@workbench-ai/workbench-core 0.0.46 → 0.0.48

package/dist/index.js

@@ -1,28 +1,30 @@
-import { createHash } from "node:crypto";
+import { createHash, randomBytes } from "node:crypto";
 import os from "node:os";
 import path from "node:path";
+import { fileURLToPath } from "node:url";
 import YAML from "yaml";
-import { adapterCommandName, assertWorkbenchAdapterOperationResultOk, collectWorkbenchAdapterAuthRequirements, readWorkbenchAdapterOperationResult, workbenchAdapterOperationCommand, workbenchAdapterOperationResultPath, } from "@workbench-ai/workbench-protocol";
-import { BENCHMARK_SPEC_FILE, engineCaseEnginePrivateFiles, engineCaseFilesForRuntimeInput, engineCaseSubjectVisibleFiles, resolveEngineCaseExecutionConfig, resolveWorkbenchResolvedSourceYaml as resolveWorkbenchResolvedSourceYamlInternal, validateWorkbenchResolvedSourceYaml as validateWorkbenchResolvedSourceYamlInternal, isWorkbenchSubjectManifestPath, } from "./generic-spec.js";
+import { adapterCommandName, assertWorkbenchAdapterOperationResultOk, collectWorkbenchAdapterAuthRequirements, parseWorkbenchAdapterManifest, readWorkbenchAdapterOperationResult, WORKBENCH_RUNTIME_CONTROL_TOKEN_ENV, WORKBENCH_RUNTIME_CONTROL_URL_ENV, workbenchAdapterOperationCommand, workbenchAdapterOperationExecutor, workbenchAdapterOperationResultPath, } from "@workbench-ai/workbench-protocol";
+import { BENCHMARK_SPEC_FILE, engineCasePrivateFiles, engineCaseFilesForRuntimeInput, engineCasePublicFiles, resolveEngineCaseExecutionConfig, resolveWorkbenchResolvedSourceYaml as resolveWorkbenchResolvedSourceYamlInternal, validateWorkbenchResolvedSourceYaml as validateWorkbenchResolvedSourceYamlInternal, isWorkbenchSubjectManifestPath, } from "./generic-spec.js";
 import { attachSandboxMetadataToJob, createWorkbenchSandboxFileStore, isSurfaceSnapshotFile, readWorkbenchExecutionSpec, } from "./sandbox-inputs.js";
-import { asRuntimeRecord, importNodeModule, jsonRecord, nodeBuiltin, quoteShellArg, resolveWorkbenchWorkerId, } from "./runtime-utils.js";
-import { executeValidatedSandboxExecution, } from "./sandbox-plane.js";
+import { asRuntimeRecord, importNodeModule, isJsonPayload, jsonRecord, nodeBuiltin, quoteShellArg, resolveWorkbenchWorkerId, } from "./runtime-utils.js";
+import { createWorkbenchExecutionCapability, createWorkbenchSandboxAllocation, collectExecutionCapabilityScopeIssues, collectSandboxAllocationScopeIssues, collectSandboxHandleScopeIssues, assertSandboxBackendSupportsNetworkPolicy, executeValidatedSandboxExecution, } from "./sandbox-plane.js";
 import { createSandboxBackendPlaneForProvider, } from "./sandbox-backends/index.js";
 import { applyWorkbenchSubjectPatch } from "./subject-patch.js";
 import { assignUsageRole, completeUsageSummary, mergeUsageSummaries, normalizeUsageSummary, usageStats, } from "./execution-usage.js";
-import { traceFilePaths, workbenchTracePhaseDirectory, } from "./trace-files.js";
+import { traceFilePaths, workbenchTraceExecutionDirectory, } from "./trace-files.js";
 import { engineCaseForCase, } from "./execution-jobs.js";
-import { createWorkbenchExecutionEventPublisher, publishCommandPhaseEvent, } from "./execution-events.js";
-import { readWorkbenchExecutionPurpose } from "./execution-phases.js";
+import { createWorkbenchExecutionEventPublisher, publishCommandStepEvent, } from "./execution-events.js";
+import { readWorkbenchExecutionPurpose } from "./execution-evidence.js";
 import { adapterAuthEnv, localWorkbenchAdapterAuthStore, normalizeWorkbenchAdapterAuthTarget, sanitizeWorkbenchAdapterAuthBundle, } from "./adapter-auth.js";
-export { BENCHMARK_SPEC_FILE, DEFAULT_EXECUTION_RESOURCES, engineCaseEnginePrivateFiles, engineCaseFilesForRuntimeInput, engineCaseSubjectVisibleFiles, engineResolveInvocationForSpec, engineResolveBindingForSpec, engineResolveBindingForSourceYaml, isWorkbenchSubjectManifestPath, parseWorkbenchSourceFiles, resolveEngineCaseExecutionConfig, resolveWorkbenchResolvedSourceYaml, resolveWorkbenchSourceFiles, serializeWorkbenchResolvedSourceYaml, validateWorkbenchResolvedSourceYaml, } from "./generic-spec.js";
-export { adapterCommandName, cloneWorkbenchAdapterManifest, collectWorkbenchAdapterAuthRequirements, collectWorkbenchAdapterInvocations, parseWorkbenchAdapterManifest, workbenchAdapterManifestRequiresAuth, workbenchAdapterManifestSupportsOperation, workbenchAdapterOperationCommand, withDefaultWorkbenchAdapterAuth, withDefaultWorkbenchAdapterAuthProfiles, } from "@workbench-ai/workbench-protocol";
+export { BENCHMARK_SPEC_FILE, DEFAULT_EXECUTION_RESOURCES, engineCasePrivateFiles, engineCaseFilesForRuntimeInput, engineCasePublicFiles, engineResolveInvocationForSpec, engineResolveBindingForSpec, engineResolveBindingForSourceYaml, isWorkbenchSubjectManifestPath, parseWorkbenchSourceFiles, resolveEngineCaseExecutionConfig, resolveWorkbenchResolvedSourceYaml, resolveWorkbenchSourceFiles, runtimeNetwork, runtimeResources, serializeWorkbenchResolvedSourceYaml, validateWorkbenchResolvedSourceYaml, } from "./generic-spec.js";
+export { composeRuntimeDockerfileWithAdapterInstallers, } from "./runtime-dockerfile.js";
+export { adapterCommandName, cloneWorkbenchAdapterManifest, collectWorkbenchAdapterAuthRequirements, collectWorkbenchAdapterInvocations, parseWorkbenchAdapterManifest, workbenchAdapterManifestRequiresAuth, workbenchAdapterManifestSupportsOperation, workbenchAdapterOperationCommand, workbenchAdapterOperationExecutor, withDefaultWorkbenchAdapterAuth, withDefaultWorkbenchAdapterAuthProfiles, } from "@workbench-ai/workbench-protocol";
 export { adapterAuthEnv, createWorkbenchAdapterAuthBundle, defaultWorkbenchAdapterAuthStoreRoot, localWorkbenchAdapterAuthStore, normalizeWorkbenchAdapterAuthTarget, parseWorkbenchAdapterAuthTarget, sanitizeWorkbenchAdapterAuthBundle, } from "./adapter-auth.js";
 export { asRuntimeRecord, importNodeModule, nodeBuiltin, normalizeWorkbenchWorkerId, normalizeRuntimeRegistry, quoteShellArg, resolveDockerRuntimeImageRef, resolveWorkbenchWorkerId, } from "./runtime-utils.js";
 export { assignUsageRole, extractExecutionUsageFromTrace, mergeUsageSummaries, } from "./execution-usage.js";
 export { createWorkbenchProgressStdoutParser, publishWorkbenchProgressStdoutEnvelope, } from "./execution-events.js";
 export { resolveSandboxTemplateImage, } from "./sandbox-backends/template-images.js";
-export { readOutputTraceFiles, workbenchTracePhaseDirectory, workbenchTraceRunDirectory, workbenchTraceRunDirectoryName, } from "./trace-files.js";
+export { readOutputTraceFiles, workbenchTraceExecutionDirectory, workbenchTraceRunDirectory, workbenchTraceRunDirectoryName, } from "./trace-files.js";
 export { assertWorkbenchAdapterOperationSupport, assertWorkbenchAdapterOperationResultOk, collectWorkbenchAdapterOperationIssues, collectWorkbenchAdapterOperationRequirements, ensureWorkbenchAdapterOutputDir, WORKBENCH_ADAPTER_RESULT_FILE, normalizeWorkbenchAdapterOperationRequest, normalizeWorkbenchAdapterOperationResult, readWorkbenchAdapterOperationRequest, readWorkbenchAdapterOperationResult, workbenchAdapterOperationResultPath, writeWorkbenchAdapterOperationResult, } from "@workbench-ai/workbench-protocol";
 export { applyWorkbenchSubjectPatch, } from "./subject-patch.js";
 export { createWorkbenchSandboxFileStore, createSandboxAdapterRequest, executionResultFromCompletedSandboxJob, materializeWorkbenchSandboxInput, readWorkbenchExecutionSpec, sanitizeWorkbenchExecutionJobForSandbox, } from "./sandbox-inputs.js";
@@ -31,8 +33,8 @@ export { createBaselineSubjectExecution, createBaselineSubjectJob, createWorkben
 export { addCapacity, capacityFits, runWorkbenchExecutionDag, subtractCapacity, workbenchJobDependencies, workbenchJobHostCost, workbenchJobResources, } from "./execution-scheduler.js";
 export { assertWorkbenchExecutionIsolation, collectWorkbenchExecutionIsolationIssues, validateWorkbenchExecutionOutputPayloads, } from "./execution-outputs.js";
 export { collectSandboxAllocationScopeIssues, collectExecutionCapabilityScopeIssues, collectSandboxHandleScopeIssues, createWorkbenchSandboxAllocation, createWorkbenchSandboxExecutionMetadata, createWorkbenchExecutionCapability, executeValidatedSandboxExecution, } from "./sandbox-plane.js";
-export { buildSubjectCasePhaseRefs, buildWorkbenchTracePhases, isWorkbenchPhaseActive, readWorkbenchExecutionId, readWorkbenchExecutionMetadataNumber, readWorkbenchExecutionMetadataString, readWorkbenchExecutionPurpose, resolveWorkbenchJobGroupStatus, } from "./execution-phases.js";
-export { finalizeWorkbenchExecutionTraceForJob, mergeWorkbenchExecutionTracesByJob, } from "./execution-traces.js";
+export { buildSubjectCaseExecutionRefs, buildWorkbenchExecutionEvidence, isWorkbenchExecutionActive, readWorkbenchExecutionId, readWorkbenchExecutionMetadataNumber, readWorkbenchExecutionMetadataString, readWorkbenchExecutionPurpose, resolveWorkbenchJobGroupStatus, } from "./execution-evidence.js";
+export { buildWorkbenchTraceSessionsFromFiles, combineWorkbenchTraceSessions, finalizeWorkbenchExecutionTraceForJob, mergeWorkbenchExecutionTracesByJob, readWorkbenchExecutionTraceFiles, traceSessionLabel, } from "./execution-traces.js";
 export { DOCKER_SANDBOX_BACKEND, assertSandboxHostHealthForProvider, createDockerSandboxBackendDescriptor, createDockerSandboxPlane, resolveWorkbenchSandboxProviderName, sandboxProviderAdmissionForResources, sandboxProviderDefaultMaxConcurrentJobs, sandboxProviderLeaseScope, sandboxHostHealthExpectationForProvider, } from "./sandbox-backends/index.js";
 export const DEFAULT_ENVIRONMENT_VERSIONS = [
     {
@@ -142,7 +144,7 @@ export const DEFAULT_ENVIRONMENTS = [
     {
         id: "env_libreoffice_agent",
         name: "LibreOffice + Agent",
-        description: "Agent runtime with soffice and Python libraries for spreadsheet-heavy skill and rubric evaluations.",
+        description: "Agent runtime with soffice and Python libraries for spreadsheet-heavy evaluations.",
         currentVersionId: "envv_libreoffice_agent",
         builtIn: true,
         createdAt: "2026-04-29T00:00:00.000Z",
@@ -278,30 +280,36 @@ function adapterProtocolCommandSpec(adapter, operation, manifests = []) {
     return {
         use: "command",
