@llm-dev-ops/agentics-cli 1.4.26 → 1.4.27

package/dist/bundled-agents/test-bench-agents/dist/execution/index.d.ts

@@ -0,0 +1,10 @@
+export type { SpanType, SpanStatus, ExecutionContext, ExecutionSpan, ExecutionArtifact, ExecutionResult, } from './types.js';
+export { generateSpanId, nowISO, REPO_NAME } from './types.js';
+export { SpanManager, validateExecutionContext, ExecutionContextError, ExecutionInvariantError, } from './span-manager.js';
+export { InstrumentedTestBench, executeWithSpans } from './instrumented.js';
+export { PolicyGate, PolicyGateError } from './policy-gate.js';
+export type { PolicyCheckResult } from './policy-gate.js';
+export { SimulationIntentSchema, GateDispositionSchema, PolicyRuleSchema, ExecutionPolicySchema, ExemptionRecordSchema, ExemptionRequestSchema, } from './policy.js';
+export type { SimulationIntent, GateDisposition, PolicyRule, ExecutionPolicy, ExemptionRecord, ExemptionRequest, } from './policy.js';
+export { DEFAULT_EXECUTION_POLICY } from './default-policy.js';
+//# sourceMappingURL=index.d.ts.map

package/dist/bundled-agents/test-bench-agents/dist/execution/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/execution/index.ts"],"names":[],"mappings":"AAWA,YAAY,EACV,QAAQ,EACR,UAAU,EACV,gBAAgB,EAChB,aAAa,EACb,iBAAiB,EACjB,eAAe,GAChB,MAAM,YAAY,CAAC;AACpB,OAAO,EAAE,cAAc,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,YAAY,CAAC;AAG/D,OAAO,EACL,WAAW,EACX,wBAAwB,EACxB,qBAAqB,EACrB,uBAAuB,GACxB,MAAM,mBAAmB,CAAC;AAG3B,OAAO,EAAE,qBAAqB,EAAE,gBAAgB,EAAE,MAAM,mBAAmB,CAAC;AAG5E,OAAO,EAAE,UAAU,EAAE,eAAe,EAAE,MAAM,kBAAkB,CAAC;AAC/D,YAAY,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AAC1D,OAAO,EACL,sBAAsB,EACtB,qBAAqB,EACrB,gBAAgB,EAChB,qBAAqB,EACrB,qBAAqB,EACrB,sBAAsB,GACvB,MAAM,aAAa,CAAC;AACrB,YAAY,EACV,gBAAgB,EAChB,eAAe,EACf,UAAU,EACV,eAAe,EACf,eAAe,EACf,gBAAgB,GACjB,MAAM,aAAa,CAAC;AACrB,OAAO,EAAE,wBAAwB,EAAE,MAAM,qBAAqB,CAAC"}

package/dist/bundled-agents/test-bench-agents/dist/execution/index.js

@@ -0,0 +1,7 @@
+export { generateSpanId, nowISO, REPO_NAME } from './types.js';
+export { SpanManager, validateExecutionContext, ExecutionContextError, ExecutionInvariantError, } from './span-manager.js';
+export { InstrumentedTestBench, executeWithSpans } from './instrumented.js';
+export { PolicyGate, PolicyGateError } from './policy-gate.js';
+export { SimulationIntentSchema, GateDispositionSchema, PolicyRuleSchema, ExecutionPolicySchema, ExemptionRecordSchema, ExemptionRequestSchema, } from './policy.js';
+export { DEFAULT_EXECUTION_POLICY } from './default-policy.js';
+//# sourceMappingURL=index.js.map

package/dist/bundled-agents/test-bench-agents/dist/execution/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/execution/index.ts"],"names":[],"mappings":"AAmBA,OAAO,EAAE,cAAc,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,YAAY,CAAC;AAG/D,OAAO,EACL,WAAW,EACX,wBAAwB,EACxB,qBAAqB,EACrB,uBAAuB,GACxB,MAAM,mBAAmB,CAAC;AAG3B,OAAO,EAAE,qBAAqB,EAAE,gBAAgB,EAAE,MAAM,mBAAmB,CAAC;AAG5E,OAAO,EAAE,UAAU,EAAE,eAAe,EAAE,MAAM,kBAAkB,CAAC;AAE/D,OAAO,EACL,sBAAsB,EACtB,qBAAqB,EACrB,gBAAgB,EAChB,qBAAqB,EACrB,qBAAqB,EACrB,sBAAsB,GACvB,MAAM,aAAa,CAAC;AASrB,OAAO,EAAE,wBAAwB,EAAE,MAAM,qBAAqB,CAAC"}

package/dist/bundled-agents/test-bench-agents/dist/execution/instrumented.d.ts

@@ -0,0 +1,39 @@
+import type { SDKConfig, CompletionRequest, CompletionResponse, BenchmarkConfig, BenchmarkResults, ComparisonResult, EvaluationConfig, CombinedEvaluationResults, ProviderName, ModelInfo } from '../types/index.js';
+import type { ExecutionContext, ExecutionResult } from './types.js';
+import type { ExecutionPolicy } from './policy.js';
+import { SpanManager } from './span-manager.js';
+export declare class InstrumentedTestBench {
+    private readonly inner;
+    private readonly gate;
+    constructor(config?: SDKConfig, policy?: ExecutionPolicy);
+    version(ctx: ExecutionContext): Promise<ExecutionResult<string>>;
+    listModels(ctx: ExecutionContext, provider?: ProviderName): Promise<ExecutionResult<ModelInfo[]>>;
+    benchmark(ctx: ExecutionContext, options: {
+        provider: ProviderName;
+        model: string;
+        prompts: string[];
+        config?: BenchmarkConfig;
+    }): Promise<ExecutionResult<BenchmarkResults>>;
+    compare(ctx: ExecutionContext, options: {
+        models: Array<{
+            provider: ProviderName;
+            model: string;
+        }>;
+        prompts: string[];
+        config?: BenchmarkConfig;
+    }): Promise<ExecutionResult<ComparisonResult>>;
+    evaluate(ctx: ExecutionContext, text: string, config: EvaluationConfig): Promise<ExecutionResult<CombinedEvaluationResults>>;
+    complete(ctx: ExecutionContext, request: CompletionRequest): Promise<ExecutionResult<CompletionResponse>>;
+    optimize(ctx: ExecutionContext, options: {
+        prompts: string[];
+        metric: 'latency' | 'cost' | 'quality';
+        maxCost?: number;
+        minQuality?: number;
+    }): Promise<ExecutionResult<{
+        provider: ProviderName;
+        model: string;
+        reason: string;
+    }>>;
+}
+export declare function executeWithSpans<T>(ctx: ExecutionContext, fn: (mgr: SpanManager) => Promise<T>): Promise<ExecutionResult<T>>;
+//# sourceMappingURL=instrumented.d.ts.map

