forward to the experimenter if present. */
+    /** Feature A: host env var names to forward to the executor spawn if present. */
     passEnv?: readonly string[];
+    /**
+     * Declared executables the subject skill ships (name + kind + howInvoked),
+     * populated by run.ts from the resolved subject's packaging config WHEN cleanly
+     * reachable (a `buildable` ref carries `packagingConfig`; a plain path source
+     * does not — issue #145 Phase T). Passed to the grader on the WITH arm ONLY as
+     * a recognition aid alongside each eval's `toolExpectations`. Absent → the
+     * grader still matches tools by the commands it sees in the transcript.
+     */
+    declaredExecutables?: Array<{
+        name: string;
+        howInvoked: string;
+        kind: string;
+    }>;
     /**
      * Opt-OUT of eval gating (for interactive use). By DEFAULT (false/absent) a
      * failing verdict returns exit EvalFailure (4) — fail-closed, so CI catches a
@@ -120,8 +155,8 @@ export interface RunHarnessResult {
     summary: string;
 }
 /**
- * Built-in cost/runtime safety ceilings for the experimenter spawn. These are the
- * SAME values the harness applies as defaults, but exported as an explicit cap so
+ * Built-in cost/runtime safety ceilings applied PER executor/grader spawn. These
+ * are the SAME values the harness applies as defaults, but exported as an explicit cap so
  * the CLI precedence layer (run.ts) can enforce a critical asymmetry:
  *
  *   - a CLI flag (explicit operator intent, typed at the terminal for THIS run)
@@ -140,21 +175,13 @@ export declare const SKILL_TEST_BUILTIN_CAPS: {
     readonly timeoutSeconds: number;
 };
 /**
- * Compute the DEFAULT wall-clock budget (ms) when the operator gave no explicit
- * `--timeout`. It scales with the number of declared evals so a large suite isn't
- * killed at the flat 5-minute default before it can finish: a 120s base plus 120s
- * per declared eval, floored at the historical 300s default (so we never shrink
- * below today's budget) and capped at 1h (so a runaway/huge suite can't request an
- * unbounded budget). An EXPLICIT `--timeout` always bypasses this (see resolveTimeoutMs).
- */
-export declare function computeDefaultTimeoutMs(declaredEvalCount: number): number;
-/**
- * Resolve the effective wall-clock timeout (ms). An explicit `--timeout` (seconds)
- * ALWAYS wins, unchanged. Otherwise the default scales with the declared eval count
- * via {@link computeDefaultTimeoutMs}; when that count is unknown/unparseable at the
- * call site, fall back to the flat 300s default.
+ * Resolve the effective PER-EVAL wall-clock timeout (ms). Each executor and
+ * grader spawn is an independent, bounded-parallel unit (issue #145), so the
+ * budget is a flat per-spawn ceiling — NOT scaled by the suite size the way the
+ * old single serial run's budget was. An explicit `--timeout` (seconds)
+ * wins; otherwise the flat {@link DEFAULT_TIMEOUT_MS} default applies.
  */
-export declare function resolveTimeoutMs(opts: RunHarnessOptions, declaredEvalCount?: number): number;
+export declare function resolveTimeoutMs(opts: RunHarnessOptions): number;
 /**
  * Map an eval verdict to a process exit code. Default behavior (fail-closed): a
  * failing verdict escalates to EvalFailure (4) — distinct from the harness-broke
@@ -203,34 +230,6 @@ export declare function renderPreflightSummary(checks: {
  * (the harness Step-6 check and the run.ts pre-build check cannot diverge).
  */
 export declare function isAcknowledged(opts: Pick<RunHarnessOptions, 'dryRun' | 'acknowledgedRunsSkillCode'>): boolean;
-/** Optional context for the timeout message (surfaced only when available). */
-export interface TimeoutMessageContext {
-    /** Number of evals the suite DECLARED (parsed suite length), if known. */
-    declaredEvalCount?: number;
-    /** Best-effort count of evals that completed before the timeout, if cheaply known. */
-    completedEvalCount?: number;
-}
-/**
- * Format the experimenter-timeout message. Pure so it can be unit-tested. Always
- * reports the raw budget (ms + a minute label). When the declared eval count is
- * known it appends the count and a hint that the default budget scales with eval
- * count (raise `--timeout`/`--stall` for larger suites); when a best-effort
- * completed count is also available it notes `completed ~N/declared`. With no
- * declared count it stays honest and omits those clauses.
- */
-export declare function formatTimeoutMessage(input: {
-    timeoutMs: number;
-} & TimeoutMessageContext): string;
-/**
- * Translate a non-success spawn outcome into an InternalHarnessError (exit 1).
- * A stall, a timeout, OR a non-zero exit are each authoritative — a non-zero exit
- * is never laundered into a PASS even if a grading.json happens to be on disk.