         command: manifest ? workbenchAdapterOperationCommand(manifest, operation) : adapterCommandName(adapter.use),
+        executor: manifest ? workbenchAdapterOperationExecutor(manifest, operation) : "sandbox",
     };
 }
-function protocolPhaseForExecution(execution, manifests) {
-    const role = executionPurposeRole(execution.purpose);
-    const operation = execution.purpose === "improve" ? "optimizer.improve" : "subject.run";
+function protocolStepForExecution(execution, manifests) {
+    if (execution.purpose !== "improve") {
+        throw new Error(`Protocol execution step only supports improve executions, not ${execution.purpose}.`);
+    }
+    const operation = "optimizer.improve";
     const command = adapterProtocolCommandSpec(execution.adapter, operation, manifests);
     return {
-        kind: role,
+        kind: "optimizer",
         label: execution.purpose,
         operation,
+        executor: command.executor,
         adapter: execution.adapter,
         command: command.command,
     };
 }
-function attemptPhasesForExecution(execution, spec, manifests) {
+function attemptStepsForExecution(execution, spec, manifests) {
     void spec;
-    const enginePhase = {
+    const command = adapterProtocolCommandSpec(execution.adapter, "engine.run", manifests);
+    const engineStep = {
         kind: "engine",
         label: "engine",
         operation: "engine.run",
+        executor: command.executor,
         adapter: execution.adapter,
-        command: adapterProtocolCommandSpec(execution.adapter, "engine.run", manifests).command,
+        command: command.command,
     };
-    return [enginePhase];
+    return [engineStep];
 }
 function adapterConfigRecord(adapter, manifests = []) {
     const config = cloneJsonRecord(jsonRecord(adapter.with));
@@ -411,7 +419,10 @@ export function materializeWorkbenchRunResult(args) {
             .sort((left, right) => compareSampleOutputs(left.output, right.output));
         const outputJobIds = new Set(outputs.flatMap(({ jobs }) => jobs.map((job) => job.id)));
         const completedSampleKeys = new Set(outputs
-            .map(({ output }) => evaluationSampleGroupKeyFromOutput(output))
+            .flatMap(({ jobs, output }) => [
+            evaluationSampleGroupKeyFromOutput(output),
+            ...jobs.map(evaluationSampleGroupKeyFromJob),
+        ])
             .filter((key) => key !== null));
         const errorSampleJobs = [
             ...subjectJobs.filter((job) => job.status === "failed"),
@@ -422,12 +433,13 @@ export function materializeWorkbenchRunResult(args) {
             ...outputs.map(({ jobs, output }) => withJobUsage(output.sample, completed, jobs[0])),
             ...errorSamples,
         ].sort((left, right) => left.index - right.index || left.id.localeCompare(right.id));
-        const evalRecord = createEvaluationRecord(subjectId, samples);
+        const subjectName = normalizedSubjectDisplayName(args.spec.subject.name);
+        const evalRecord = createEvaluationRecord(subjectId, subjectName, samples);
         const usage = mergeUsageSummaries([
             subjectRevision.usage,
             ...samples.map((sample) => sample.usage),
         ]);
-        const metrics = evaluationMeanMetrics(createEvaluationRecord(subjectId, samples));
+        const metrics = evaluationMeanMetrics(evalRecord);
         const attemptIndex = subjectRevision.attemptIndex;
         const evaluationTraces = [
             ...outputs.flatMap(({ output }) => output.traces),
@@ -457,6 +469,7 @@ export function materializeWorkbenchRunResult(args) {
         }
         const record = {
             id: subjectId,
+            ...(subjectName ? { name: subjectName } : {}),
             ordinal: args.existingSubjectCount + subjects.length,
             benchmarkFingerprint: args.benchmarkFingerprint,
             subjectFingerprint: args.subjectFingerprint ?? materializedSubjectFingerprint(args.spec, subjectRevision.files),
@@ -472,7 +485,7 @@ export function materializeWorkbenchRunResult(args) {
             meta,
         };
         subjects.push(record);
-        evaluations.push(createEvaluationResultRecord({
+        evaluations.push(createEvaluationScorecard({
             runId: args.runId,
             benchmarkFingerprint: args.benchmarkFingerprint,
             createdAt: args.startedAt,
@@ -528,6 +541,8 @@ function materializedSubjectFingerprint(spec, files) {
     hash.update("workbench-subject-v1\0");
     hash.update("materialized\0runner\0");
     hash.update(JSON.stringify(spec.run));
+    hash.update("prepare");
+    hash.update(JSON.stringify(spec.subject.prepare ?? null));
     for (const file of filterSubjectSourceFiles(files).slice().sort((left, right) => left.path.localeCompare(right.path))) {
         hash.update("\0file\0");
         hash.update(file.path);
@@ -547,14 +562,15 @@ function materializedSubjectFiles(args) {
     }
     return [...byPath.values()].sort((left, right) => left.path.localeCompare(right.path));
 }
-function createEvaluationResultRecord(args) {
+function createEvaluationScorecard(args) {
     const evaluation = args.evaluation;
     return {
-        id: evaluationResultId(args.runId, args.subject.id),
+        id: evaluationScorecardId(args.runId, args.subject.id),
         runId: args.runId,
         benchmarkFingerprint: args.benchmarkFingerprint,
         subjectFingerprint: args.subject.subjectFingerprint,
         subjectId: args.subject.id,
+        ...(args.subject.name ? { subjectName: args.subject.name } : {}),
         createdAt: args.createdAt,
         updatedAt: evaluation.finishedAt ?? args.createdAt,
         status: evaluation.status,
@@ -568,7 +584,7 @@ function createEvaluationResultRecord(args) {
         evaluation,
     };
 }
-function evaluationResultId(runId, subjectId) {
+export function evaluationScorecardId(runId, subjectId) {
     const runPart = runId.replace(/[^a-z0-9]+/giu, "_").replace(/^_+|_+$/gu, "").slice(-24);
     const subjectPart = subjectId.replace(/[^a-z0-9]+/giu, "_").replace(/^_+|_+$/gu, "").slice(-24);
     return `eval_${runPart}_${subjectPart}`;
@@ -584,7 +600,7 @@ export function isWorkbenchInternalOutputPath(filePath) {
         normalized === "sandbox-environment.json" ||
         normalized === "sandbox_error.log" ||
         normalized === "exit_code" ||
-        /^[a-z-]+_(stdout\.log|stderr\.log|exit_code)$/u.test(normalized));
+        /^[a-z_-]+_(stdout\.log|stderr\.log|exit_code)$/u.test(normalized));
 }
 export function createSubjectRevisionTraceInputFiles(args) {
     const files = [];
@@ -620,6 +636,23 @@ export function createSubjectRevisionTraceInputFiles(args) {
     }, null, 2)}\n`));
     return dedupeSurfaceFiles(files);
 }
+export function createSubjectEvaluationTraceInputFiles(args) {
+    const subject = args.subject;
+    if (!subject?.eval && !subject?.metrics) {
+        return [];
+    }
+    const filePath = normalizeRelativePath(args.path ?? `base-subject/${subject.id}/evaluation.json`);
+    const payload = {
+        kind: "subject_evaluation",
+        subjectId: subject.id,
+        status: subject.status,
+        metrics: subject.metrics ?? null,
+        fileChanges: subject.fileChanges,
+        eval: subject.eval ?? null,
+        prompt: subject.prompt ?? null,
+    };
+    return [textSurfaceFile(filePath, `${JSON.stringify(payload, null, 2)}\n`)];
+}
 function isTerminalExecutionJob(job) {
     return job.kind === "execute" && (job.status === "succeeded" ||
         job.status === "failed" ||
@@ -866,16 +899,14 @@ export function createSubjectFilePreview(args) {
     };
 }
 export function createCaseReview(args) {
-    const preferredSampleIndex = uniquePhaseSampleIndex(args.phases ?? []);
-    const sampleMatchesCase = (sample) => sample.id === args.caseId ||
-        sample.id.startsWith(`${args.caseId}__`) ||
-        (sample.cases ?? []).some((entry) => entry.id === args.caseId || entry.id.startsWith(`${args.caseId}__`));
+    const preferredSampleIndex = uniqueExecutionSampleIndex(args.executions ?? []);
+    const sampleMatchesCase = (sample) => (sample.cases ?? []).some((entry) => entry.id === args.caseId);
     const samples = args.subject.eval?.samples ?? [];
     const sampleResult = samples.find((sample) => typeof preferredSampleIndex === "number" &&
         sample.index === preferredSampleIndex &&
         sampleMatchesCase(sample)) ?? samples.find(sampleMatchesCase);
-    const caseResult = sampleResult?.cases?.find((entry) => entry.id === args.caseId || entry.id.startsWith(`${args.caseId}__`));
-    if (!sampleResult && (args.phases?.length ?? 0) > 0) {
+    const caseResult = sampleResult?.cases?.find((entry) => entry.id === args.caseId);
+    if (!sampleResult && (args.executions?.length ?? 0) > 0) {
         return {
             subjectId: args.subject.id,
             caseId: args.caseId,
@@ -884,7 +915,7 @@ export function createCaseReview(args) {
                 ? { sampleIndex: preferredSampleIndex }
                 : {}),
             metrics: {},
-            phases: args.phases ?? [],
+            executions: args.executions ?? [],
             criteria_results: [],
         };
     }
@@ -893,28 +924,21 @@ export function createCaseReview(args) {
     }
     const durationMs = typeof caseResult?.durationMs === "number"
         ? caseResult.durationMs
-        : sampleResult?.cases?.length === 1 &&
-            typeof sampleResult.durationMs === "number"
-            ? sampleResult.durationMs
-            : !caseResult && typeof sampleResult.durationMs === "number"
-                ? sampleResult.durationMs
-                : undefined;
-    const sampleStatus = sampleResult.status === "planned" ? undefined : sampleResult.status;
-    const status = caseResult?.status ?? sampleStatus;
+        : undefined;
     return {
         subjectId: args.subject.id,
-        caseId: caseResult?.id ?? sampleResult.id,
+        caseId: caseResult?.id ?? args.caseId,
         caseLabel: caseResult?.label ?? args.caseId,
         sampleId: sampleResult.id,
         sampleIndex: sampleResult.index,
-        ...(status ? { status } : {}),
-        metrics: caseResult?.metrics ?? sampleResult.metrics ?? {},
+        ...(caseResult?.status ? { status: caseResult.status } : {}),