package/dist/bundled-agents/test-bench-agents/dist/execution/instrumented.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"instrumented.d.ts","sourceRoot":"","sources":["../../src/execution/instrumented.ts"],"names":[],"mappings":"AAaA,OAAO,KAAK,EACV,SAAS,EACT,iBAAiB,EACjB,kBAAkB,EAClB,eAAe,EACf,gBAAgB,EAChB,gBAAgB,EAChB,gBAAgB,EAChB,yBAAyB,EACzB,YAAY,EACZ,SAAS,EACV,MAAM,mBAAmB,CAAC;AAE3B,OAAO,KAAK,EAAE,gBAAgB,EAAE,eAAe,EAAE,MAAM,YAAY,CAAC;AACpE,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,aAAa,CAAC;AACnD,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAyBhD,qBAAa,qBAAqB;IAChC,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAe;IACrC,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAa;gBAEtB,MAAM,GAAE,SAAc,EAAE,MAAM,CAAC,EAAE,eAAe;IAQtD,OAAO,CAAC,GAAG,EAAE,gBAAgB,GAAG,OAAO,CAAC,eAAe,CAAC,MAAM,CAAC,CAAC;IAkBhE,UAAU,CACd,GAAG,EAAE,gBAAgB,EACrB,QAAQ,CAAC,EAAE,YAAY,GACtB,OAAO,CAAC,eAAe,CAAC,SAAS,EAAE,CAAC,CAAC;IAmBlC,SAAS,CACb,GAAG,EAAE,gBAAgB,EACrB,OAAO,EAAE;QACP,QAAQ,EAAE,YAAY,CAAC;QACvB,KAAK,EAAE,MAAM,CAAC;QACd,OAAO,EAAE,MAAM,EAAE,CAAC;QAClB,MAAM,CAAC,EAAE,eAAe,CAAC;KAC1B,GACA,OAAO,CAAC,eAAe,CAAC,gBAAgB,CAAC,CAAC;IA6CvC,OAAO,CACX,GAAG,EAAE,gBAAgB,EACrB,OAAO,EAAE;QACP,MAAM,EAAE,KAAK,CAAC;YAAE,QAAQ,EAAE,YAAY,CAAC;YAAC,KAAK,EAAE,MAAM,CAAA;SAAE,CAAC,CAAC;QACzD,OAAO,EAAE,MAAM,EAAE,CAAC;QAClB,MAAM,CAAC,EAAE,eAAe,CAAC;KAC1B,GACA,OAAO,CAAC,eAAe,CAAC,gBAAgB,CAAC,CAAC;IA6BvC,QAAQ,CACZ,GAAG,EAAE,gBAAgB,EACrB,IAAI,EAAE,MAAM,EACZ,MAAM,EAAE,gBAAgB,GACvB,OAAO,CAAC,eAAe,CAAC,yBAAyB,CAAC,CAAC;IA6BhD,QAAQ,CACZ,GAAG,EAAE,gBAAgB,EACrB,OAAO,EAAE,iBAAiB,GACzB,OAAO,CAAC,eAAe,CAAC,kBAAkB,CAAC,CAAC;IA4BzC,QAAQ,CACZ,GAAG,EAAE,gBAAgB,EACrB,OAAO,EAAE;QACP,OAAO,EAAE,MAAM,EAAE,CAAC;QAClB,MAAM,EAAE,SAAS,GAAG,MAAM,GAAG,SAAS,CAAC;QACvC,OAAO,CAAC,EAAE,MAAM,CAAC;QACjB,UAAU,CAAC,EAAE,MAAM,CAAC;KACrB,GACA,OAAO,CACR,eAAe,CAAC;QAAE,QAAQ,EAAE,YAAY,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,CAAC,CAC3E;CA6BF;AA0BD,wBAAsB,gBAAgB,CAAC,CAAC,EACtC,GAAG,EAAE,gBAAgB,EACrB,EAAE,EAAE,CAAC,GAAG,EAAE,WAAW,KAAK,OAAO,CAAC,CAAC,CAAC,GACnC,OAAO,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC,CAU7B"}

package/dist/bundled-agents/test-bench-agents/dist/execution/instrumented.js

@@ -0,0 +1,178 @@
+import { LLMTestBench } from '../core/llm-test-bench.js';
+import { SpanManager } from './span-manager.js';
+import { PolicyGate } from './policy-gate.js';
+export class InstrumentedTestBench {
+    inner;
+    gate;
+    constructor(config = {}, policy) {
+        this.inner = new LLMTestBench(config);
+        this.gate = new PolicyGate(policy);
+    }
+    async version(ctx) {
+        const { mgr } = this.gate.enforce('benchmark', ctx);
+        const agent = mgr.startAgent('version-resolver');
+        try {
+            const version = await this.inner.version();
+            mgr.endAgent(agent.span_id, 'completed');
+            return mgr.finalize(version);
+        }
+        catch (err) {
+            const reason = err instanceof Error ? err.message : String(err);
+            mgr.endAgent(agent.span_id, 'failed', [reason]);
+            return mgr.finalize(undefined, 'failed');
+        }
+    }
+    async listModels(ctx, provider) {
+        const { mgr } = this.gate.enforce('benchmark', ctx);
+        const agent = mgr.startAgent('model-lister');
+        try {
+            const models = await this.inner.listModels(provider);
+            mgr.endAgent(agent.span_id, 'completed');
+            return mgr.finalize(models);
+        }
+        catch (err) {
+            const reason = err instanceof Error ? err.message : String(err);
+            mgr.endAgent(agent.span_id, 'failed', [reason]);
+            return mgr.finalize(undefined, 'failed');
+        }
+    }
+    async benchmark(ctx, options) {
+        const { mgr } = this.gate.enforce('benchmark', ctx);
+        const validator = mgr.startAgent('input-validator');
+        try {
+            if (!options.provider || !options.model || !options.prompts?.length) {
+                throw new Error('Missing required benchmark parameters');
+            }
+            mgr.endAgent(validator.span_id, 'completed');
+        }
+        catch (err) {
+            const reason = err instanceof Error ? err.message : String(err);
+            mgr.endAgent(validator.span_id, 'failed', [reason]);
+            return mgr.finalize(undefined, 'failed');
+        }
+        const runner = mgr.startAgent('benchmark-runner');
+        try {
+            const results = await this.inner.benchmark(options);
+            mgr.addArtifact({
+                reference: `benchmark-${mgr.repoSpanId}-results.json`,
+                artifact_type: 'benchmark_results',
+                agent_span_id: runner.span_id,
+                metadata: {
+                    provider: options.provider,
+                    model: options.model,
+                    prompt_count: options.prompts.length,
+                },
+            });
+            mgr.endAgent(runner.span_id, 'completed');
+            return mgr.finalize(results);
+        }
+        catch (err) {
+            const reason = err instanceof Error ? err.message : String(err);
+            mgr.endAgent(runner.span_id, 'failed', [reason]);
+            return mgr.finalize(undefined, 'failed');
+        }
+    }
+    async compare(ctx, options) {
+        const { mgr } = this.gate.enforce('compare', ctx);
+        const agent = mgr.startAgent('model-comparator');
+        try {
+            const result = await this.inner.compare(options);
+            mgr.addArtifact({
+                reference: `comparison-${mgr.repoSpanId}-results.json`,
+                artifact_type: 'comparison_results',
+                agent_span_id: agent.span_id,
+                metadata: {
+                    model_count: options.models.length,
+                    prompt_count: options.prompts.length,
+                },
+            });
+            mgr.endAgent(agent.span_id, 'completed');
+            return mgr.finalize(result);
+        }
+        catch (err) {
+            const reason = err instanceof Error ? err.message : String(err);
+            mgr.endAgent(agent.span_id, 'failed', [reason]);
+            return mgr.finalize(undefined, 'failed');
+        }
+    }
+    async evaluate(ctx, text, config) {
+        const { mgr } = this.gate.enforce('evaluate', ctx);
+        const agent = mgr.startAgent('evaluator');
+        try {
+            const result = await this.inner.evaluate(text, config);
+            mgr.addArtifact({
+                reference: `evaluation-${mgr.repoSpanId}-results.json`,
+                artifact_type: 'evaluation_results',
+                agent_span_id: agent.span_id,
+                metadata: {
+                    evaluators: config.evaluators,
+                    text_length: text.length,
+                },
+            });
+            mgr.endAgent(agent.span_id, 'completed');
+            return mgr.finalize(result);
+        }
+        catch (err) {
+            const reason = err instanceof Error ? err.message : String(err);
+            mgr.endAgent(agent.span_id, 'failed', [reason]);
+            return mgr.finalize(undefined, 'failed');
+        }
+    }
+    async complete(ctx, request) {
+        const { mgr } = this.gate.enforce('complete', ctx);
+        const agent = mgr.startAgent('completion-agent');
+        try {
+            const result = await this.inner.complete(request);
+            mgr.addArtifact({
+                reference: `completion-${mgr.repoSpanId}-response.json`,
+                artifact_type: 'completion_response',
+                agent_span_id: agent.span_id,
+                metadata: {
+                    model: request.model,
+                },
+            });
+            mgr.endAgent(agent.span_id, 'completed');
+            return mgr.finalize(result);
+        }
+        catch (err) {
+            const reason = err instanceof Error ? err.message : String(err);
+            mgr.endAgent(agent.span_id, 'failed', [reason]);
+            return mgr.finalize(undefined, 'failed');
+        }
+    }
+    async optimize(ctx, options) {
+        const { mgr } = this.gate.enforce('optimize', ctx);
+        const agent = mgr.startAgent('optimizer');
+        try {
+            const result = await this.inner.optimize(options);
+            mgr.addArtifact({
+                reference: `optimization-${mgr.repoSpanId}-recommendation.json`,
+                artifact_type: 'optimization_recommendation',
+                agent_span_id: agent.span_id,
+                metadata: {
+                    metric: options.metric,
+                    prompt_count: options.prompts.length,
+                },
+            });
+            mgr.endAgent(agent.span_id, 'completed');
+            return mgr.finalize(result);
+        }
+        catch (err) {
+            const reason = err instanceof Error ? err.message : String(err);
+            mgr.endAgent(agent.span_id, 'failed', [reason]);
+            return mgr.finalize(undefined, 'failed');
+        }
+    }
+}
+export async function executeWithSpans(ctx, fn) {
+    const mgr = SpanManager.create(ctx);
+    try {
+        const data = await fn(mgr);
+        return mgr.finalize(data);
+    }
+    catch (err) {
+        const reason = err instanceof Error ? err.message : String(err);
+        return mgr.finalizeFailure([reason]);
+    }
+}
+//# sourceMappingURL=instrumented.js.map