- */
-export declare function assertExperimenterSucceeded(spawnResult: {
-    stalled: boolean;
-    timedOut: boolean;
-    status: number;
-}, stallMs: number | undefined, timeoutMs: number, timeoutContext?: TimeoutMessageContext): void;
 /**
  * Format a friction report for human consumption — one line per entry as
  * `[<severity>] <category>: <message>`. Pure; returns the empty string for no
@@ -247,14 +246,15 @@ export declare function formatFrictionReport(items: readonly FrictionItem[]): st
 export declare function subjectSkillName(opts: RunHarnessOptions): string;
 export declare function resolveScaffoldEvalsPath(opts: RunHarnessOptions, repoRoot: string, evalsSubpath: string): string;
 /** Parse the staged eval suite and materialize each eval's input `files` into
- * `<harnessRoot>/workspaces/<id>/`. Returns the workspaces root AND the number of
- * evals the suite declared (so the caller can scale the default timeout without
- * re-reading the suite). The dir is wiped first so a reused harness root cannot
- * leak a prior run's inputs. Throws {@link EvalInputError} (mapped by the caller to
- * exit 2) on a bad suite or a missing input file. */
+ * `<harnessRoot>/workspaces/<id>/`. Returns the workspaces root, the parsed
+ * {@link EvalSuite} (so the eval loop has the entries without re-reading), and the
+ * declared eval count (derived from the suite). The dir is wiped first so a reused
+ * harness root cannot leak a prior run's inputs. Throws {@link EvalInputError}
+ * (mapped by the caller to exit 2) on a bad suite or a missing input file. */
 export declare function stageWorkspacesForRun(evalsPath: string, harnessRoot: string): {
     workspacesRoot: string;
     declaredEvalCount: number;
+    suite: EvalSuite;
 };
 /** Inputs for the dry-run summary string. */
 export interface DryRunSummaryInput {
@@ -271,8 +271,14 @@ export interface DryRunSummaryInput {
     provenanceFingerprint: string;
     /** Number of entries in the staged manifest. */
     provenanceEntryCount: number;
-    /** The assembled model flag string (e.g. `--model claude-opus-4-8`). */
+    /** The assembled model flag string for the executor (e.g. `--model claude-opus-4-8`). */
     modelFlag: string;
+    /** Number of declared evals a real run would execute. */
+    evalCount: number;
+    /** Bounded-parallel executor→grader pipeline width a real run would use. */
+    concurrency: number;
+    /** Pinned grader/judge model a real run would grade with. */
+    graderModel: string;
 }
 /**
  * Build the dry-run summary string. Pure function so it can be unit-tested
@@ -311,6 +317,132 @@ export interface CleanupHarnessOptions {
  * could otherwise delete the symlink's target outside tmp.
  */
 export declare function cleanupHarness(harnessRoot: string, opts: CleanupHarnessOptions): void;
+/** One unit of executor→grader work: an eval + which arm (with/without skill). */
+export interface EvalWorkItem {
+    entry: EvalEntry;
+    arm: 'with' | 'without';
+}
+/**
+ * Build the per-eval work items for a set of evals (one tier's worth, or a whole
+ * suite). Every eval gets a WITH arm (skill present). When `baseline` is set,
+ * every eval ALSO gets a WITHOUT arm (skill absent) so vat can record an
+ * informational A/B — the WITHOUT arm never contributes to the pass/fail verdict
+ * (see {@link partitionFragmentsByArm}) and never drives tier gating. Pure +
+ * unit-testable.
+ */
+export declare function buildEvalWorkItems(evals: readonly EvalEntry[], baseline: boolean): EvalWorkItem[];
+/**
+ * Partition graded fragments into the WITH arm (the authoritative verdict +
+ * grading.json) and the WITHOUT arm (baseline.json, informational only). A
+ * fragment with no `arm` (or `arm: 'with'`) is a WITH-arm fragment. Pure +
+ * unit-testable.
+ */
+export declare function partitionFragmentsByArm(fragments: EvalFragment[]): {
+    withArm: EvalFragment[];
+    withoutArm: EvalFragment[];
+};
+/**
+ * The vat-only grader dir for a run: `<tmp>/vat-skill-grade-<dirToken>/`. It is
+ * deliberately OUTSIDE the harness root (the skill's `--add-dir` sandbox) and
+ * created 0700, so it is invisible to Claude's own permission model and to any
+ * OTHER OS user. Pure (derives a path only).
+ *
+ * SCOPE OF THE GUARANTEE — read honestly. `--add-dir`/`bypassPermissions` is
+ * Claude's permission model, NOT an OS sandbox: the executor's skill code runs
+ * as the SAME OS uid as vat, so it CAN read a 0700 dir this process owns. The
+ * layered defenses here — dir outside the sandbox, named by an unpredictable
+ * `dirToken` (distinct from the integrity nonce, which never touches the dir
+ * name or any argv and travels only via grader stdin), the nonce echoed back
+ * per fragment, and each fragment file unlinked the instant vat reads it — RAISE
+ * THE BAR against same-uid forgery (a forger must now win a per-fragment
+ * read→overwrite race against a secret it cannot predict, with no persisted copy
+ * to harvest at leisure). They do NOT amount to true isolation from same-uid
+ * code. The complete fix is running the grader under a SEPARATE OS uid /
+ * container; that is tracked as a follow-up (see CHANGELOG "Security" notes) and
+ * is the only thing that closes the residual race outright.