+        metrics: caseResult?.metrics ?? {},
         ...(typeof durationMs === "number" ? { durationMs } : {}),
         ...(caseResult?.source ? { source: caseResult.source } : {}),
-        ...((caseResult?.feedback ?? sampleResult.feedback) !== undefined
-            ? { feedback: caseResult?.feedback ?? sampleResult.feedback }
+        ...(caseResult?.feedback !== undefined
+            ? { feedback: caseResult.feedback }
             : {}),
-        phases: args.phases ?? [],
+        executions: args.executions ?? [],
         criteria_results: (caseResult?.criteria ?? []).map((criterion) => ({
             criterion_id: criterion.criterion_id,
             pass: criterion.pass,
@@ -924,9 +948,9 @@ export function createCaseReview(args) {
         })),
     };
 }
-function uniquePhaseSampleIndex(phases) {
-    const sampleIndices = new Set(phases
-        .map((phase) => phase.sampleIndex)
+function uniqueExecutionSampleIndex(executions) {
+    const sampleIndices = new Set(executions
+        .map((execution) => execution.sampleIndex)
         .filter((index) => typeof index === "number"));
     if (sampleIndices.size !== 1) {
         return null;
@@ -951,6 +975,7 @@ function parseAuthoredWorkbenchSourceSpec(source) {
             name: resolved.subject.name,
             description: resolved.subject.description,
             files: { path: resolved.subject.files.path },
+            ...(resolved.subject.prepare ? { prepare: { ...resolved.subject.prepare } } : {}),
             run: runSpecFromInvocation(resolved.run),
         },
         ...(resolved.optimizer
@@ -1101,11 +1126,18 @@ export async function executeWorkbenchExecutionJob(args, options) {
         const runtimeArgs = adapterAuthProfiles.length > 0
             ? { ...args, adapterAuthProfiles }
             : args;
-        const executionForSandbox = readWorkbenchExecutionSpec(runtimeArgs.job);
+        const executionForRuntime = readWorkbenchExecutionSpec(runtimeArgs.job);
+        const executor = workbenchExecutionExecutorForRuntimeInput(runtimeArgs);
+        if (executor === "host") {
+            return await withWorkbenchRuntimeControlServer(runtimeArgs, options, startedAt, async (adapterRuntimeEnv) => executeAdapterInCurrentRuntime({
+                ...runtimeArgs,
+                adapterRuntimeEnv,
+            }, executionForRuntime, startedAt, createWorkbenchExecutionCapability(executionForRuntime, { now: startedAt })));
+        }
         const fileStore = createWorkbenchSandboxFileStore(runtimeArgs);
         const planeFactory = options.createSandboxPlaneForProvider ?? createSandboxBackendPlaneForProvider;
         const plane = planeFactory(options.sandboxProvider, runtimeArgs, startedAt, fileStore);
-        const validated = await executeValidatedSandboxExecution(plane, executionForSandbox, {
+        const validated = await executeValidatedSandboxExecution(plane, executionForRuntime, {
             now: startedAt,
             runnerId: resolveWorkbenchWorkerId([
                 process.env.WORKBENCH_WORKER_ID,
@@ -1121,6 +1153,215 @@ export async function executeWorkbenchExecutionJob(args, options) {
         return failWorkbenchRunJob(args.job, startedAt, error);
     }
 }
+export function workbenchExecutionExecutorForRuntimeInput(args) {
+    if (args.runtimeControlOperation) {
+        return "sandbox";
+    }
+    const execution = readWorkbenchExecutionSpec(args.job);
+    const operation = adapterOperationForExecutionPurpose(execution.purpose);
+    if (!operation) {
+        return "sandbox";
+    }
+    const manifest = args.adapterManifests?.find((entry) => entry.id === execution.adapter.use);
+    return manifest ? workbenchAdapterOperationExecutor(manifest, operation) : "sandbox";
+}
+function adapterOperationForExecutionPurpose(purpose) {
+    if (purpose === "improve") {
+        return "optimizer.improve";
+    }
+    if (purpose === "attempt") {
+        return "engine.run";
+    }
+    return null;
+}
+const RUNTIME_CONTROL_MAX_BODY_BYTES = 512 * 1024 * 1024;
+async function withWorkbenchRuntimeControlServer(args, options, startedAt, run) {
+    const [{ createServer }] = await Promise.all([
+        importNodeModule(nodeBuiltin("http")),
+    ]);
+    const token = randomBytes(24).toString("base64url");
+    const server = createServer((request, response) => {
+        void handleWorkbenchRuntimeControlHttpRequest({
+            request,
+            response,
+            token,
+            args,
+            options,
+            startedAt,
+        });
+    });
+    const url = await new Promise((resolve, reject) => {
+        server.once("error", reject);
+        server.listen(0, "127.0.0.1", () => {
+            server.off("error", reject);
+            const address = server.address();
+            if (!address || typeof address === "string") {
+                reject(new Error("Workbench runtime-control server did not expose a local TCP address."));
+                return;
+            }
+            resolve(`http://127.0.0.1:${address.port}`);
+        });
+    });
+    try {
+        return await run({
+            [WORKBENCH_RUNTIME_CONTROL_URL_ENV]: url,
+            [WORKBENCH_RUNTIME_CONTROL_TOKEN_ENV]: token,
+        });
+    }
+    finally {
+        await new Promise((resolve) => server.close(() => resolve()));
+    }
+}
+async function handleWorkbenchRuntimeControlHttpRequest(args) {
+    const { request, response } = args;
+    try {
+        if (request.method !== "POST" || request.url !== "/v1/operation-sequence") {
+            writeRuntimeControlJson(response, 404, { error: "Unknown Workbench runtime-control endpoint." });
+            return;
+        }
+        if (request.headers.authorization !== `Bearer ${args.token}`) {
+            writeRuntimeControlJson(response, 401, { error: "Workbench runtime-control token is invalid." });
+            return;
+        }
+        const parsed = JSON.parse(await readRuntimeControlBody(request));
+        const controlRequest = normalizeRuntimeControlOperationSequenceRequest(parsed);
+        const result = await executeRuntimeControlOperationSequenceInSandbox(args.args, args.options, args.startedAt, controlRequest);
+        writeRuntimeControlJson(response, 200, result);
+    }
+    catch (error) {
+        writeRuntimeControlJson(response, 500, {
+            error: error instanceof Error ? error.stack ?? error.message : String(error),
+        });
+    }
+}
+function writeRuntimeControlJson(response, statusCode, payload) {
+    response.statusCode = statusCode;
+    response.setHeader("content-type", "application/json");
+    response.end(`${JSON.stringify(payload, null, 2)}\n`);
+}
+function readRuntimeControlBody(request) {
+    return new Promise((resolve, reject) => {
+        const chunks = [];
+        let size = 0;
+        request.on("data", (chunk) => {
+            size += chunk.length;
+            if (size > RUNTIME_CONTROL_MAX_BODY_BYTES) {
+                reject(new Error("Workbench runtime-control request body is too large."));
+                request.destroy();
+                return;
+            }
+            chunks.push(chunk);
+        });
+        request.on("error", reject);
+        request.on("end", () => resolve(Buffer.concat(chunks).toString("utf8")));
+    });
+}
+function normalizeRuntimeControlOperationSequenceRequest(value) {
+    if (!value || typeof value !== "object" || Array.isArray(value)) {
+        throw new Error("Workbench runtime-control operation sequence request must be an object.");
+    }
+    const record = value;
+    if (!Array.isArray(record.operations) || record.operations.length === 0) {
+        throw new Error("Workbench runtime-control operation sequence requires at least one operation.");
+    }
+    const inputs = normalizeRuntimeControlInputs(record.inputs);
+    return {
+        ...(inputs ? { inputs } : {}),
+        operations: record.operations.map((entry, index) => normalizeRuntimeControlOperation(entry, `operations[${index}]`)),
+        ...(typeof record.prepare === "boolean" ? { prepare: record.prepare } : {}),
+        ...(typeof record.collectWorkspace === "boolean" ? { collectWorkspace: record.collectWorkspace } : {}),
+    };
+}
+function normalizeRuntimeControlInputs(value) {
+    if (value === undefined) {
+        return undefined;
+    }
+    if (!value || typeof value !== "object" || Array.isArray(value)) {
+        throw new Error("Workbench runtime-control inputs must be an object.");
+    }
+    const record = value;
+    const inputs = {};
+    if (hasOwn(record, "subject")) {
+        inputs.subject = normalizeRuntimeControlFiles(record.subject, "inputs.subject");
+    }
+    if (hasOwn(record, "case")) {
+        inputs.case = normalizeRuntimeControlFiles(record.case, "inputs.case");
+    }
+    if (hasOwn(record, "enginePrivate")) {
+        inputs.enginePrivate = normalizeRuntimeControlFiles(record.enginePrivate, "inputs.enginePrivate");
+    }
+    if (hasOwn(record, "traces")) {
+        inputs.traces = normalizeRuntimeControlFiles(record.traces, "inputs.traces");
+    }
+    if (hasOwn(record, "workspace")) {
+        inputs.workspace = normalizeRuntimeControlFiles(record.workspace, "inputs.workspace");
+    }
+    if (hasOwn(record, "output")) {
+        inputs.output = normalizeRuntimeControlFiles(record.output, "inputs.output");
+    }
+    return inputs;
+}
+function normalizeRuntimeControlFiles(value, label) {
+    if (value === undefined) {
+        return [];
+    }
+    if (!Array.isArray(value)) {
+        throw new Error(`Workbench runtime-control ${label} must be an array.`);
+    }
+    return value.map((entry, index) => {
+        if (!isSurfaceSnapshotFile(entry)) {
+            throw new Error(`Workbench runtime-control ${label}[${index}] must be a surface snapshot file.`);
+        }
+        return { ...entry, path: normalizeRelativePath(entry.path) };
+    });
+}
+function hasOwn(value, key) {
+    return Object.prototype.hasOwnProperty.call(value, key);
+}
+function normalizeRuntimeControlOperation(value, label) {
+    if (!value || typeof value !== "object" || Array.isArray(value)) {
+        throw new Error(`Workbench runtime-control ${label} must be an object.`);