package/dist/bundled-agents/test-bench-agents/dist/execution/instrumented.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"instrumented.js","sourceRoot":"","sources":["../../src/execution/instrumented.ts"],"names":[],"mappings":"AAyBA,OAAO,EAAE,YAAY,EAAE,MAAM,2BAA2B,CAAC;AAGzD,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,EAAE,UAAU,EAAE,MAAM,kBAAkB,CAAC;AAwB9C,MAAM,OAAO,qBAAqB;IACf,KAAK,CAAe;IACpB,IAAI,CAAa;IAElC,YAAY,SAAoB,EAAE,EAAE,MAAwB;QAC1D,IAAI,CAAC,KAAK,GAAG,IAAI,YAAY,CAAC,MAAM,CAAC,CAAC;QACtC,IAAI,CAAC,IAAI,GAAG,IAAI,UAAU,CAAC,MAAM,CAAC,CAAC;IACrC,CAAC;IAKD,KAAK,CAAC,OAAO,CAAC,GAAqB;QACjC,MAAM,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,WAAW,EAAE,GAAG,CAAC,CAAC;QACpD,MAAM,KAAK,GAAG,GAAG,CAAC,UAAU,CAAC,kBAAkB,CAAC,CAAC;QAEjD,IAAI,CAAC;YACH,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;YAC3C,GAAG,CAAC,QAAQ,CAAC,KAAK,CAAC,OAAO,EAAE,WAAW,CAAC,CAAC;YACzC,OAAO,GAAG,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;QAC/B,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,MAAM,MAAM,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAChE,GAAG,CAAC,QAAQ,CAAC,KAAK,CAAC,OAAO,EAAE,QAAQ,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC;YAChD,OAAO,GAAG,CAAC,QAAQ,CAAS,SAAS,EAAE,QAAQ,CAAC,CAAC;QACnD,CAAC;IACH,CAAC;IAKD,KAAK,CAAC,UAAU,CACd,GAAqB,EACrB,QAAuB;QAEvB,MAAM,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,WAAW,EAAE,GAAG,CAAC,CAAC;QACpD,MAAM,KAAK,GAAG,GAAG,CAAC,UAAU,CAAC,cAAc,CAAC,CAAC;QAE7C,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,UAAU,CAAC,QAAQ,CAAC,CAAC;YACrD,GAAG,CAAC,QAAQ,CAAC,KAAK,CAAC,OAAO,EAAE,WAAW,CAAC,CAAC;YACzC,OAAO,GAAG,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;QAC9B,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,MAAM,MAAM,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAChE,GAAG,CAAC,QAAQ,CAAC,KAAK,CAAC,OAAO,EAAE,QAAQ,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC;YAChD,OAAO,GAAG,CAAC,QAAQ,CAAc,SAAS,EAAE,QAAQ,CAAC,CAAC;QACxD,CAAC;IACH,CAAC;IAMD,KAAK,CAAC,SAAS,CACb,GAAqB,EACrB,OAKC;QAED,MAAM,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,WAAW,EAAE,GAAG,CAAC,CAAC;QAGpD,MAAM,SAAS,GAAG,GAAG,CAAC,UAAU,CAAC,iBAAiB,CAAC,CAAC;QACpD,IAAI,CAAC;YACH,IAAI,CAAC,OAAO,CAAC,QAAQ,IAAI,CAAC,OAAO,CAAC,KAAK,IAAI,CAAC,OAAO,CAAC,OAAO,EAAE,MAAM,EAAE,CAAC;gBACpE,MAAM,IAAI,KAAK,CAAC,uCAAuC,CAAC,CAAC;YAC3D,CAAC;YACD,GAAG,CAAC,QAAQ,CAAC,SAAS,CAAC,OAAO,EAAE,WAAW,CAAC,CAAC;QAC/C,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,MAAM,MAAM,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAChE,GAAG,CAAC,QAAQ,CAAC,SAAS,CAAC,OAAO,EAAE,QAAQ,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC;YACpD,OAAO,GAAG,CAAC,QAAQ,CAAmB,SAAS,EAAE,QAAQ,CAAC,CAAC;QAC7D,CAAC;QAGD,MAAM,MAAM,GAAG,GAAG,CAAC,UAAU,CAAC,kBAAkB,CAAC,CAAC;QAClD,IAAI,CAAC;YACH,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC;YAGpD,GAAG,CAAC,WAAW,CAAC;gBACd,SAAS,EAAE,aAAa,GAAG,CAAC,UAAU,eAAe;gBACrD,aAAa,EAAE,mBAAmB;gBAClC,aAAa,EAAE,MAAM,CAAC,OAAO;gBAC7B,QAAQ,EAAE;oBACR,QAAQ,EAAE,OAAO,CAAC,QAAQ;oBAC1B,KAAK,EAAE,OAAO,CAAC,KAAK;oBACpB,YAAY,EAAE,OAAO,CAAC,OAAO,CAAC,MAAM;iBACrC;aACF,CAAC,CAAC;YAEH,GAAG,CAAC,QAAQ,CAAC,MAAM,CAAC,OAAO,EAAE,WAAW,CAAC,CAAC;YAC1C,OAAO,GAAG,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;QAC/B,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,MAAM,MAAM,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAChE,GAAG,CAAC,QAAQ,CAAC,MAAM,CAAC,OAAO,EAAE,QAAQ,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC;YACjD,OAAO,GAAG,CAAC,QAAQ,CAAmB,SAAS,EAAE,QAAQ,CAAC,CAAC;QAC7D,CAAC;IACH,CAAC;IAKD,KAAK,CAAC,OAAO,CACX,GAAqB,EACrB,OAIC;QAED,MAAM,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,SAAS,EAAE,GAAG,CAAC,CAAC;QAClD,MAAM,KAAK,GAAG,GAAG,CAAC,UAAU,CAAC,kBAAkB,CAAC,CAAC;QAEjD,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;YAEjD,GAAG,CAAC,WAAW,CAAC;gBACd,SAAS,EAAE,cAAc,GAAG,CAAC,UAAU,eAAe;gBACtD,aAAa