+ */
+export declare function resolveGraderOutDir(dirToken: string): string;
+/**
+ * The executor working directory for one eval: its staged input workspace
+ * `<workspacesRoot>/<id>` when the eval declares input `files`, else undefined
+ * (the executor then defaults to the staged subject dir). Pure + unit-testable.
+ */
+export declare function resolvePerEvalWorkspaceDir(entry: EvalEntry, workspacesRoot: string): string | undefined;
+/** Best-effort removal of the vat-only grader dir (never throws — runs from cleanup). */
+export declare function removeGraderOutDir(graderOutDir: string | undefined): void;
+/** The results/ artifacts vat is the SOLE writer of, resolved for one run. */
+export interface ArtifactPaths {
+    gradingOut: string;
+    frictionOut: string;
+    baselineOut: string;
+    toolEvalOut: string;
+}
+/** Resolve the run's grading/friction/baseline/tool-eval artifact paths under
+ *  `resultsDir`. Single source of truth for the filenames (used by the
+ *  pre-pipeline stale wipe AND the post-merge writer). */
+export declare function resolveArtifactPaths(resultsDir: string): ArtifactPaths;
+/**
+ * Remove any PRIOR run's artifacts before this run writes its own. The harness
+ * root is deterministic per skill-set and reused across runs (`--keep`/`--out`, or
+ * after a crash/SIGKILL that preempted cleanup), so a stale grading/friction/
+ * baseline/tool-eval.json can otherwise survive into a run that throws BEFORE the merge —
+ * where the `finally` would then echo the PRIOR run's friction as if it were this
+ * run's. Wiping all three up front closes that cross-run leak. Best-effort
+ * (`force: true`) — a missing file is fine.
+ */
+export declare function wipeStaleArtifacts(paths: ArtifactPaths): void;
+/**
+ * The COMPOSITE run verdict (issue #145 Phase T): the run passes only when BOTH
+ * the prose-expectation grading passed AND every tool-expectation verdict passed.
+ * Tool verdicts live in tool-eval.json (a SEPARATE channel — C2); this combines
+ * the two at the exit-code layer WITHOUT mixing the channels' data. When no eval
+ * declared `toolExpectations`, `toolEval.evals` is empty and this equals
+ * `outputAllPassed`. Pure + unit-testable.
+ */
+export declare function computeCompositeVerdict(outputAllPassed: boolean, toolEval: ToolEvalReport): boolean;
+/**
+ * The run's final pass/fail after cost-tiered fail-fast: the {@link
+ * computeCompositeVerdict} of the tiers that RAN, AND no tiers were skipped. A
+ * fail-fast run that gated higher tiers is NEVER a pass (skipped ≠ passed) — this
+ * forces `false` so the exit code is EvalFailure (4), never downgraded to 0 by the
+ * composite path alone. Pure + unit-testable.
+ */
+export declare function resolveCompositeAllPassed(outputAllPassed: boolean, toolEval: ToolEvalReport, skipped: SkippedEvalsSummary | undefined): boolean;
+/**
+ * Running total of spend across every executor+grader session in a run (adopter
+ * follow-up). `sessions` counts only sessions that REPORTED a `total_cost_usd`, so
+ * `≈$${totalUsd} across ${sessions} sessions` is always internally consistent (the
+ * sum is over exactly those sessions). Mutated in place by {@link recordSessionCost}
+ * — safe under the cooperative (single-threaded) pipeline concurrency.
+ */
+export interface RunCostSummary {
+    totalUsd: number;
+    sessions: number;
+}
+/** Fold one session's `total_cost_usd` into the accumulator; a non-number (mock spawn / missing result) is ignored. */
+export declare function recordSessionCost(acc: RunCostSummary, totalCostUsd: number | undefined): void;
+/**
+ * The ` | ≈$0.42 across 6 sessions` spend suffix for the summary line, or `''`
+ * when no session reported a cost (e.g. every spawn was a test mock) so the suffix
+ * never adds noise to a run with no cost signal. Pure + unit-testable.
+ */
+export declare function formatRunCostSuffix(cost: RunCostSummary | undefined): string;
+/**
+ * The run's human summary line, computed from the COMPOSITE verdict so an
+ * output-pass with a failing tool verdict still reads FAIL. The prose-expectation
+ * counts (`passed/total`) come from the grading verdict; when any tool-expectation
+ * verdict failed, a `(N tool)` suffix names how many — so a composite FAIL whose
+ * OUTPUT counts look all-green (e.g. `FAIL 3/3 (1 tool)`) is self-explaining. Pure.
+ */
+export declare function buildRunSummary(verdict: GradingVerdict, toolEval: ToolEvalReport, compositeAllPassed: boolean): string;
+/**
+ * The run summary line, appending the fail-fast SKIPPED note when the tier gate
+ * stopped higher tiers. Legibility is required — the skipped tiers are named on
+ * their own line, never silently dropped. Pure. Note that a run with skipped
+ * tiers ALWAYS reads FAIL (skipped ≠ passed forces `compositeAllPassed` false at
+ * the call site), so the base line is already FAIL when the note is present.