+    }
+    const record = value;
+    const operation = record.operation;
+    if (operation !== "engine.resolve" &&
+        operation !== "engine.run" &&
+        operation !== "subject.run" &&
+        operation !== "optimizer.improve") {
+        throw new Error(`Workbench runtime-control ${label}.operation is invalid.`);
+    }
+    const invocation = record.invocation;
+    if (!invocation || typeof invocation !== "object" || Array.isArray(invocation)) {
+        throw new Error(`Workbench runtime-control ${label}.invocation must be an object.`);
+    }
+    const invocationRecord = invocation;
+    if (typeof invocationRecord.use !== "string" || invocationRecord.use.length === 0) {
+        throw new Error(`Workbench runtime-control ${label}.invocation.use is required.`);
+    }
+    const withConfig = invocationRecord.with === undefined
+        ? {}
+        : isJsonPayload(invocationRecord.with)
+            ? invocationRecord.with
+            : null;
+    if (withConfig === null) {
+        throw new Error(`Workbench runtime-control ${label}.invocation.with must be JSON.`);
+    }
+    if (invocationRecord.auth !== undefined && !isJsonPayload(invocationRecord.auth)) {
+        throw new Error(`Workbench runtime-control ${label}.invocation.auth must be JSON.`);
+    }
+    return {
+        operation,
+        invocation: {
+            use: invocationRecord.use,
+            with: withConfig,
+            ...(invocationRecord.auth !== undefined ? { auth: invocationRecord.auth } : {}),
+            ...(typeof invocationRecord.command === "string" && invocationRecord.command.trim()
+                ? { command: invocationRecord.command }
+                : {}),
+        },
+        ...(typeof record.label === "string" && record.label.trim() ? { label: record.label } : {}),
+    };
+}
 async function explicitAdapterAuthProfilesForExecution(execution, args, loadLocalAdapterProfiles) {
     const required = requiredAdapterAuthTargetsForExecution(execution, args);
     if (required.length === 0) {
@@ -1155,7 +1396,7 @@ function adapterAuthTargetKey(target) {
 export function workbenchExecutionPurpose(job) {
     return readWorkbenchExecutionPurpose(job);
 }
-export async function executeAdapterInCurrentSandboxRuntime(args, execution, startedAt, capability) {
+export async function executeAdapterInCurrentRuntime(args, execution, startedAt, capability) {
     const eventPublisher = createWorkbenchExecutionEventPublisher({
         projectId: args.job.projectId,
         runId: args.job.runId,
@@ -1174,10 +1415,10 @@ export async function executeAdapterInCurrentSandboxRuntime(args, execution, sta
     };
     try {
         if (execution.purpose === "improve") {
-            return await executeSubjectRevisionExecutionInSandbox(runtimeInput, execution, startedAt, capability, eventPublisher);
+            return await executeSubjectRevisionExecutionInCurrentRuntime(runtimeInput, execution, startedAt, capability, eventPublisher);
         }
         if (execution.purpose === "attempt") {
-            return await executeAttemptExecutionInSandbox(runtimeInput, execution, startedAt, capability, eventPublisher);
+            return await executeAttemptExecutionInCurrentRuntime(runtimeInput, execution, startedAt, capability, eventPublisher);
         }
         throw new Error(`Unsupported execution purpose ${execution.purpose}.`);
     }
@@ -1274,7 +1515,7 @@ function adapterAuthRequest(bundles, root, currentAdapterId) {
     }
     return entries;
 }
-function adapterAuthRequestForPhase(args, adapterId) {
+function adapterAuthRequestForStep(args, adapterId) {
     const profiles = (args.adapterAuthProfiles ?? [])
         .map((bundle) => sanitizeWorkbenchAdapterAuthBundle(bundle));
     if (profiles.length === 0) {
@@ -1295,12 +1536,19 @@ function adapterAuthProfilesForExecution(execution, args) {
 }
 function requiredAdapterAuthTargetsForExecution(execution, args) {
     const manifests = args.adapterManifests ?? [];
-    return collectWorkbenchAdapterAuthRequirements(adapterInvocationsForExecution(execution, args.spec), manifests)
+    return collectWorkbenchAdapterAuthRequirements(adapterInvocationsForExecution(execution, args), manifests)
         .map((target) => normalizeWorkbenchAdapterAuthTarget(target));
 }
-function adapterInvocationsForExecution(execution, spec) {
+function adapterInvocationsForExecution(execution, args) {
+    if (args.runtimeControlOperation) {
+        return uniqueAdapterInvocations(args.runtimeControlOperation.operations.map((operation) => ({
+            use: operation.invocation.use,
+            with: operation.invocation.with ?? {},
+            ...(operation.invocation.auth !== undefined ? { auth: operation.invocation.auth } : {}),
+        })));
+    }
     if (execution.purpose === "attempt") {
-        return uniqueAdapterInvocations([execution.adapter, spec.run]);
+        return uniqueAdapterInvocations([execution.adapter, args.spec.run]);
     }
     return [execution.adapter];
 }
@@ -1341,7 +1589,7 @@ function completedJobFromSandboxResult(fallbackJob, startedAt, result) {
     }
     return attachSandboxMetadataToJob(failWorkbenchRunJob(fallbackJob, result.startedAt || startedAt, result.error ?? `Sandbox execution ${result.status}.`, result.finishedAt), asRuntimeRecord(result.metadata).sandbox);
 }
-async function executeSubjectRevisionExecutionInSandbox(args, execution, startedAt, capability, eventPublisher) {
+async function executeSubjectRevisionExecutionInCurrentRuntime(args, execution, startedAt, capability, eventPublisher) {
     const { workload, result } = await runHostedProtocolExecutionResult(args, execution, startedAt, capability, eventPublisher);
     if (result.error || (result.exitCode ?? 0) !== 0) {
         return failWorkbenchRunJob(args.job, startedAt, result.error ?? `Adapter ${execution.adapter.use} exited with status ${result.exitCode}.`, result.finishedAt, result);
@@ -1382,7 +1630,7 @@ async function executeSubjectRevisionExecutionInSandbox(args, execution, started
         },
     };
 }
-async function executeAttemptExecutionInSandbox(args, execution, startedAt, capability, eventPublisher) {
+async function executeAttemptExecutionInCurrentRuntime(args, execution, startedAt, capability, eventPublisher) {
     const workload = createWorkbenchRunWorkload({
         job: args.job,
         spec: args.spec,
@@ -1391,7 +1639,7 @@ async function executeAttemptExecutionInSandbox(args, execution, startedAt, capa
         engineCases: args.engineCases,
         traceFiles: args.traceFiles,
     });
-    const workloadResult = await runHostedCommandExecutionPhases(args, workload, attemptPhasesForExecution(execution, args.spec, args.adapterManifests), startedAt, {
+    const workloadResult = await runHostedCommandExecutionSteps(args, workload, attemptStepsForExecution(execution, args.spec, args.adapterManifests), startedAt, {
         capability,
         eventPublisher,
     });
@@ -1405,10 +1653,7 @@ async function executeAttemptExecutionInSandbox(args, execution, startedAt, capa
         return failWorkbenchRunJob(args.job, startedAt, "Attempt engine must return a workbench-result result with a finite numeric score.", workloadResult.finishedAt, workloadResult);
     }
     const finishedAt = workloadResult.finishedAt ?? new Date().toISOString();
-    const usage = mergeUsageSummaries([
-        workloadResult.usage,
-        engineResult.usage,
-    ]);
+    const usage = attemptUsageSummary(workloadResult.usage, engineResult.usage);
     const sample = evaluateSample({
         subjectId: workload.subjectId,
         files: workloadResult.files,
@@ -1453,6 +1698,282 @@ async function executeAttemptExecutionInSandbox(args, execution, startedAt, capa
         },
     };
 }
+export async function executeRuntimeControlOperationSequenceInCurrentRuntime(args, execution, startedAt, capability) {
+    void execution;
+    void capability;
+    if (!args.runtimeControlOperation) {
+        throw new Error("Runtime-control operation sequence is missing from the sandbox request.");
+    }
+    const childExecution = readWorkbenchExecutionSpec(args.job);
+    const workload = createWorkbenchRunWorkload({
+        job: args.job,
+        spec: args.spec,
+        baseFiles: args.baseFiles,
+        engineResolveFiles: args.engineResolveFiles,
+        engineCases: args.engineCases,
+        traceFiles: args.traceFiles,
+    });
+    const runtimeArgs = { ...args };
+    delete runtimeArgs.adapterRuntimeEnv;
+    const adapterAuth = await materializeSandboxAdapterAuth(runtimeArgs, childExecution);
+    let result;
+    try {
+        result = await runHostedCommandExecutionSteps({
+            ...runtimeArgs,
+            ...(adapterAuth.root ? { adapterAuthRoot: adapterAuth.root } : {}),
+            ...(Object.keys(adapterAuth.env).length > 0
+                ? { adapterAuthEnv: adapterAuth.env }
+                : {}),
+        }, workload, args.runtimeControlOperation.operations.map((operation, index) => runtimeControlStepForOperation(operation, index, args.adapterManifests)), startedAt, {
+            runSubjectPrepare: args.runtimeControlOperation.prepare ?? false,
+            workspaceFiles: args.runtimeControlOperation.inputs?.workspace ?? [],
+            outputFiles: args.runtimeControlOperation.inputs?.output ?? [],
+            collectWorkspace: args.runtimeControlOperation.collectWorkspace ?? false,
+        });
+    }
+    finally {
+        if (adapterAuth.cleanup) {
+            await adapterAuth.cleanup().catch(() => undefined);
+        }
+    }
+    const finishedAt = result.finishedAt ?? new Date().toISOString();
+    const failed = Boolean(result.error) || (result.exitCode ?? 0) !== 0;
+    return {
+        ...args.job,
+        status: failed ? "failed" : "succeeded",
+        attempt: Math.max(1, args.job.attempt),
+        startedAt,
+        finishedAt,
+        updatedAt: finishedAt,
+        ...(failed ? { error: result.error ?? `Runtime-control operation sequence exited with status ${result.exitCode}.` } : {}),
+        output: runtimeControlJobOutput(result, !failed),
+    };
+}
+async function executeRuntimeControlOperationSequenceInSandbox(args, options, startedAt, request) {