,EAAE,oBAAoB;gBACnC,aAAa,EAAE,KAAK,CAAC,OAAO;gBAC5B,QAAQ,EAAE;oBACR,WAAW,EAAE,OAAO,CAAC,MAAM,CAAC,MAAM;oBAClC,YAAY,EAAE,OAAO,CAAC,OAAO,CAAC,MAAM;iBACrC;aACF,CAAC,CAAC;YAEH,GAAG,CAAC,QAAQ,CAAC,KAAK,CAAC,OAAO,EAAE,WAAW,CAAC,CAAC;YACzC,OAAO,GAAG,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;QAC9B,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,MAAM,MAAM,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAChE,GAAG,CAAC,QAAQ,CAAC,KAAK,CAAC,OAAO,EAAE,QAAQ,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC;YAChD,OAAO,GAAG,CAAC,QAAQ,CAAmB,SAAS,EAAE,QAAQ,CAAC,CAAC;QAC7D,CAAC;IACH,CAAC;IAKD,KAAK,CAAC,QAAQ,CACZ,GAAqB,EACrB,IAAY,EACZ,MAAwB;QAExB,MAAM,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,UAAU,EAAE,GAAG,CAAC,CAAC;QACnD,MAAM,KAAK,GAAG,GAAG,CAAC,UAAU,CAAC,WAAW,CAAC,CAAC;QAE1C,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;YAEvD,GAAG,CAAC,WAAW,CAAC;gBACd,SAAS,EAAE,cAAc,GAAG,CAAC,UAAU,eAAe;gBACtD,aAAa,EAAE,oBAAoB;gBACnC,aAAa,EAAE,KAAK,CAAC,OAAO;gBAC5B,QAAQ,EAAE;oBACR,UAAU,EAAE,MAAM,CAAC,UAAU;oBAC7B,WAAW,EAAE,IAAI,CAAC,MAAM;iBACzB;aACF,CAAC,CAAC;YAEH,GAAG,CAAC,QAAQ,CAAC,KAAK,CAAC,OAAO,EAAE,WAAW,CAAC,CAAC;YACzC,OAAO,GAAG,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;QAC9B,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,MAAM,MAAM,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAChE,GAAG,CAAC,QAAQ,CAAC,KAAK,CAAC,OAAO,EAAE,QAAQ,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC;YAChD,OAAO,GAAG,CAAC,QAAQ,CAA4B,SAAS,EAAE,QAAQ,CAAC,CAAC;QACtE,CAAC;IACH,CAAC;IAKD,KAAK,CAAC,QAAQ,CACZ,GAAqB,EACrB,OAA0B;QAE1B,MAAM,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,UAAU,EAAE,GAAG,CAAC,CAAC;QACnD,MAAM,KAAK,GAAG,GAAG,CAAC,UAAU,CAAC,kBAAkB,CAAC,CAAC;QAEjD,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;YAElD,GAAG,CAAC,WAAW,CAAC;gBACd,SAAS,EAAE,cAAc,GAAG,CAAC,UAAU,gBAAgB;gBACvD,aAAa,EAAE,qBAAqB;gBACpC,aAAa,EAAE,KAAK,CAAC,OAAO;gBAC5B,QAAQ,EAAE;oBACR,KAAK,EAAE,OAAO,CAAC,KAAK;iBACrB;aACF,CAAC,CAAC;YAEH,GAAG,CAAC,QAAQ,CAAC,KAAK,CAAC,OAAO,EAAE,WAAW,CAAC,CAAC;YACzC,OAAO,GAAG,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;QAC9B,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,MAAM,MAAM,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAChE,GAAG,CAAC,QAAQ,CAAC,KAAK,CAAC,OAAO,EAAE,QAAQ,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC;YAChD,OAAO,GAAG,CAAC,QAAQ,CAAqB,SAAS,EAAE,QAAQ,CAAC,CAAC;QAC/D,CAAC;IACH,CAAC;IAKD,KAAK,CAAC,QAAQ,CACZ,GAAqB,EACrB,OAKC;QAID,MAAM,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,UAAU,EAAE,GAAG,CAAC,CAAC;QACnD,MAAM,KAAK,GAAG,GAAG,CAAC,UAAU,CAAC,WAAW,CAAC,CAAC;QAE1C,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;YAElD,GAAG,CAAC,WAAW,CAAC;gBACd,SAAS,EAAE,gBAAgB,GAAG,CAAC,UAAU,sBAAsB;gBAC/D,aAAa,EAAE,6BAA6B;gBAC5C,aAAa,EAAE,KAAK,CAAC,OAAO;gBAC5B,QAAQ,EAAE;oBACR,MAAM,EAAE,OAAO,CAAC,MAAM;oBACtB,YAAY,EAAE,OAAO,CAAC,OAAO,CAAC,MAAM;iBACrC;aACF,CAAC,CAAC;YAEH,GAAG,CAAC,QAAQ,CAAC,KAAK,CAAC,OAAO,EAAE,WAAW,CAAC,CAAC;YACzC,OAAO,GAAG,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;QAC9B,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,MAAM,MAAM,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAChE,GAAG,CAAC,QAAQ,CAAC,KAAK,CAAC,OAAO,EAAE,QAAQ,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC;YAChD,OAAO,GAAG,CAAC,QAAQ,CAIhB,SAAS,EAAE,QAAQ,CAAC,CAAC;QAC1B,CAAC;IACH,CAAC;CACF;AA0BD,MAAM,CAAC,KAAK,UAAU,gBAAgB,CACpC,GAAqB,EACrB,EAAoC;IAEpC,MAAM,GAAG,GAAG,WAAW,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;IAEpC,IAAI,CAAC;QACH,MAAM,IAAI,GAAG,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC;QAC3B,OAAO,GAAG,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;IAC5B,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,MAAM,MAAM,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;QAChE,OAAO,GAAG,CAAC,eAAe,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC;IACvC,CAAC;AACH,CAAC"}