+ */
+export declare function buildRunSummaryWithSkips(verdict: GradingVerdict, toolEval: ToolEvalReport, compositeAllPassed: boolean, skipped: SkippedEvalsSummary | undefined, cost?: RunCostSummary): string;
 /**
  * Domain orchestrator for `vat skill test run`. Pure of process.exit — all
  * exit-code decisions live in the caller (run.ts).

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

@@ -1 +1 @@
-{"version":3,"file":"run-harness.d.ts","sourceRoot":"","sources":["../../src/skill-test/run-harness.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAOH,OAAO,KAAK,EAAE,qBAAqB,EAAE,MAAM,+BAA+B,CAAC;AAe3E,OAAO,KAAK,EAAuB,yBAAyB,EAAE,WAAW,EAAE,MAAM,0BAA0B,CAAC;AAK5G,OAAO,EAIL,KAAK,sBAAsB,EAC5B,MAAM,iBAAiB,CAAC;AAOzB,OAAO,EAAwB,KAAK,YAAY,EAAE,MAAM,sBAAsB,CAAC;AAK/E,OAAO,EAAgB,KAAK,cAAc,EAAE,MAAM,gBAAgB,CAAC;AACnE,OAAO,EAAoC,KAAK,SAAS,EAAE,MAAM,cAAc,CAAC;AAsBhF,2CAA2C;AAC3C,MAAM,MAAM,eAAe,GAAG,SAAS,GAAG,cAAc,GAAG,SAAS,GAAG,MAAM,CAAC;AAE9E,mCAAmC;AACnC,MAAM,MAAM,oBAAoB,GAAG,cAAc,GAAG,SAAS,CAAC;AAE9D,MAAM,WAAW,iBAAiB;IAChC,mDAAmD;IACnD,MAAM,EAAE,MAAM,EAAE,CAAC;IAEjB;;;;;OAKG;IACH,QAAQ,CAAC,EAAE,MAAM,CAAC;IAElB;;;OAGG;IACH,YAAY,CAAC,EAAE,MAAM,CAAC;IAEtB;;;OAGG;IACH,WAAW,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,qBAAqB,CAAC,CAAC;IAEpD,4DAA4D;IAC5D,YAAY,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,qBAAqB,CAAC,CAAC;IAErD,+EAA+E;IAC/E,OAAO,CAAC,EAAE,MAAM,CAAC;IAEjB,kEAAkE;IAClE,GAAG,CAAC,EAAE,MAAM,CAAC;IAEb,mEAAmE;IACnE,OAAO,CAAC,EAAE,OAAO,CAAC;IAElB,iEAAiE;IACjE,IAAI,CAAC,EAAE,OAAO,CAAC;IAEf,2CAA2C;IAC3C,IAAI,CAAC,EAAE,eAAe,CAAC;IAEvB,+BAA+B;IAC/B,WAAW,CAAC,EAAE,oBAAoB,CAAC;IAEnC,wCAAwC;IACxC,cAAc,CAAC,EAAE,MAAM,CAAC;IAExB,oDAAoD;IACpD,QAAQ,CAAC,EAAE,OAAO,CAAC;IAEnB,2DAA2D;IAC3D,0BAA0B,CAAC,EAAE,OAAO,CAAC;IAErC,qDAAqD;IACrD,MAAM,CAAC,EAAE,OAAO,CAAC;IAEjB,qEAAqE;IACrE,yBAAyB,CAAC,EAAE,OAAO,CAAC;IAGpC,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,0BAA0B;IAC1B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,iCAAiC;IACjC,KAAK,CAAC,EAAE,MAAM,CAAC;IAEf;;;;OAIG;IACH,aAAa,CAAC,EAAE,WAAW,CAAC;IAE5B;;;;;;OAMG;IACH,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAE5B;;;OAGG;IACH,OAAO,CAAC,EAAE,OAAO,CAAC;IAElB;;;;OAIG;IACH,UAAU,CAAC,EAAE,OAAO,CAAC;IAErB;;;;;OAKG;IACH,wBAAwB,CAAC,EAAE,OAAO,CAAC;IAEnC,2EAA2E;IAC3E,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC7B,+EAA+E;IAC/E,OAAO,CAAC,EAAE,SAAS,MAAM,EAAE,CAAC;IAE5B;;;;;;OAMG;IACH,mBAAmB,CAAC,EAAE,OAAO,CAAC;CAC/B;AAED,MAAM,WAAW,gBAAgB;IAC/B,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