+    const childArgs = createRuntimeControlSandboxInput(args, request);
+    const execution = readWorkbenchExecutionSpec(childArgs.job);
+    const fileStore = createWorkbenchSandboxFileStore(childArgs);
+    const planeFactory = options.createSandboxPlaneForProvider ?? createSandboxBackendPlaneForProvider;
+    const plane = planeFactory(options.sandboxProvider, childArgs, startedAt, fileStore);
+    assertSandboxBackendSupportsNetworkPolicy(plane.backend, execution);
+    const sandboxOptions = {
+        now: startedAt,
+        runnerId: resolveWorkbenchWorkerId([
+            process.env.WORKBENCH_WORKER_ID,
+            process.env.EC2_INSTANCE_ID,
+            os.hostname(),
+            process.env.HOSTNAME,
+        ], "local-runner"),
+        fileStore,
+    };
+    const inputs = await fileStore.materializeInputs(execution);
+    const environment = plane.prepareEnvironment
+        ? await plane.prepareEnvironment(execution, sandboxOptions)
+        : {
+            backend: plane.backend.name,
+            kind: execution.sandbox.kind,
+            ref: execution.sandbox.ref,
+        };
+    const allocation = createWorkbenchSandboxAllocation(execution, {
+        backend: plane.backend.name,
+        runnerId: sandboxOptions.runnerId,
+        now: startedAt,
+    });
+    const capability = createWorkbenchExecutionCapability(execution, { now: startedAt });
+    assertRuntimeControlScope("Runtime-control sandbox allocation", collectSandboxAllocationScopeIssues(allocation, execution, { now: startedAt }));
+    assertRuntimeControlScope("Runtime-control execution capability", collectExecutionCapabilityScopeIssues(capability, execution, { now: startedAt }));
+    const sandbox = await plane.createSandbox({
+        execution,
+        environment,
+        allocation,
+        capability,
+        inputs,
+    }, sandboxOptions);
+    assertRuntimeControlScope("Runtime-control sandbox handle", collectSandboxHandleScopeIssues(sandbox, allocation, execution));
+    let result;
+    try {
+        result = await plane.exec({
+            execution,
+            environment,
+            sandbox,
+            allocation,
+            capability,
+            inputs,
+        }, sandboxOptions);
+    }
+    finally {
+        await plane.destroySandbox(sandbox, sandboxOptions);
+    }
+    const completedJob = completedJobFromSandboxResult(childArgs.job, startedAt, result);
+    return runtimeControlResultFromCompletedJob(completedJob);
+}
+function createRuntimeControlSandboxInput(args, request) {
+    const parentExecution = readWorkbenchExecutionSpec(args.job);
+    const parentWorkload = createWorkbenchRunWorkload({
+        job: args.job,
+        spec: args.spec,
+        baseFiles: args.baseFiles,
+        engineResolveFiles: args.engineResolveFiles,
+        engineCases: args.engineCases,
+        traceFiles: args.traceFiles,
+    });
+    const nonce = runtimeControlNonce();
+    const childExecutionId = `${parentExecution.id}:runtime:${nonce}`;
+    const childJobId = `${args.job.id}:runtime:${nonce}`;
+    const parentInput = asRuntimeRecord(args.job.input);
+    const publicFiles = runtimeControlInputFiles(request.inputs, "case", parentWorkload.engineCase ? engineCasePublicFiles(parentWorkload.engineCase) : []);
+    const privateFiles = runtimeControlInputFiles(request.inputs, "enginePrivate", parentWorkload.engineCase ? engineCasePrivateFiles(parentWorkload.engineCase) : []);
+    const subjectFiles = runtimeControlInputFiles(request.inputs, "subject", parentWorkload.subjectFiles);
+    const traceFiles = runtimeControlInputFiles(request.inputs, "traces", parentWorkload.traceFiles);
+    const adapter = request.operations[request.operations.length - 1]?.invocation;
+    const childExecution = {
+        ...parentExecution,
+        id: childExecutionId,
+        outputs: [],
+        adapter: adapter
+            ? {
+                use: adapter.use,
+                with: adapter.with ?? {},
+                ...(adapter.auth !== undefined ? { auth: adapter.auth } : {}),
+            }
+            : parentExecution.adapter,
+        metadata: {
+            ...asRuntimeRecord(parentExecution.metadata),
+            runtimeControl: true,
+            caseId: parentWorkload.caseId,
+        },
+    };
+    const engineCase = {
+        id: parentWorkload.caseId,
+        case: parentWorkload.engineCaseSpec ?? {
+            version: 3,
+            prompt: parentWorkload.prompt,
+        },
+        files: {
+            public: publicFiles,
+            private: privateFiles,
+        },
+    };
+    const childJob = {
+        ...args.job,
+        id: childJobId,
+        input: {
+            ...parentInput,
+            execution: childExecution,
+            caseId: parentWorkload.caseId,
+        },
+    };
+    const childArgs = {
+        ...args,
+        job: childJob,
+        baseFiles: subjectFiles,
+        engineResolveFiles: [...publicFiles, ...privateFiles],
+        engineCases: [engineCase],
+        traceFiles,
+        runtimeControlOperation: request,
+    };
+    delete childArgs.adapterRuntimeEnv;
+    delete childArgs.workspaceRoot;
+    return childArgs;
+}
+function runtimeControlInputFiles(inputs, key, fallback) {
+    if (inputs && Object.prototype.hasOwnProperty.call(inputs, key)) {
+        return cloneSurfaceFiles(inputs[key] ?? []);
+    }
+    return cloneSurfaceFiles(fallback);
+}
+function runtimeControlStepForOperation(operation, index, manifests = []) {
+    const command = operation.invocation.command?.trim()
+        || adapterProtocolCommandSpec({
+            use: operation.invocation.use,
+            with: operation.invocation.with ?? {},
+            ...(operation.invocation.auth !== undefined ? { auth: operation.invocation.auth } : {}),
+        }, operation.operation, manifests).command;
+    return {
+        kind: operation.operation === "subject.run"
+            ? "subject"
+            : operation.operation === "optimizer.improve"
+                ? "optimizer"
+                : "engine",
+        label: operation.label ?? `${operation.operation.replace(".", "_")}_${index + 1}`,
+        operation: operation.operation,
+        executor: "sandbox",
+        adapter: {
+            use: operation.invocation.use,
+            with: operation.invocation.with ?? {},
+            ...(operation.invocation.auth !== undefined ? { auth: operation.invocation.auth } : {}),
+        },
+        command,
+    };
+}
+function runtimeControlResultFromCompletedJob(job) {
+    return normalizeRuntimeControlResultOutput(asRuntimeRecord(job.output), job.status === "succeeded", job.error);
+}
+function runtimeControlJobOutput(result, ok) {
+    return normalizeRuntimeControlResultOutput({
+        ok,
+        files: result.files,
+        fileChanges: result.fileChanges,
+        ...(result.operationResults ? { operationResults: result.operationResults } : {}),
+        ...(result.workspaceFiles ? { workspaceFiles: result.workspaceFiles } : {}),
+        ...(result.result ? { result: result.result } : {}),
+        ...(result.usage ? { usage: result.usage } : {}),
+        ...(result.summary !== undefined ? { summary: result.summary } : {}),
+        ...(result.feedback !== undefined ? { feedback: result.feedback } : {}),
+        ...(result.error ? { error: result.error } : {}),
+    }, ok, result.error);
+}
+function normalizeRuntimeControlResultOutput(output, ok, fallbackError) {
+    const files = Array.isArray(output.files)
+        ? output.files.filter(isSurfaceSnapshotFile)
+        : [];
+    const workspaceFiles = Array.isArray(output.workspaceFiles)
+        ? output.workspaceFiles.filter(isSurfaceSnapshotFile)
+        : undefined;
+    const operationResults = Array.isArray(output.operationResults)
+        ? output.operationResults.filter(isWorkbenchAdapterOperationResult)
+        : [];
+    return {
+        ok: ok && output.ok !== false,
+        files,
+        fileChanges: Array.isArray(output.fileChanges)
+            ? output.fileChanges.filter((entry) => typeof entry === "string")
+            : files.map((file) => file.path),
+        operationResults,
+        ...(workspaceFiles ? { workspaceFiles } : {}),
+        ...(output.result && typeof output.result === "object" && !Array.isArray(output.result)
+            ? { result: output.result }
+            : {}),
+        ...(output.usage && typeof output.usage === "object" && !Array.isArray(output.usage)
+            ? { usage: output.usage }
+            : {}),
+        ...(typeof output.summary === "string" ? { summary: output.summary } : {}),
+        ...(output.feedback !== undefined && isJsonPayload(output.feedback) ? { feedback: output.feedback } : {}),
+        ...(typeof output.error === "string" ? { error: output.error } : fallbackError ? { error: fallbackError } : {}),
+    };
+}
+function isWorkbenchAdapterOperationResult(value) {
+    if (!value || typeof value !== "object" || Array.isArray(value)) {
+        return false;
+    }
+    const record = value;
+    return record.protocol === "workbench.adapter-result.v1" &&
+        (record.operation === "engine.resolve" ||
+            record.operation === "engine.run" ||
+            record.operation === "subject.run" ||
+            record.operation === "optimizer.improve");
+}
+function cloneSurfaceFiles(files) {
+    return files.map((file) => ({ ...file, path: normalizeRelativePath(file.path) }));
+}
+function runtimeControlNonce() {
+    return randomBytes(6).toString("hex");
+}
+function assertRuntimeControlScope(label, issues) {
+    if (issues.length > 0) {
+        throw new Error(`${label} failed validation:\n${issues.join("\n")}`);
+    }
+}
 async function runHostedProtocolExecutionResult(args, execution, startedAt, capability, eventPublisher) {
     const workload = createWorkbenchRunWorkload({
         job: args.job,
@@ -1462,13 +1983,13 @@ async function runHostedProtocolExecutionResult(args, execution, startedAt, capa
         engineCases: args.engineCases,
         traceFiles: args.traceFiles,
     });
-    const result = await runHostedCommandExecutionPhases(args, workload, [protocolPhaseForExecution(execution, args.adapterManifests)], startedAt, {