package/dist/bundled-agents/test-bench-agents/dist/execution/policy-gate.d.ts

@@ -0,0 +1,25 @@
+import type { ExecutionContext } from './types.js';
+import type { SimulationIntent, GateDisposition, ExecutionPolicy, ExemptionRequest, ExemptionRecord } from './policy.js';
+import { SpanManager } from './span-manager.js';
+export declare class PolicyGateError extends Error {
+    readonly intent: SimulationIntent;
+    readonly disposition: GateDisposition;
+    constructor(message: string, intent: SimulationIntent, disposition: GateDisposition);
+}
+export interface PolicyCheckResult {
+    allowed: boolean;
+    disposition: GateDisposition;
+    exemption?: ExemptionRecord;
+    error?: string;
+}
+export declare class PolicyGate {
+    private readonly policy;
+    constructor(policy?: ExecutionPolicy);
+    getPolicy(): Readonly<ExecutionPolicy>;
+    check(intent: SimulationIntent, ctx: ExecutionContext | null, exemption?: ExemptionRequest): PolicyCheckResult;
+    enforce(intent: SimulationIntent, ctx: ExecutionContext): {
+        mgr: SpanManager;
+    };
+    recordExemption(mgr: SpanManager, agentSpanId: string, exemption: ExemptionRecord): void;
+}
+//# sourceMappingURL=policy-gate.d.ts.map

package/dist/bundled-agents/test-bench-agents/dist/execution/policy-gate.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"policy-gate.d.ts","sourceRoot":"","sources":["../../src/execution/policy-gate.ts"],"names":[],"mappings":"AASA,OAAO,KAAK,EAAE,gBAAgB,EAAqB,MAAM,YAAY,CAAC;AACtE,OAAO,KAAK,EACV,gBAAgB,EAChB,eAAe,EACf,eAAe,EACf,gBAAgB,EAChB,eAAe,EAChB,MAAM,aAAa,CAAC;AACrB,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAOhD,qBAAa,eAAgB,SAAQ,KAAK;IACxC,SAAgB,MAAM,EAAE,gBAAgB,CAAC;IACzC,SAAgB,WAAW,EAAE,eAAe,CAAC;gBAG3C,OAAO,EAAE,MAAM,EACf,MAAM,EAAE,gBAAgB,EACxB,WAAW,EAAE,eAAe;CAO/B;AAKD,MAAM,WAAW,iBAAiB;IAChC,OAAO,EAAE,OAAO,CAAC;IACjB,WAAW,EAAE,eAAe,CAAC;IAC7B,SAAS,CAAC,EAAE,eAAe,CAAC;IAC5B,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AASD,qBAAa,UAAU;IACrB,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAkB;gBAE7B,MAAM,CAAC,EAAE,eAAe;IAKpC,SAAS,IAAI,QAAQ,CAAC,eAAe,CAAC;IAWtC,KAAK,CACH,MAAM,EAAE,gBAAgB,EACxB,GAAG,EAAE,gBAAgB,GAAG,IAAI,EAC5B,SAAS,CAAC,EAAE,gBAAgB,GAC3B,iBAAiB;IAqDpB,OAAO,CACL,MAAM,EAAE,gBAAgB,EACxB,GAAG,EAAE,gBAAgB,GACpB;QAAE,GAAG,EAAE,WAAW,CAAA;KAAE;IAevB,eAAe,CACb,GAAG,EAAE,WAAW,EAChB,WAAW,EAAE,MAAM,EACnB,SAAS,EAAE,eAAe,GACzB,IAAI;CAUR"}

package/dist/bundled-agents/test-bench-agents/dist/execution/policy-gate.js

@@ -0,0 +1,75 @@
+import { SpanManager } from './span-manager.js';
+import { DEFAULT_EXECUTION_POLICY } from './default-policy.js';
+import { nowISO } from './types.js';
+export class PolicyGateError extends Error {
+    intent;
+    disposition;
+    constructor(message, intent, disposition) {
+        super(message);
+        this.name = 'PolicyGateError';
+        this.intent = intent;
+        this.disposition = disposition;
+    }
+}
+export class PolicyGate {
+    policy;
+    constructor(policy) {
+        this.policy = policy ?? DEFAULT_EXECUTION_POLICY;
+    }
+    getPolicy() {
+        return this.policy;
+    }
+    check(intent, ctx, exemption) {
+        const rule = this.policy.rules.find((r) => r.intent === intent);
+        const disposition = rule?.disposition ?? this.policy.default_disposition;
+        if (ctx !== null) {
+            return { allowed: true, disposition };
+        }
+        switch (disposition) {
+            case 'optional':
+                return { allowed: true, disposition };
+            case 'recommended':
+                return { allowed: true, disposition };
+            case 'mandatory': {
+                if (!exemption) {
+                    return {
+                        allowed: false,
+                        disposition,
+                        error: `Intent '${intent}' has mandatory policy gate. ` +
+                            `Provide an ExecutionContext or an ExemptionRequest. ` +
+                            `Rationale: ${rule?.rationale ?? 'policy default'}`,
+                    };
+                }
+                const record = {
+                    intent,
+                    reason: exemption.reason,
+                    granted_by: exemption.granted_by,
+                    granted_at: nowISO(),
+                    execution_id: 'EXEMPTED',
+                    parent_span_id: 'EXEMPTED',
+                };
+                return { allowed: true, disposition, exemption: record };
+            }
+            default:
+                return { allowed: true, disposition };
+        }
+    }
+    enforce(intent, ctx) {
+        const result = this.check(intent, ctx);
+        if (!result.allowed) {
+            throw new PolicyGateError(result.error ?? 'Policy gate check failed', intent, result.disposition);
+        }
+        return { mgr: SpanManager.create(ctx) };
+    }
+    recordExemption(mgr, agentSpanId, exemption) {
+        const artifact = {
+            reference: `policy-exemption-${exemption.intent}-${Date.now()}.json`,
+            artifact_type: 'policy_exemption',
+            agent_span_id: agentSpanId,
+            metadata: { ...exemption },
+        };
+        mgr.addArtifact(artifact);
+        mgr.addExemption(exemption);
+    }
+}
+//# sourceMappingURL=policy-gate.js.map