,MAAM,CAAC;CACjB;AAUD;;;;;;;;;;;;;;GAcG;AACH,eAAO,MAAM,uBAAuB;;;;CAI1B,CAAC;AAEX;;;;;;;GAOG;AACH,wBAAgB,uBAAuB,CAAC,iBAAiB,EAAE,MAAM,GAAG,MAAM,CAEzE;AAED;;;;;GAKG;AACH,wBAAgB,gBAAgB,CAAC,IAAI,EAAE,iBAAiB,EAAE,iBAAiB,CAAC,EAAE,MAAM,GAAG,MAAM,CAG5F;AAED;;;;;;GAMG;AACH,wBAAgB,eAAe,CAAC,SAAS,EAAE,OAAO,EAAE,mBAAmB,EAAE,OAAO,GAAG,sBAAsB,CAExG;AAED,wBAAgB,cAAc,CAAC,IAAI,EAAE,iBAAiB,GAAG,MAAM,GAAG,SAAS,CAE1E;AAED,wBAAgB,YAAY,CAAC,IAAI,EAAE,iBAAiB,GAAG;IACrD,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB,CASA;AAED;;;;;;;GAOG;AACH,wBAAgB,sBAAsB,CACpC,MAAM,EAAE,WAAW,EACnB,QAAQ,EAAE,MAAM,GACf,SAAS,CAAC,cAAc,CAAC,GAAG,SAAS,CAIvC;AAED,wBAAgB,aAAa,CAC3B,IAAI,EAAE,MAAM,EACZ,MAAM,EAAE,WAAW,EACnB,QAAQ,EAAE,MAAM,EAChB,IAAI,EAAE,SAAS,GAAG,SAAS,GAC1B,SAAS,CAQX;AAED,wBAAgB,eAAe,CAAC,IAAI,EAAE,iBAAiB,EAAE,QAAQ,EAAE,MAAM,GAAG,SAAS,EAAE,CA2BtF;AAED,wBAAgB,eAAe,CAAC,WAAW,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,GAAG,yBAAyB,CAMhG;AAkBD;;;GAGG;AACH,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAEtD;AA4BD,wBAAgB,mBAAmB,CACjC,SAAS,EAAE,MAAM,EACjB,UAAU,EAAE,MAAM,EAAE,EACpB,IAAI,EAAE,iBAAiB,EACvB,KAAK,EAAE;IAAE,YAAY,EAAE,MAAM,CAAA;CAAE,GAC9B,cAAc,CA6BhB;AAED,wBAAgB,sBAAsB,CAAC,MAAM,EAAE;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,OAAO,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,EAAE,GAAG,MAAM,CAK3G;AAED;;;;;;GAMG;AACH,wBAAgB,cAAc,CAC5B,IAAI,EAAE,IAAI,CAAC,iBAAiB,EAAE,QAAQ,GAAG,2BAA2B,CAAC,GACpE,OAAO,CAET;AAED,+EAA+E;AAC/E,MAAM,WAAW,qBAAqB;IACpC,0EAA0E;IAC1E,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,sFAAsF;IACtF,kBAAkB,CAAC,EAAE,MAAM,CAAC;CAC7B;AAED;;;;;;;GAOG;AACH,wBAAgB,oBAAoB,CAAC,KAAK,EAAE;IAAE,SAAS,EAAE,MAAM,CAAA;CAAE,GAAG,qBAAqB,GAAG,MAAM,CAcjG;AAED;;;;GAIG;AACH,wBAAgB,2BAA2B,CACzC,WAAW,EAAE;IAAE,OAAO,EAAE,OAAO,CAAC;IAAC,QAAQ,EAAE,OAAO,CAAC;IAAC,MAAM,EAAE,MAAM,CAAA;CAAE,EACpE,OAAO,EAAE,MAAM,GAAG,SAAS,EAC3B,SAAS,EAAE,MAAM,EACjB,cAAc,CAAC,EAAE,qBAAqB,GACrC,IAAI,CAUN;AAED;;;;GAIG;AACH,wBAAgB,oBAAoB,CAAC,KAAK,EAAE,SAAS,YAAY,EAAE,GAAG,MAAM,CAE3E;AA2BD;;;;;GAKG;AACH,uFAAuF;AACvF,wBAAgB,gBAAgB,CAAC,IAAI,EAAE,iBAAiB,GAAG,MAAM,CAGhE;AAED,wBAAgB,wBAAwB,CAAC,IAAI,EAAE,iBAAiB,EAAE,QAAQ,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,GAAG,MAAM,CAchH;AAkDD;;;;;qDAKqD;AACrD,wBAAgB,qBAAqB,CACnC,SAAS,EAAE,MAAM,EACjB,WAAW,EAAE,MAAM,GAClB;IAAE,cAAc,EAAE,MAAM,CAAC;IAAC,iBAAiB,EAAE,MAAM,CAAA;CAAE,CAUvD;AAiFD,6CAA6C;AAC7C,MAAM,WAAW,kBAAkB;IACjC,uFAAuF;IACvF,UAAU,EAAE,OAAO,CAAC;IACpB;;;OAGG;IACH,wBAAwB,CAAC,EAAE,OAAO,CAAC;IACnC,sEAAsE;IACtE,cAAc,EAAE,MAAM,CAAC;IACvB,oDAAoD;IACpD,qBAAqB,EAAE,MAAM,CAAC;IAC9B,gDAAgD;IAChD,oBAAoB,EAAE,MAAM,CAAC;IAC7B,wEAAwE;IACxE,SAAS,EAAE,MAAM,CAAC;CACnB;AAED;;;;;;;;;;;;GAYG;AACH,wBAAgB,kBAAkB,CAAC,KAAK,EAAE,kBAAkB,GAAG,MAAM,CA+BpE;AAMD,MAAM,WAAW,qBAAqB;IACpC,sEAAsE;IACtE,IAAI,EAAE,OAAO,CAAC;IACd;;;;OAIG;IACH,OAAO,EAAE,OAAO,CAAC;CAClB;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,cAAc,CAAC,WAAW,EAAE,MAAM,EAAE,IAAI,EAAE,qBAAqB,GAAG,IAAI,CAcrF;AAMD;;;GAGG;AACH,wBAAsB,mBAAmB,CAAC,IAAI,EAAE,iBAAiB,GAAG,OAAO,CAAC,gBAAgB,CAAC,CAsS5F"}
+{"version":3,"file":"run-harness.d.ts","sourceRoot":"","sources":["../../src/skill-test/run-harness.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAOH,OAAO,KAAK,EAAE,qBAAqB,EAAE,MAAM,+BAA+B,CAAC;AAC3E,OAAO,EASL,KAAK,mBAAmB,EAGzB,MAAM,2BAA2B,CAAC;AAGnC,OAAO,KAAK,EAAuB,yBAAyB,EAAE,WAAW,EAAE,MAAM,0BAA0B,CAAC;AAI5G,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAEvD,OAAO,EAAuD,KAAK,SAAS,EAAE,KAAK,SAAS,EAAE,MAAM,kBAAkB,CAAC;AAGvH,OAAO,EAIL,KAAK,sBAAsB,EAC5B,MAAM,iBAAiB,CAAC;AAEzB,OAAO,EAAwB,KAAK,YAAY,EAAE,MAAM,sBAAsB,CAAC;AAE/E,OAAO,EAAoB,KAAK,cAAc,EAAE,MAAM,sBAAsB,CAAC;AAM7E,OAAO,EAAgB,KAAK,cAAc,EAAE,MAAM,gBAAgB,CAAC;AACnE,OAAO,EAAoC,KAAK,SAAS,EAAE,MAAM,cAAc,CAAC;AAChF,OAAO,EAKL,KAAK,mBAAmB,EACzB,MAAM,gBAAgB,CAAC;AACxB,OAAO,EAAwB,KAAK,cAAc,EAAE,MAAM,uBAAuB,CAAC;AA8BlF,2CAA2C;AAC3C,MAAM,MAAM,eAAe,GAAG,SAAS,GAAG,cAAc,GAAG,SAAS,GAAG,MAAM,CAAC;AAE9E,mCAAmC;AACnC,MAAM,MAAM,oBAAoB,GAAG,cAAc,GAAG,SAAS,CAAC;AAE9D,MAAM,WAAW,iBAAiB;IAChC,mDAAmD;IACnD,MAAM,EAAE,MAAM,EAAE,CAAC;IAEjB;;;;;OAKG;IACH,QAAQ,CAAC,EAAE,MAAM,CAAC;IAElB;;;OAGG;IACH,YAAY,CAAC,EAAE,MAAM,CAAC;IAEtB;;;OAGG;IACH,WAAW,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,qBAAqB,CAAC,CAAC;IAEpD,4DAA4D;IAC5D,YAAY,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,qBAAqB,CAAC,CAAC;IAErD,+EAA+E;IAC/E,OAAO,CAAC,EAAE,MAAM,CAAC;IAEjB,kEAAkE;IAClE,GAAG,CAAC,EAAE,MAAM,CAAC;IAEb,mEAAmE;IACnE,OAAO,CAAC,EAAE,OAAO,CAAC;IAElB,iEAAiE;IACjE,IAAI,CAAC,EAAE,OAAO,CAAC;IAEf,2CAA2C;IAC3C,IAAI,CAAC,EAAE,eAAe,CAAC;IAEvB,+BAA+B;IAC/B,WAAW,CAAC,EAAE,oBAAoB,CAAC;IAEnC;;;;;OAKG;IACH,WAAW,CAAC,EAAE,MAAM,CAAC;IAErB,gGAAgG;IAChG,WAAW,CAAC,EAAE,MAAM,CAAC;IAErB;;;;;OAKG;IACH,KAAK,CAAC,EAAE,OAAO,mBAAmB,CAAC;IAEnC,oDAAoD;IACpD,QAAQ,CAAC,EAAE,OAAO,CAAC;IAEnB,2DAA2D;IAC3D,0BAA0B,CAAC,EAAE,OAAO,CAAC;IAErC,qDAAqD;IACrD,MAAM,CAAC,EAAE,OAAO,CAAC;IAEjB,qEAAqE;IACrE,yBAAyB,CAAC,EAAE,OAAO,CAAC;IAGpC,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,0BAA0B;IAC1B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