+    const result = await runHostedCommandExecutionSteps(args, workload, [protocolStepForExecution(execution, args.adapterManifests)], startedAt, {
         capability,
         eventPublisher,
     });
     return { workload, result };
 }
-async function runHostedCommandExecutionPhases(args, workload, phases, startedAt, options = {}) {
+async function runHostedCommandExecutionSteps(args, workload, steps, startedAt, options = {}) {
     const [{ execFile }, fs, os, path, { promisify }] = await Promise.all([
         importNodeModule(nodeBuiltin("child_process")),
         importNodeModule(nodeBuiltin("fs/promises")),
@@ -1489,9 +2010,22 @@ async function runHostedCommandExecutionPhases(args, workload, phases, startedAt
     const workspace = await createRuntimeWorkspaceRoot(args, fs, os, path, "workbench-execution-sandbox-");
     try {
         await stageWorkbenchRunWorkload(workspace.root, workload);
+        if (options.workspaceFiles && options.workspaceFiles.length > 0) {
+            await stageInitialWorkspaceFiles(workspace.root, options.workspaceFiles);
+        }
+        if (options.outputFiles && options.outputFiles.length > 0) {
+            await writeSurfaceFiles(outputDir(workspace.root), options.outputFiles);
+        }
+        const execution = readWorkbenchExecutionSpec(workload.job);
+        const hostAdapterIds = new Set(steps.flatMap((step) => step.executor === "host"
+            ? [step.adapter?.use ?? execution.adapter.use]
+            : []));
+        const hostAdapterRoots = hostAdapterIds.size > 0
+            ? await materializeHostAdapterRoots(workspace.root, args.adapterFiles ?? [], hostAdapterIds)
+            : new Map();
         let exitCode = 0;
         let runtimeError;
-        const phaseResults = [];
+        const operationResults = [];
         try {
             if (!environmentVersion) {
                 throw new Error("environment is required for adapter command executions.");
@@ -1503,49 +2037,64 @@ async function runHostedCommandExecutionPhases(args, workload, phases, startedAt
                     network: environmentVersion.spec.network,
                 }, null, 2)}\n`);
             }
-            const phaseTimeoutMs = environmentVersion
+            const stepTimeoutMs = environmentVersion
                 ? environmentVersionTimeoutMs(environmentVersion)
                 : 5 * 60 * 1000;
-            const execution = readWorkbenchExecutionSpec(workload.job);
-            for (const phase of phases) {
-                await resetHostedWorkloadPhaseOutput(workspace.root, phase);
-                if (phase.kind === "engine" && execution.purpose === "attempt") {
-                    await stageAttemptScoringInputs(workspace.root, workload);
+            const shouldRunSubjectPrepare = options.runSubjectPrepare ?? steps.some((step) => step.executor === "sandbox");
+            if (shouldRunSubjectPrepare) {
+                await runSubjectPrepareCommand({
+                    root: workspace.root,
+                    workload,
+                    execution,
+                    execFileAsync,
+                    timeoutMs: stepTimeoutMs,
+                    eventPublisher: options.eventPublisher,
+                });
+            }
+            let enginePrivateStaged = false;
+            for (const step of steps) {
+                if (step.kind === "engine" && !enginePrivateStaged) {
+                    await stageWorkbenchEnginePrivateFiles(workspace.root, workload);
+                    enginePrivateStaged = true;
                 }
-                const adapterRequestPath = await writeWorkbenchAdapterRequest(workspace.root, workload, execution, phase, adapterAuthRequestForPhase(args, phase.adapter?.use ?? execution.adapter.use), args.adapterManifests);
-                const phaseRole = phaseEventRole(phase);
-                await publishCommandPhaseEvent(options.eventPublisher, {
-                    phase: phase.label,
+                await resetHostedWorkloadStepOutput(workspace.root);
+                const adapterRequestPath = await writeWorkbenchAdapterRequest(workspace.root, workload, execution, step, adapterAuthRequestForStep(args, step.adapter?.use ?? execution.adapter.use), args.adapterManifests);
+                const stepRole = stepEventRole(step);
+                await publishCommandStepEvent(options.eventPublisher, {
+                    step: step.label,
                     status: "started",
-                    ...(phaseRole ? { role: phaseRole } : {}),
+                    ...(stepRole ? { role: stepRole } : {}),
                 });
                 try {
-                    if (!phase.command) {
-                        throw new Error(`Adapter phase ${phase.label} is missing a command.`);
+                    if (!step.command) {
+                        throw new Error(`Adapter step ${step.label} is missing a command.`);
                     }
-                    const command = createHostedWorkloadShellCommand(workspace.root, phase.command, phase.label, phase.okExitCodes);
+                    const adapterRoot = step.executor === "host"
+                        ? hostAdapterRoots.get(step.adapter?.use ?? execution.adapter.use)
+                        : undefined;
+                    const command = createHostedWorkloadShellCommand(workspace.root, step.command, step.label, step.okExitCodes);
                     await execFileAsync("sh", ["-c", command], {
-                        cwd: workspace.root,
-                        env: createHostedWorkloadPhaseEnv(workspace.root, adapterRequestPath, args.adapterAuthEnv),
+                        cwd: adapterRoot ?? workspace.root,
+                        env: createHostedWorkloadAdapterEnv(workspace.root, adapterRequestPath, args.adapterAuthEnv, adapterRoot ? { adapterRoot } : undefined, args.adapterRuntimeEnv),
                         maxBuffer: 10 * 1024 * 1024,
-                        timeout: phaseTimeoutMs,
+                        timeout: stepTimeoutMs,
                     });
-                    const operationResult = await readWorkbenchAdapterOperationResult(outputDir(workspace.root), phase.operation);
-                    assertWorkbenchAdapterOperationResultOk(operationResult, `Adapter ${phase.adapter?.use ?? execution.adapter.use} ${phase.operation}`);
-                    phaseResults.push(operationResult);
-                    await publishCommandPhaseEvent(options.eventPublisher, {
-                        phase: phase.label,
+                    const operationResult = await readWorkbenchAdapterOperationResult(outputDir(workspace.root), step.operation);
+                    assertWorkbenchAdapterOperationResultOk(operationResult, `Adapter ${step.adapter?.use ?? execution.adapter.use} ${step.operation}`);
+                    operationResults.push(operationResult);
+                    await publishCommandStepEvent(options.eventPublisher, {
+                        step: step.label,
                         status: "succeeded",
-                        ...(phaseRole ? { role: phaseRole } : {}),
+                        ...(stepRole ? { role: stepRole } : {}),
                     });
                 }
                 catch (error) {
-                    await publishCommandPhaseEvent(options.eventPublisher, {
-                        phase: phase.label,
+                    await publishCommandStepEvent(options.eventPublisher, {
+                        step: step.label,
                         status: "failed",
                         exitCode: readExitCode(error),
                         error: error instanceof Error ? error.message : String(error),
-                        ...(phaseRole ? { role: phaseRole } : {}),
+                        ...(stepRole ? { role: stepRole } : {}),
                     });
                     throw error;
                 }
@@ -1569,16 +2118,56 @@ async function runHostedCommandExecutionPhases(args, workload, phases, startedAt
                 startedAt,
             });
         }
-        return await readWorkbenchRunWorkloadResult(workspace.root, workload, {
+        const result = await readWorkbenchRunWorkloadResult(workspace.root, workload, {
             exitCode,
             startedAt,
-            phaseResults,
+            operationResults,
         });
+        if (options.collectWorkspace) {
+            result.workspaceFiles = await readMutableWorkspaceSnapshotFiles(workspace.root);
+        }
+        return result;
     }
     finally {
         await workspace.cleanup();
     }
 }
+async function runSubjectPrepareCommand(args) {
+    const command = args.workload.spec.subject.prepare?.command;
+    if (!command) {
+        return;
+    }
+    const role = args.execution.purpose === "improve" ? "optimizer" : "runner";
+    await publishCommandStepEvent(args.eventPublisher, {
+        step: "subject_prepare",
+        status: "started",
+        role,
+    });
+    try {
+        const shellCommand = createHostedWorkloadShellCommand(args.root, command, "subject_prepare");
+        await args.execFileAsync("sh", ["-c", shellCommand], {
+            cwd: args.root,
+            env: createHostedWorkloadPrepareEnv(args.root),
+            maxBuffer: 10 * 1024 * 1024,
+            timeout: args.timeoutMs,
+        });
+        await publishCommandStepEvent(args.eventPublisher, {
+            step: "subject_prepare",
+            status: "succeeded",
+            role,
+        });
+    }
+    catch (error) {
+        await publishCommandStepEvent(args.eventPublisher, {
+            step: "subject_prepare",
+            status: "failed",
+            exitCode: readExitCode(error),
+            error: error instanceof Error ? error.message : String(error),
+            role,
+        });
+        throw new Error(`Subject prepare command failed: ${error instanceof Error ? error.message : String(error)}`);
+    }
+}
 async function createRuntimeWorkspaceRoot(args, fs, os, path, prefix) {
     if (args.workspaceRoot) {
         await fs.mkdir(args.workspaceRoot, { recursive: true });
@@ -1614,19 +2203,22 @@ async function createRuntimeWorkspaceRoot(args, fs, os, path, prefix) {
         },
     };
 }
-function phaseEventRole(phase) {