package/dist/bundled-agents/test-bench-agents/dist/execution/policy-gate.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"policy-gate.js","sourceRoot":"","sources":["../../src/execution/policy-gate.ts"],"names":[],"mappings":"AAiBA,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,EAAE,wBAAwB,EAAE,MAAM,qBAAqB,CAAC;AAC/D,OAAO,EAAE,MAAM,EAAE,MAAM,YAAY,CAAC;AAKpC,MAAM,OAAO,eAAgB,SAAQ,KAAK;IACxB,MAAM,CAAmB;IACzB,WAAW,CAAkB;IAE7C,YACE,OAAe,EACf,MAAwB,EACxB,WAA4B;QAE5B,KAAK,CAAC,OAAO,CAAC,CAAC;QACf,IAAI,CAAC,IAAI,GAAG,iBAAiB,CAAC;QAC9B,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;QACrB,IAAI,CAAC,WAAW,GAAG,WAAW,CAAC;IACjC,CAAC;CACF;AAmBD,MAAM,OAAO,UAAU;IACJ,MAAM,CAAkB;IAEzC,YAAY,MAAwB;QAClC,IAAI,CAAC,MAAM,GAAG,MAAM,IAAI,wBAAwB,CAAC;IACnD,CAAC;IAGD,SAAS;QACP,OAAO,IAAI,CAAC,MAAM,CAAC;IACrB,CAAC;IASD,KAAK,CACH,MAAwB,EACxB,GAA4B,EAC5B,SAA4B;QAE5B,MAAM,IAAI,GAAG,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAqB,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,MAAM,CAAC,CAAC;QACpF,MAAM,WAAW,GAAG,IAAI,EAAE,WAAW,IAAI,IAAI,CAAC,MAAM,CAAC,mBAAmB,CAAC;QAGzE,IAAI,GAAG,KAAK,IAAI,EAAE,CAAC;YACjB,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,WAAW,EAAE,CAAC;QACxC,CAAC;QAGD,QAAQ,WAAW,EAAE,CAAC;YACpB,KAAK,UAAU;gBACb,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,WAAW,EAAE,CAAC;YAExC,KAAK,aAAa;gBAChB,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,WAAW,EAAE,CAAC;YAExC,KAAK,WAAW,CAAC,CAAC,CAAC;gBACjB,IAAI,CAAC,SAAS,EAAE,CAAC;oBACf,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,WAAW;wBACX,KAAK,EACH,WAAW,MAAM,+BAA+B;4BAChD,sDAAsD;4BACtD,cAAc,IAAI,EAAE,SAAS,IAAI,gBAAgB,EAAE;qBACtD,CAAC;gBACJ,CAAC;gBAED,MAAM,MAAM,GAAoB;oBAC9B,MAAM;oBACN,MAAM,EAAE,SAAS,CAAC,MAAM;oBACxB,UAAU,EAAE,SAAS,CAAC,UAAU;oBAChC,UAAU,EAAE,MAAM,EAAE;oBACpB,YAAY,EAAE,UAAU;oBACxB,cAAc,EAAE,UAAU;iBAC3B,CAAC;gBAEF,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,WAAW,EAAE,SAAS,EAAE,MAAM,EAAE,CAAC;YAC3D,CAAC;YAED;gBACE,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,WAAW,EAAE,CAAC;QAC1C,CAAC;IACH,CAAC;IASD,OAAO,CACL,MAAwB,EACxB,GAAqB;QAErB,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;QACvC,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;YACpB,MAAM,IAAI,eAAe,CACvB,MAAM,CAAC,KAAK,IAAI,0BAA0B,EAC1C,MAAM,EACN,MAAM,CAAC,WAAW,CACnB,CAAC;QACJ,CAAC;QACD,OAAO,EAAE,GAAG,EAAE,WAAW,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC;IAC1C,CAAC;IAKD,eAAe,CACb,GAAgB,EAChB,WAAmB,EACnB,SAA0B;QAE1B,MAAM,QAAQ,GAAsB;YAClC,SAAS,EAAE,oBAAoB,SAAS,CAAC,MAAM,IAAI,IAAI,CAAC,GAAG,EAAE,OAAO;YACpE,aAAa,EAAE,kBAAkB;YACjC,aAAa,EAAE,WAAW;YAC1B,QAAQ,EAAE,EAAE,GAAG,SAAS,EAAE;SAC3B,CAAC;QACF,GAAG,CAAC,WAAW,CAAC,QAAQ,CAAC,CAAC;QAC1B,GAAG,CAAC,YAAY,CAAC,SAAS,CAAC,CAAC;IAC9B,CAAC;CACF"}