,iCAAiC;IACjC,KAAK,CAAC,EAAE,MAAM,CAAC;IAEf;;;;OAIG;IACH,aAAa,CAAC,EAAE,WAAW,CAAC;IAE5B;;;;;;OAMG;IACH,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAE5B;;;OAGG;IACH,OAAO,CAAC,EAAE,OAAO,CAAC;IAElB;;;;OAIG;IACH,UAAU,CAAC,EAAE,OAAO,CAAC;IAErB;;;;;OAKG;IACH,wBAAwB,CAAC,EAAE,OAAO,CAAC;IAEnC,2EAA2E;IAC3E,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC7B,iFAAiF;IACjF,OAAO,CAAC,EAAE,SAAS,MAAM,EAAE,CAAC;IAE5B;;;;;;;OAOG;IACH,mBAAmB,CAAC,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAEhF;;;;;;OAMG;IACH,mBAAmB,CAAC,EAAE,OAAO,CAAC;CAC/B;AAED,MAAM,WAAW,gBAAgB;IAC/B,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,MAAM,CAAC;CACjB;AAUD;;;;;;;;;;;;;;GAcG;AACH,eAAO,MAAM,uBAAuB;;;;CAI1B,CAAC;AAEX;;;;;;GAMG;AACH,wBAAgB,gBAAgB,CAAC,IAAI,EAAE,iBAAiB,GAAG,MAAM,CAEhE;AAED;;;;;;GAMG;AACH,wBAAgB,eAAe,CAAC,SAAS,EAAE,OAAO,EAAE,mBAAmB,EAAE,OAAO,GAAG,sBAAsB,CAExG;AAED,wBAAgB,cAAc,CAAC,IAAI,EAAE,iBAAiB,GAAG,MAAM,GAAG,SAAS,CAE1E;AAED,wBAAgB,YAAY,CAAC,IAAI,EAAE,iBAAiB,GAAG;IACrD,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB,CASA;AAED;;;;;;;GAOG;AACH,wBAAgB,sBAAsB,CACpC,MAAM,EAAE,WAAW,EACnB,QAAQ,EAAE,MAAM,GACf,SAAS,CAAC,cAAc,CAAC,GAAG,SAAS,CAIvC;AAED,wBAAgB,aAAa,CAC3B,IAAI,EAAE,MAAM,EACZ,MAAM,EAAE,WAAW,EACnB,QAAQ,EAAE,MAAM,EAChB,IAAI,EAAE,SAAS,GAAG,SAAS,GAC1B,SAAS,CAQX;AAED,wBAAgB,eAAe,CAAC,IAAI,EAAE,iBAAiB,EAAE,QAAQ,EAAE,MAAM,GAAG,SAAS,EAAE,CA2BtF;AAED,wBAAgB,eAAe,CAAC,WAAW,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,GAAG,yBAAyB,CAMhG;AAkBD;;;GAGG;AACH,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAEtD;AA4BD,wBAAgB,mBAAmB,CACjC,SAAS,EAAE,MAAM,EACjB,UAAU,EAAE,MAAM,EAAE,EACpB,IAAI,EAAE,iBAAiB,EACvB,KAAK,EAAE;IAAE,YAAY,EAAE,MAAM,CAAA;CAAE,GAC9B,cAAc,CA6BhB;AAED,wBAAgB,sBAAsB,CAAC,MAAM,EAAE;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,OAAO,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,EAAE,GAAG,MAAM,CAK3G;AAED;;;;;;GAMG;AACH,wBAAgB,cAAc,CAC5B,IAAI,EAAE,IAAI,CAAC,iBAAiB,EAAE,QAAQ,GAAG,2BAA2B,CAAC,GACpE,OAAO,CAET;AAED;;;;GAIG;AACH,wBAAgB,oBAAoB,CAAC,KAAK,EAAE,SAAS,YAAY,EAAE,GAAG,MAAM,CAE3E;AA2BD;;;;;GAKG;AACH,uFAAuF;AACvF,wBAAgB,gBAAgB,CAAC,IAAI,EAAE,iBAAiB,GAAG,MAAM,CAGhE;AAED,wBAAgB,wBAAwB,CAAC,IAAI,EAAE,iBAAiB,EAAE,QAAQ,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,GAAG,MAAM,CAchH;AAkDD;;;;;8EAK8E;AAC9E,wBAAgB,qBAAqB,CACnC,SAAS,EAAE,MAAM,EACjB,WAAW,EAAE,MAAM,GAClB;IAAE,cAAc,EAAE,MAAM,CAAC;IAAC,iBAAiB,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,SAAS,CAAA;CAAE,CAWzE;AA8FD,6CAA6C;AAC7C,MAAM