-    if (phase.kind === "optimizer") {
+function stepEventRole(step) {
+    if (step.kind === "optimizer") {
         return "optimizer";
     }
-    if (phase.kind === "runner") {
+    if (step.kind === "subject") {
         return "runner";
     }
-    if (phase.kind === "engine") {
+    if (step.kind === "engine") {
         return "engine";
     }
     return undefined;
 }
 function adapterOperationUsageSummary(result) {
+    if (hasExplicitUsageRole(result.usage)) {
+        return completeUsageSummary(result.usage);
+    }
     if (result.operation === "optimizer.improve") {
         return assignUsageRole("optimizer", result.usage);
     }
@@ -1638,11 +2230,16 @@ function adapterOperationUsageSummary(result) {
     }
     return result.usage;
 }
-function executionPurposeRole(purpose) {
-    if (purpose === "improve") {
-        return "optimizer";
-    }
-    return "runner";
+function attemptUsageSummary(workloadUsage, resultUsage) {
+    const normalizedWorkloadUsage = completeUsageSummary(workloadUsage);
+    const legacyEngineUsage = normalizedWorkloadUsage?.engine
+        ? undefined
+        : assignUsageRole("engine", resultUsage);
+    return mergeUsageSummaries([normalizedWorkloadUsage, legacyEngineUsage]);
+}
+function hasExplicitUsageRole(usage) {
+    const normalized = completeUsageSummary(usage);
+    return Boolean(normalized?.optimizer || normalized?.runner || normalized?.engine);
 }
 function createSubjectPatchFromResult(result, spec) {
     if (result.subjectPatch) {
@@ -1720,47 +2317,103 @@ export async function stageWorkbenchRunWorkload(root, workload) {
         fs
             .rm(runtimePrivateDir(root), { recursive: true, force: true })
             .catch(() => undefined),
-        fs
-            .rm(runtimeLogsDir(root), { recursive: true, force: true })
-            .catch(() => undefined),
     ]);
     await fs.mkdir(inputDir(root), { recursive: true });
     await fs.mkdir(outputDir(root), { recursive: true });
     if (purpose === "attempt") {
-        assertMutableWorkspaceFiles(workload.subjectFiles, "Subject files");
         await fs.mkdir(subjectDir(root), { recursive: true });
         await fs.mkdir(caseDir(root), { recursive: true });
-        await fs.mkdir(runtimeLogsAgentDir(root), { recursive: true });
-        await fs.mkdir(runtimeLogsVerifierDir(root), { recursive: true });
         const engineCase = requireWorkloadEngineCase(workload, "Attempt staging");
         await writeSurfaceFiles(subjectDir(root), workload.subjectFiles);
-        await writeSurfaceFiles(caseDir(root), engineCaseSubjectVisibleFiles(engineCase));
-        await writeSurfaceFiles(root, workload.subjectFiles);
+        await writeSurfaceFiles(caseDir(root), engineCasePublicFiles(engineCase));
         return;
     }
     if (purpose === "improve") {
-        assertMutableWorkspaceFiles(workload.subjectFiles, "Subject files");
         await fs.mkdir(subjectDir(root), { recursive: true });
         await writeSurfaceFiles(subjectDir(root), workload.subjectFiles);
-        await writeSurfaceFiles(root, workload.subjectFiles);
         await fs.mkdir(tracesDir(root), { recursive: true });
         await writeSurfaceFiles(tracesDir(root), workload.traceFiles);
     }
 }
-async function stageAttemptScoringInputs(root, workload) {
+async function stageWorkbenchEnginePrivateFiles(root, workload) {
+    if (readWorkloadExecutionPurpose(workload) !== "attempt") {
+        return;
+    }
     const fs = await importNodeModule(nodeBuiltin("fs/promises"));
-    const engineCase = requireWorkloadEngineCase(workload, "Attempt scoring");
-    await Promise.all([
-        fs
-            .rm(runtimeEnginePrivateDir(root), { recursive: true, force: true })
-            .catch(() => undefined),
-        fs
-            .rm(runtimeLogsVerifierDir(root), { recursive: true, force: true })
-            .catch(() => undefined),
-    ]);
     await fs.mkdir(runtimeEnginePrivateDir(root), { recursive: true });
-    await fs.mkdir(runtimeLogsVerifierDir(root), { recursive: true });
-    await writeSurfaceFiles(runtimeEnginePrivateDir(root), engineCaseEnginePrivateFiles(engineCase));
+    await writeSurfaceFiles(runtimeEnginePrivateDir(root), engineCasePrivateFiles(requireWorkloadEngineCase(workload, "Engine-private staging")));
+}
+async function stageInitialWorkspaceFiles(root, files) {
+    await writeSurfaceFiles(root, files.filter((file) => isMutableWorkspaceSnapshotPath(file.path)));
+}
+async function readMutableWorkspaceSnapshotFiles(root) {
+    return (await readSurfaceFiles(root))
+        .filter((file) => isMutableWorkspaceSnapshotPath(file.path))
+        .sort((left, right) => left.path.localeCompare(right.path));
+}
+function isMutableWorkspaceSnapshotPath(filePath) {
+    const normalized = normalizeRelativePath(filePath);
+    return Boolean(normalized &&
+        !normalized.startsWith("../") &&
+        normalized !== "input" &&
+        !normalized.startsWith("input/") &&
+        normalized !== "private" &&
+        !normalized.startsWith("private/") &&
+        normalized !== "output" &&
+        !normalized.startsWith("output/") &&
+        normalized !== ".workbench" &&
+        !normalized.startsWith(".workbench/"));
+}
+async function materializeHostAdapterRoots(root, adapterFiles, adapterIds) {
+    if (adapterFiles.length === 0 || adapterIds.size === 0) {
+        return new Map();
+    }
+    const fs = await importNodeModule(nodeBuiltin("fs/promises"));
+    const path = await importNodeModule(nodeBuiltin("path"));
+    const sourceRoots = hostAdapterSourceRoots(adapterFiles, adapterIds);
+    const roots = new Map();
+    for (const [adapterId, sourceRoot] of sourceRoots) {
+        const targetRoot = path.join(root, ".workbench", "adapters", adapterId);
+        const files = adapterFiles.flatMap((file) => {
+            const relativePath = adapterFilePathWithinRoot(file.path, sourceRoot);
+            return relativePath === null
+                ? []
+                : [{ ...file, path: relativePath }];
+        });
+        await fs.rm(targetRoot, { recursive: true, force: true }).catch(() => undefined);
+        await fs.mkdir(targetRoot, { recursive: true });
+        await writeSurfaceFiles(targetRoot, files);
+        roots.set(adapterId, await fs.realpath(targetRoot));
+    }
+    return roots;
+}
+function hostAdapterSourceRoots(adapterFiles, adapterIds) {
+    const roots = new Map();
+    for (const file of adapterFiles) {
+        const normalized = normalizeRelativePath(file.path);
+        if (!normalized.endsWith("workbench.adapter.yaml")) {
+            continue;
+        }
+        const manifest = parseWorkbenchAdapterManifest(file.content);
+        if (!adapterIds.has(manifest.id)) {
+            continue;
+        }
+        const sourceRoot = normalized === "workbench.adapter.yaml"
+            ? ""
+            : normalized.slice(0, -"workbench.adapter.yaml".length).replace(/\/+$/u, "");
+        roots.set(manifest.id, sourceRoot);
+    }
+    return roots;
+}
+function adapterFilePathWithinRoot(filePath, sourceRoot) {
+    const normalized = normalizeRelativePath(filePath);
+    if (!sourceRoot) {
+        return normalized;
+    }
+    if (!normalized.startsWith(`${sourceRoot}/`)) {
+        return null;
+    }
+    return normalized.slice(sourceRoot.length + 1);
 }
 async function readHostedRunFailureResult(root, workload, options) {
     const traceFiles = await readRuntimeTraceFiles(root, workload);
@@ -1788,16 +2441,16 @@ async function readWorkbenchRunWorkloadResult(root, workload, options = {}) {
     const primaryOperation = purpose === "improve"
         ? "optimizer.improve"
         : "engine.run";
-    const primaryResult = [...(options.phaseResults ?? [])]
+    const primaryResult = [...(options.operationResults ?? [])]
         .reverse()
         .find((result) => result.operation === primaryOperation);
     const resultPayload = jsonRecord(primaryResult?.value);
     const usage = mergeUsageSummaries([
         options.usage,
-        ...(options.phaseResults ?? []).map(adapterOperationUsageSummary),
+        ...(options.operationResults ?? []).map(adapterOperationUsageSummary),
     ]);
-    const metrics = normalizeRewardMetrics(resultPayload.metrics);
-    const cases = normalizeRewardCases(resultPayload.cases);
+    const metrics = normalizeResultMetrics(resultPayload.metrics);
+    const cases = normalizeResultCases(resultPayload.cases);
     const includeResultScoring = purpose === "attempt";
     const files = [...outputFiles, ...traceFiles].sort((left, right) => left.path.localeCompare(right.path));
     const subjectPatch = purpose === "improve" ? primaryResult?.value : undefined;
@@ -1809,6 +2462,7 @@ async function readWorkbenchRunWorkloadResult(root, workload, options = {}) {
     return {
         files,
         fileChanges: declaredChanges,
+        ...(options.operationResults ? { operationResults: [...options.operationResults] } : {}),
         ...(subjectPatch ? { subjectPatch } : {}),
         ...(engineResult ? { result: engineResult } : {}),
         ...(includeResultScoring && metrics ? { metrics } : {}),
@@ -1835,10 +2489,10 @@ async function readRuntimeTraceFiles(root, workload) {
     const path = await importNodeModule(nodeBuiltin("path"));
     const traceRoot = path.join(outputDir(root), ".workbench", "traces", workload.job.id);
     const purpose = readWorkloadExecutionPurpose(workload);
-    const outputTraceRoot = workbenchTracePhaseDirectory({
+    const outputTraceRoot = workbenchTraceExecutionDirectory({
         sequence: 1,
         runId: workload.job.runId,
-        phase: purpose,
+        purpose,
     });
     return (await readSurfaceFiles(traceRoot)).map((file) => ({
         ...file,
@@ -1868,13 +2522,13 @@ function createHostedWorkloadShellCommand(root, command, prefix = "", okExitCode
         'exit "$status"',
     ].join("; ");
 }
-async function resetHostedWorkloadPhaseOutput(root, _phase) {
+async function resetHostedWorkloadStepOutput(root) {
     const fs = await importNodeModule(nodeBuiltin("fs/promises"));
     await fs
         .rm(workbenchAdapterOperationResultPath(outputDir(root)), { force: true })