package/dist/bundled-agents/test-bench-agents/dist/execution/policy.d.ts

@@ -0,0 +1,108 @@
+import { z } from 'zod';
+export declare const SimulationIntentSchema: z.ZodEnum<["benchmark", "compare", "evaluate", "complete", "optimize", "load_ramp", "spike", "soak", "extreme_input", "adversarial", "rate_limit_probe", "timeout_boundary", "token_limit", "context_overflow", "malformed_request"]>;
+export type SimulationIntent = z.infer<typeof SimulationIntentSchema>;
+export declare const GateDispositionSchema: z.ZodEnum<["mandatory", "recommended", "optional"]>;
+export type GateDisposition = z.infer<typeof GateDispositionSchema>;
+export declare const PolicyRuleSchema: z.ZodObject<{
+    intent: z.ZodEnum<["benchmark", "compare", "evaluate", "complete", "optimize", "load_ramp", "spike", "soak", "extreme_input", "adversarial", "rate_limit_probe", "timeout_boundary", "token_limit", "context_overflow", "malformed_request"]>;
+    disposition: z.ZodEnum<["mandatory", "recommended", "optional"]>;
+    rationale: z.ZodString;
+}, "strip", z.ZodTypeAny, {
+    intent: "compare" | "complete" | "optimize" | "benchmark" | "evaluate" | "load_ramp" | "spike" | "soak" | "extreme_input" | "adversarial" | "rate_limit_probe" | "timeout_boundary" | "token_limit" | "context_overflow" | "malformed_request";
+    disposition: "mandatory" | "recommended" | "optional";
+    rationale: string;
+}, {
+    intent: "compare" | "complete" | "optimize" | "benchmark" | "evaluate" | "load_ramp" | "spike" | "soak" | "extreme_input" | "adversarial" | "rate_limit_probe" | "timeout_boundary" | "token_limit" | "context_overflow" | "malformed_request";
+    disposition: "mandatory" | "recommended" | "optional";
+    rationale: string;
+}>;
+export type PolicyRule = z.infer<typeof PolicyRuleSchema>;
+export declare const ExecutionPolicySchema: z.ZodObject<{
+    policy_version: z.ZodString;
+    updated_at: z.ZodString;
+    description: z.ZodString;
+    rules: z.ZodEffects<z.ZodArray<z.ZodObject<{
+        intent: z.ZodEnum<["benchmark", "compare", "evaluate", "complete", "optimize", "load_ramp", "spike", "soak", "extreme_input", "adversarial", "rate_limit_probe", "timeout_boundary", "token_limit", "context_overflow", "malformed_request"]>;
+        disposition: z.ZodEnum<["mandatory", "recommended", "optional"]>;
+        rationale: z.ZodString;
+    }, "strip", z.ZodTypeAny, {
+        intent: "compare" | "complete" | "optimize" | "benchmark" | "evaluate" | "load_ramp" | "spike" | "soak" | "extreme_input" | "adversarial" | "rate_limit_probe" | "timeout_boundary" | "token_limit" | "context_overflow" | "malformed_request";
+        disposition: "mandatory" | "recommended" | "optional";
+        rationale: string;
+    }, {
+        intent: "compare" | "complete" | "optimize" | "benchmark" | "evaluate" | "load_ramp" | "spike" | "soak" | "extreme_input" | "adversarial" | "rate_limit_probe" | "timeout_boundary" | "token_limit" | "context_overflow" | "malformed_request";
+        disposition: "mandatory" | "recommended" | "optional";
+        rationale: string;
+    }>, "many">, {
+        intent: "compare" | "complete" | "optimize" | "benchmark" | "evaluate" | "load_ramp" | "spike" | "soak" | "extreme_input" | "adversarial" | "rate_limit_probe" | "timeout_boundary" | "token_limit" | "context_overflow" | "malformed_request";
+        disposition: "mandatory" | "recommended" | "optional";
+        rationale: string;
+    }[], {
+        intent: "compare" | "complete" | "optimize" | "benchmark" | "evaluate" | "load_ramp" | "spike" | "soak" | "extreme_input" | "adversarial" | "rate_limit_probe" | "timeout_boundary" | "token_limit" | "context_overflow" | "malformed_request";
+        disposition: "mandatory" | "recommended" | "optional";
+        rationale: string;
+    }[]>;
+    default_disposition: z.ZodEnum<["mandatory", "recommended", "optional"]>;
+}, "strip", z.ZodTypeAny, {
+    policy_version: string;
+    updated_at: string;
+    description: string;
+    rules: {
+        intent: "compare" | "complete" | "optimize" | "benchmark" | "evaluate" | "load_ramp" | "spike" | "soak" | "extreme_input" | "adversarial" | "rate_limit_probe" | "timeout_boundary" | "token_limit" | "context_overflow" | "malformed_request";
+        disposition: "mandatory" | "recommended" | "optional";
+        rationale: string;
+    }[];
+    default_disposition: "mandatory" | "recommended" | "optional";
+}, {
+    policy_version: string;
+    updated_at: string;
+    description: string;
+    rules: {
+        intent: "compare" | "complete" | "optimize" | "benchmark" | "evaluate" | "load_ramp" | "spike" | "soak" | "extreme_input" | "adversarial" | "rate_limit_probe" | "timeout_boundary" | "token_limit" | "context_overflow" | "malformed_request";
+        disposition: "mandatory" | "recommended" | "optional";
+        rationale: string;
+    }[];
+    default_disposition: "mandatory" | "recommended" | "optional";
+}>;
+export type ExecutionPolicy = z.infer<typeof ExecutionPolicySchema>;
+export declare const ExemptionRecordSchema: z.ZodObject<{
+    intent: z.ZodEnum<["benchmark", "compare", "evaluate", "complete", "optimize", "load_ramp", "spike", "soak", "extreme_input", "adversarial", "rate_limit_probe", "timeout_boundary", "token_limit", "context_overflow", "malformed_request"]>;
+    reason: z.ZodString;
+    granted_by: z.ZodString;
+    granted_at: z.ZodString;
+    expires_at: z.ZodOptional<z.ZodString>;
+    execution_id: z.ZodString;
+    parent_span_id: z.ZodString;
+}, "strip", z.ZodTypeAny, {
+    intent: "compare" | "complete" | "optimize" | "benchmark" | "evaluate" | "load_ramp" | "spike" | "soak" | "extreme_input" | "adversarial" | "rate_limit_probe" | "timeout_boundary" | "token_limit" | "context_overflow" | "malformed_request";
+    reason: string;
+    granted_by: string;
+    granted_at: string;
+    execution_id: string;
+    parent_span_id: string;
+    expires_at?: string | undefined;
+}, {
+    intent: "compare" | "complete" | "optimize" | "benchmark" | "evaluate" | "load_ramp" | "spike" | "soak" | "extreme_input" | "adversarial" | "rate_limit_probe" | "timeout_boundary" | "token_limit" | "context_overflow" | "malformed_request";
+    reason: string;
+    granted_by: string;
+    granted_at: string;
+    execution_id: string;
+    parent_span_id: string;
+    expires_at?: string | undefined;
+}>;
+export type ExemptionRecord = z.infer<typeof ExemptionRecordSchema>;
+export declare const ExemptionRequestSchema: z.ZodObject<{
+    intent: z.ZodEnum<["benchmark", "compare", "evaluate", "complete", "optimize", "load_ramp", "spike", "soak", "extreme_input", "adversarial", "rate_limit_probe", "timeout_boundary", "token_limit", "context_overflow", "malformed_request"]>;
+    reason: z.ZodString;
+    granted_by: z.ZodString;
+}, "strip", z.ZodTypeAny, {
+    intent: "compare" | "complete" | "optimize" | "benchmark" | "evaluate" | "load_ramp" | "spike" | "soak" | "extreme_input" | "adversarial" | "rate_limit_probe" | "timeout_boundary" | "token_limit" | "context_overflow" | "malformed_request";
+    reason: string;
+    granted_by: string;
+}, {
+    intent: "compare" | "complete" | "optimize" | "benchmark" | "evaluate" | "load_ramp" | "spike" | "soak" | "extreme_input" | "adversarial" | "rate_limit_probe" | "timeout_boundary" | "token_limit" | "context_overflow" | "malformed_request";
+    reason: string;
+    granted_by: string;
+}>;
+export type ExemptionRequest = z.infer<typeof ExemptionRequestSchema>;
+//# sourceMappingURL=policy.d.ts.map

package/dist/bundled-agents/test-bench-agents/dist/execution/policy.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"policy.d.ts","sourceRoot":"","sources":["../../src/execution/policy.ts"],"names":[],"mappings":"AASA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAUxB,eAAO,MAAM,sBAAsB,uOAkBjC,CAAC;AAEH,MAAM,MAAM,gBAAgB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,sBAAsB,CAAC,CAAC;AAatE,eAAO,MAAM,qBAAqB,qDAIhC,CAAC;AAEH,MAAM,MAAM,eAAe,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,qBAAqB,CAAC,CAAC;AASpE,eAAO,MAAM,gBAAgB;;;;;;;;;;;;EAI3B,CAAC;AAEH,MAAM,MAAM,UAAU,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,gBAAgB,CAAC,CAAC;AAU1D,eAAO,MAAM,qBAAqB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAYhC,CAAC;AAEH,MAAM,MAAM,eAAe,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,qBAAqB,CAAC,CAAC;AAWpE,eAAO,MAAM,qBAAqB;;;;;;;;;;;;;;;;;;;;;;;;EAQhC,CAAC;AAEH,MAAM,MAAM,eAAe,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,qBAAqB,CAAC,CAAC;AAKpE,eAAO,MAAM,sBAAsB;;;;;;;;;;;;EAIjC,CAAC;AAEH,MAAM,MAAM,gBAAgB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,sBAAsB,CAAC,CAAC"}

package/dist/bundled-agents/test-bench-agents/dist/execution/policy.js

@@ -0,0 +1,53 @@
+import { z } from 'zod';
+export const SimulationIntentSchema = z.enum([
+    'benchmark',
+    'compare',
+    'evaluate',
+    'complete',
+    'optimize',
+    'load_ramp',
+    'spike',
+    'soak',
+    'extreme_input',
+    'adversarial',
+    'rate_limit_probe',
+    'timeout_boundary',
+    'token_limit',
+    'context_overflow',
+    'malformed_request',
+]);
+export const GateDispositionSchema = z.enum([
+    'mandatory',
+    'recommended',
+    'optional',
+]);
+export const PolicyRuleSchema = z.object({
+    intent: SimulationIntentSchema,
+    disposition: GateDispositionSchema,
+    rationale: z.string().min(1),
+});
+export const ExecutionPolicySchema = z.object({
+    policy_version: z.string().regex(/^\d+\.\d+\.\d+$/),
+    updated_at: z.string().datetime(),
+    description: z.string(),
+    rules: z.array(PolicyRuleSchema).refine((rules) => {
+        const intents = rules.map((r) => r.intent);
+        return new Set(intents).size === intents.length;
+    }, { message: 'Duplicate intent in policy rules' }),
+    default_disposition: GateDispositionSchema,
+});
+export const ExemptionRecordSchema = z.object({
+    intent: SimulationIntentSchema,
+    reason: z.string().min(1),
+    granted_by: z.string().min(1),
+    granted_at: z.string().datetime(),
+    expires_at: z.string().datetime().optional(),
+    execution_id: z.string(),
+    parent_span_id: z.string(),
+});
+export const ExemptionRequestSchema = z.object({
+    intent: SimulationIntentSchema,
+    reason: z.string().min(1),
+    granted_by: z.string().min(1),
+});
+//# sourceMappingURL=policy.js.map