,WAAW,kBAAkB;IACjC,uFAAuF;IACvF,UAAU,EAAE,OAAO,CAAC;IACpB;;;OAGG;IACH,wBAAwB,CAAC,EAAE,OAAO,CAAC;IACnC,sEAAsE;IACtE,cAAc,EAAE,MAAM,CAAC;IACvB,oDAAoD;IACpD,qBAAqB,EAAE,MAAM,CAAC;IAC9B,gDAAgD;IAChD,oBAAoB,EAAE,MAAM,CAAC;IAC7B,yFAAyF;IACzF,SAAS,EAAE,MAAM,CAAC;IAClB,yDAAyD;IACzD,SAAS,EAAE,MAAM,CAAC;IAClB,4EAA4E;IAC5E,WAAW,EAAE,MAAM,CAAC;IACpB,6DAA6D;IAC7D,WAAW,EAAE,MAAM,CAAC;CACrB;AAED;;;;;;;;;;;;GAYG;AACH,wBAAgB,kBAAkB,CAAC,KAAK,EAAE,kBAAkB,GAAG,MAAM,CAiCpE;AAMD,MAAM,WAAW,qBAAqB;IACpC,sEAAsE;IACtE,IAAI,EAAE,OAAO,CAAC;IACd;;;;OAIG;IACH,OAAO,EAAE,OAAO,CAAC;CAClB;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,cAAc,CAAC,WAAW,EAAE,MAAM,EAAE,IAAI,EAAE,qBAAqB,GAAG,IAAI,CAcrF;AAMD,kFAAkF;AAClF,MAAM,WAAW,YAAY;IAC3B,KAAK,EAAE,SAAS,CAAC;IACjB,GAAG,EAAE,MAAM,GAAG,SAAS,CAAC;CACzB;AAED;;;;;;;GAOG;AACH,wBAAgB,kBAAkB,CAAC,KAAK,EAAE,SAAS,SAAS,EAAE,EAAE,QAAQ,EAAE,OAAO,GAAG,YAAY,EAAE,CAOjG;AAED;;;;;GAKG;AACH,wBAAgB,uBAAuB,CAAC,SAAS,EAAE,YAAY,EAAE,GAAG;IAClE,OAAO,EAAE,YAAY,EAAE,CAAC;IACxB,UAAU,EAAE,YAAY,EAAE,CAAC;CAC5B,CAQA;AAED;;;;;;;;;;;;;;;;;;;GAmBG;AACH,wBAAgB,mBAAmB,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,CAE5D;AAED;;;;GAIG;AACH,wBAAgB,0BAA0B,CAAC,KAAK,EAAE,SAAS,EAAE,cAAc,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAGvG;AAED,yFAAyF;AACzF,wBAAgB,kBAAkB,CAAC,YAAY,EAAE,MAAM,GAAG,SAAS,GAAG,IAAI,CAWzE;AA2ID,8EAA8E;AAC9E,MAAM,WAAW,aAAa;IAC5B,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,MAAM,CAAC;IACpB,WAAW,EAAE,MAAM,CAAC;IACpB,WAAW,EAAE,MAAM,CAAC;CACrB;AAED;;0DAE0D;AAC1D,wBAAgB,oBAAoB,CAAC,UAAU,EAAE,MAAM,GAAG,aAAa,CAOtE;AAED;;;;;;;;GAQG;AACH,wBAAgB,kBAAkB,CAAC,KAAK,EAAE,aAAa,GAAG,IAAI,CAK7D;AA4CD;;;;;;;GAOG;AACH,wBAAgB,uBAAuB,CAAC,eAAe,EAAE,OAAO,EAAE,QAAQ,EAAE,cAAc,GAAG,OAAO,CAEnG;AAED;;;;;;GAMG;AACH,wBAAgB,yBAAyB,CACvC,eAAe,EAAE,OAAO,EACxB,QAAQ,EAAE,cAAc,EACxB,OAAO,EAAE,mBAAmB,GAAG,SAAS,GACvC,OAAO,CAET;AAED;;;;;;GAMG;AACH,MAAM,WAAW,cAAc;IAC7B,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;CAClB;AAED,uHAAuH;AACvH,wBAAgB,iBAAiB,CAAC,GAAG,EAAE,cAAc,EAAE,YAAY,EAAE,MAAM,GAAG,SAAS,GAAG,IAAI,CAI7F;AAED;;;;GAIG;AACH,wBAAgB,mBAAmB,CAAC,IAAI,EAAE,cAAc,GAAG,SAAS,GAAG,MAAM,CAG5E;AAED;;;;;;GAMG;AACH,wBAAgB,eAAe,CAC7B,OAAO,EAAE,cAAc,EACvB,QAAQ,EAAE,cAAc,EACxB,kBAAkB,EAAE,OAAO,GAC1B,MAAM,CAIR;AAED;;;;;;GAMG;AACH,wBAAgB,wBAAwB,CACtC,OAAO,EAAE,cAAc,EACvB,QAAQ,EAAE,cAAc,EACxB,kBAAkB,EAAE,OAAO,EAC3B,OAAO,EAAE,mBAAmB,GAAG,SAAS,EACxC,IAAI,CAAC,EAAE,cAAc,GACpB,MAAM,CAKR;AAgDD;;;GAGG;AACH,wBAAsB,mBAAmB,CAAC,IAAI,EAAE,iBAAiB,GAAG,OAAO,CAAC,gBAAgB,CAAC,CAmU5F"}