package/dist/bundled-agents/test-bench-agents/dist/execution/policy.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"policy.js","sourceRoot":"","sources":["../../src/execution/policy.ts"],"names":[],"mappings":"AASA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAUxB,MAAM,CAAC,MAAM,sBAAsB,GAAG,CAAC,CAAC,IAAI,CAAC;IAE3C,WAAW;IACX,SAAS;IACT,UAAU;IACV,UAAU;IACV,UAAU;IAEV,WAAW;IACX,OAAO;IACP,MAAM;IACN,eAAe;IACf,aAAa;IACb,kBAAkB;IAClB,kBAAkB;IAClB,aAAa;IACb,kBAAkB;IAClB,mBAAmB;CACpB,CAAC,CAAC;AAeH,MAAM,CAAC,MAAM,qBAAqB,GAAG,CAAC,CAAC,IAAI,CAAC;IAC1C,WAAW;IACX,aAAa;IACb,UAAU;CACX,CAAC,CAAC;AAWH,MAAM,CAAC,MAAM,gBAAgB,GAAG,CAAC,CAAC,MAAM,CAAC;IACvC,MAAM,EAAE,sBAAsB;IAC9B,WAAW,EAAE,qBAAqB;IAClC,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;CAC7B,CAAC,CAAC;AAYH,MAAM,CAAC,MAAM,qBAAqB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC5C,cAAc,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,KAAK,CAAC,iBAAiB,CAAC;IACnD,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IACjC,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE;IACvB,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,gBAAgB,CAAC,CAAC,MAAM,CACrC,CAAC,KAAwE,EAAE,EAAE;QAC3E,MAAM,OAAO,GAAG,KAAK,CAAC,GAAG,CAAC,CAAC,CAAqB,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;QAC/D,OAAO,IAAI,GAAG,CAAC,OAAO,CAAC,CAAC,IAAI,KAAK,OAAO,CAAC,MAAM,CAAC;IAClD,CAAC,EACD,EAAE,OAAO,EAAE,kCAAkC,EAAE,CAChD;IACD,mBAAmB,EAAE,qBAAqB;CAC3C,CAAC,CAAC;AAaH,MAAM,CAAC,MAAM,qBAAqB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC5C,MAAM,EAAE,sBAAsB;IAC9B,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACzB,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC7B,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IACjC,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAC5C,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE;IACxB,cAAc,EAAE,CAAC,CAAC,MAAM,EAAE;CAC3B,CAAC,CAAC;AAOH,MAAM,CAAC,MAAM,sBAAsB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC7C,MAAM,EAAE,sBAAsB;IAC9B,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACzB,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;CAC9B,CAAC,CAAC"}

package/dist/bundled-agents/test-bench-agents/dist/execution/span-manager.d.ts

@@ -0,0 +1,28 @@
+import type { ExecutionContext, ExecutionSpan, ExecutionArtifact, ExecutionResult, SpanStatus } from './types.js';
+import type { ExemptionRecord } from './policy.js';
+export declare function validateExecutionContext(ctx: unknown): ctx is ExecutionContext;
+export declare class ExecutionContextError extends Error {
+    constructor(message: string);
+}
+export declare class ExecutionInvariantError extends Error {
+    constructor(message: string);
+}
+export declare class SpanManager {
+    private readonly repoSpan;
+    private readonly agentSpans;
+    private readonly artifacts;
+    private readonly exemptions;
+    private readonly activeAgents;
+    private finalized;
+    private constructor();
+    static create(ctx: unknown): SpanManager;
+    get repoSpanId(): string;
+    get executionId(): string;
+    startAgent(agentName: string): ExecutionSpan;
+    endAgent(spanId: string, status: 'completed' | 'failed', failureReasons?: string[]): void;
+    addArtifact(artifact: ExecutionArtifact): void;
+    addExemption(record: ExemptionRecord): void;
+    finalize<T>(data?: T, forceStatus?: SpanStatus): ExecutionResult<T>;
+    finalizeFailure(reasons: string[]): ExecutionResult<never>;
+}
+//# sourceMappingURL=span-manager.d.ts.map

package/dist/bundled-agents/test-bench-agents/dist/execution/span-manager.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"span-manager.d.ts","sourceRoot":"","sources":["../../src/execution/span-manager.ts"],"names":[],"mappings":"AAcA,OAAO,KAAK,EACV,gBAAgB,EAChB,aAAa,EACb,iBAAiB,EACjB,eAAe,EACf,UAAU,EACX,MAAM,YAAY,CAAC;AAEpB,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,aAAa,CAAC;AAMnD,wBAAgB,wBAAwB,CAAC,GAAG,EAAE,OAAO,GAAG,GAAG,IAAI,gBAAgB,CAS9E;AAKD,qBAAa,qBAAsB,SAAQ,KAAK;gBAClC,OAAO,EAAE,MAAM;CAI5B;AAKD,qBAAa,uBAAwB,SAAQ,KAAK;gBACpC,OAAO,EAAE,MAAM;CAI5B;AAaD,qBAAa,WAAW;IACtB,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAgB;IACzC,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAyC;IACpE,OAAO,CAAC,QAAQ,CAAC,SAAS,CAA2B;IACrD,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAyB;IACpD,OAAO,CAAC,QAAQ,CAAC,YAAY,CAA0B;IACvD,OAAO,CAAC,SAAS,CAAS;IAE1B,OAAO;IAeP,MAAM,CAAC,MAAM,CAAC,GAAG,EAAE,OAAO,GAAG,WAAW;IAYxC,IAAI,UAAU,IAAI,MAAM,CAEvB;IAGD,IAAI,WAAW,IAAI,MAAM,CAGxB;IAMD,UAAU,CAAC,SAAS,EAAE,MAAM,GAAG,aAAa;IAuB5C,QAAQ,CACN,MAAM,EAAE,MAAM,EACd,MAAM,EAAE,WAAW,GAAG,QAAQ,EAC9B,cAAc,CAAC,EAAE,MAAM,EAAE,GACxB,IAAI;IAuBP,WAAW,CAAC,QAAQ,EAAE,iBAAiB,GAAG,IAAI;IAc9C,YAAY,CAAC,MAAM,EAAE,eAAe,GAAG,IAAI;IAY3C,QAAQ,CAAC,CAAC,EAAE,IAAI,CAAC,EAAE,CAAC,EAAE,WAAW,CAAC,EAAE,UAAU,GAAG,eAAe,CAAC,CAAC,CAAC;IA0EnE,eAAe,CAAC,OAAO,EAAE,MAAM,EAAE,GAAG,eAAe,CAAC,KAAK,CAAC;CAsB3D"}