         .catch(() => undefined);
 }
-async function writeWorkbenchAdapterRequest(root, workload, execution, phase, auth, manifests) {
+async function writeWorkbenchAdapterRequest(root, workload, execution, step, auth, manifests) {
     const [fs, path] = await Promise.all([
         importNodeModule(nodeBuiltin("fs/promises")),
         importNodeModule(nodeBuiltin("path")),
@@ -1882,13 +2536,13 @@ async function writeWorkbenchAdapterRequest(root, workload, execution, phase, au
     const requestPath = path.join(root, ".workbench", "request.json");
     await fs.mkdir(path.dirname(requestPath), { recursive: true });
     const casePrompt = workload.engineCaseSpec?.prompt;
-    const adapter = phase.adapter ?? execution.adapter;
+    const adapter = step.adapter ?? execution.adapter;
     const subjectCommand = adapterProtocolCommandSpec(workload.spec.run, "subject.run", manifests).command;
     await fs.writeFile(requestPath, `${JSON.stringify({
         protocol: "workbench.adapter.v3",
         id: execution.id,
         jobId: workload.job.id,
-        operation: phase.operation,
+        operation: step.operation,
         invocation: {
             use: adapter.use,
             with: adapterConfigRecord(adapter, manifests),
@@ -1903,6 +2557,7 @@ async function writeWorkbenchAdapterRequest(root, workload, execution, phase, au
             subject: {
                 id: workload.subjectId,
                 path: workload.spec.subject.files.path,
+                ...(workload.spec.subject.prepare ? { prepare: { ...workload.spec.subject.prepare } } : {}),
                 run: {
                     ...workload.spec.run,
                     command: subjectCommand,
@@ -1923,14 +2578,12 @@ async function writeWorkbenchAdapterRequest(root, workload, execution, phase, au
         },
         paths: {
             workspace: root,
-            cwd: root,
             output: outputDir(root),
             result: workbenchAdapterOperationResultPath(outputDir(root)),
             subject: subjectDir(root),
             ...(workload.engineCaseSpec ? { case: caseDir(root) } : {}),
             traces: tracesDir(root),
-            ...(phase.kind === "engine" ? { enginePrivate: runtimeEnginePrivateDir(root) } : {}),
-            logs: runtimeLogsDir(root),
+            ...(step.kind === "engine" ? { enginePrivate: runtimeEnginePrivateDir(root) } : {}),
         },
     }, null, 2)}\n`);
     return requestPath;
@@ -1945,7 +2598,29 @@ function requireOptimizerEdits(spec) {
     }
     return edits;
 }
-function createHostedWorkloadPhaseEnv(root, adapterRequestPath, adapterEnv = {}) {
+function createHostedWorkloadAdapterEnv(root, adapterRequestPath, adapterEnv = {}, options = {}, runtimeEnv = {}) {
+    const env = createHostedWorkloadBaseEnv();
+    env.WORKBENCH_ADAPTER_REQUEST = adapterRequestPath;
+    env.WORKBENCH_OUTPUT = outputDir(root);
+    env.WORKBENCH_RESULT = workbenchAdapterOperationResultPath(outputDir(root));
+    if (options.adapterRoot) {
+        env.WORKBENCH_ADAPTER_ROOT = options.adapterRoot;
+        env.WORKBENCH_WORKSPACE_ROOT = root;
+        env.PATH = [
+            `${options.adapterRoot}/node_modules/.bin`,
+            env.PATH,
+        ].filter(Boolean).join(":");
+    }
+    Object.assign(env, adapterEnv);
+    Object.assign(env, runtimeEnv);
+    return env;
+}
+function createHostedWorkloadPrepareEnv(root) {
+    const env = createHostedWorkloadBaseEnv();
+    env.WORKBENCH_OUTPUT = outputDir(root);
+    return env;
+}
+function createHostedWorkloadBaseEnv() {
     const env = {};
     for (const [key, value] of Object.entries(process.env)) {
         if (typeof value === "string") {
@@ -1957,20 +2632,52 @@ function createHostedWorkloadPhaseEnv(root, adapterRequestPath, adapterEnv = {})
             delete env[key];
         }
     }
-    const runtimeBins = [
+    const runtimeBins = uniquePathEntries([
+        ...nodeModuleBinDirsForAncestors(process.cwd()),
+        ...nodeModuleBinDirsForAncestors(path.dirname(fileURLToPath(import.meta.url))),
+        "/app/node_modules/.bin",
         "/workbench-runtime/node_modules/.bin",
         "/workbench-runtime/products/workbench/node_modules/.bin",
-    ].join(":");
-    const systemBins = "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin";
-    env.PATH = process.env.PATH
-        ? `${systemBins}:${runtimeBins}:${process.env.PATH}`
-        : `${systemBins}:${runtimeBins}`;
-    env.WORKBENCH_ADAPTER_REQUEST = adapterRequestPath;
-    env.WORKBENCH_OUTPUT = outputDir(root);
-    env.WORKBENCH_RESULT = workbenchAdapterOperationResultPath(outputDir(root));
-    Object.assign(env, adapterEnv);
+    ]);
+    env.PATH = uniquePathEntries([
+        path.dirname(process.execPath),
+        "/usr/local/sbin",
+        "/usr/local/bin",
+        "/usr/sbin",
+        "/usr/bin",
+        "/sbin",
+        "/bin",
+        ...runtimeBins,
+        ...(process.env.PATH ? process.env.PATH.split(path.delimiter) : []),
+    ]).join(path.delimiter);
     return env;
 }
+function nodeModuleBinDirsForAncestors(start) {
+    const dirs = [];
+    let current = path.resolve(start);
+    for (let depth = 0; depth < 12; depth += 1) {
+        dirs.push(path.join(current, "node_modules", ".bin"));
+        const parent = path.dirname(current);
+        if (parent === current) {
+            break;
+        }
+        current = parent;
+    }
+    return dirs;
+}
+function uniquePathEntries(entries) {
+    const seen = new Set();
+    const output = [];
+    for (const entry of entries) {
+        const trimmed = entry.trim();
+        if (!trimmed || seen.has(trimmed)) {
+            continue;
+        }
+        seen.add(trimmed);
+        output.push(trimmed);
+    }
+    return output;
+}
 function readWorkloadExecutionPurpose(workload) {
     const purpose = workbenchExecutionPurpose(workload.job);
     if (purpose === "improve" || purpose === "attempt") {
@@ -2005,35 +2712,6 @@ function runtimePrivateDir(root) {
 function runtimeEnginePrivateDir(root) {
     return `${runtimePrivateDir(root)}/engine`;
 }
-function runtimeLogsDir(root) {
-    return `${root}/logs`;
-}
-function runtimeLogsAgentDir(root) {
-    return `${runtimeLogsDir(root)}/agent`;
-}
-function runtimeLogsVerifierDir(root) {
-    return `${runtimeLogsDir(root)}/verifier`;
-}
-function assertMutableWorkspaceFiles(files, label) {
-    const reserved = files
-        .map((file) => normalizeRelativePath(file.path))
-        .filter(isRuntimeReservedWorkspacePath);
-    if (reserved.length > 0) {
-        throw new Error(`${label} cannot target runtime-reserved workspace paths: ${reserved.join(", ")}.`);
-    }
-}
-function isRuntimeReservedWorkspacePath(normalizedPath) {
-    return normalizedPath === ".workbench" ||
-        normalizedPath.startsWith(".workbench/") ||
-        normalizedPath === "input" ||
-        normalizedPath.startsWith("input/") ||
-        normalizedPath === "output" ||
-        normalizedPath.startsWith("output/") ||
-        normalizedPath === "logs" ||
-        normalizedPath.startsWith("logs/") ||
-        normalizedPath === "private" ||
-        normalizedPath.startsWith("private/");
-}
 async function writeSurfaceFiles(root, files) {
     const fs = await importNodeModule(nodeBuiltin("fs/promises"));
     const path = await importNodeModule(nodeBuiltin("path"));
@@ -2097,7 +2775,7 @@ function encodeSurfaceSnapshotContent(body, utf8Decoder) {
         };
     }
 }
-function normalizeRewardMetrics(value) {
+function normalizeResultMetrics(value) {
     if (!value || typeof value !== "object" || Array.isArray(value)) {
         return undefined;
     }
@@ -2109,7 +2787,7 @@ function normalizeRewardMetrics(value) {
     }
     return Object.keys(metrics).length > 0 ? metrics : undefined;
 }
-function normalizeRewardCases(value) {
+function normalizeResultCases(value) {
     if (!Array.isArray(value)) {
         return undefined;
     }
@@ -2122,7 +2800,7 @@ function normalizeRewardCases(value) {
         if (!id) {
             return [];
         }
-        const metrics = normalizeRewardMetrics(record.metrics) ?? {};
+        const metrics = normalizeResultMetrics(record.metrics) ?? {};
         const status = record.status === "completed" || record.status === "error"
             ? record.status
             : undefined;
@@ -2146,9 +2824,7 @@ function normalizeRewardCases(value) {
                     : undefined;
                 const pass = typeof criterionRecord.pass === "boolean"
                     ? criterionRecord.pass
-                    : score !== undefined
-                        ? score >= 0.5
-                        : undefined;
+                    : undefined;
                 if (!criterionId || score === undefined || pass === undefined) {
                     return [];
                 }
@@ -2261,13 +2937,13 @@ function evaluateSample(args) {
     if (typeof sampleScore !== "number" || !Number.isFinite(sampleScore)) {
         throw new Error("Evaluation sample requires an engine result with a finite numeric score.");
     }
-    const cases = args.workload.cases?.length ? args.workload.cases : undefined;
     const metrics = args.workload.metrics ?? {
         score: sampleScore,
     };
     if (metrics.score === undefined) {
         metrics.score = sampleScore;
     }
+    const cases = args.workload.cases?.length ? args.workload.cases : undefined;
     const feedback = {
         ...(args.workload.summary !== undefined
             ? { summary: args.workload.summary }
@@ -2295,7 +2971,7 @@ function evaluateSample(args) {
         feedback,
     };
 }
-function normalizeSampleJobOutput(value, fallbackFiles = []) {
+function normalizeSampleJobOutput(value) {
     if (!value || typeof value !== "object" || Array.isArray(value)) {
         return null;
     }
@@ -2314,9 +2990,6 @@ function normalizeSampleJobOutput(value, fallbackFiles = []) {
         !Number.isFinite(record.attemptIndex)) {
         return null;
     }
-    const sampleFiles = files.length > 0
-        ? files
-        : fallbackFiles.map((file) => ({ ...file }));
     return {
         subjectId: record.subjectId,
         attemptIndex: record.attemptIndex,
@@ -2324,10 +2997,10 @@ function normalizeSampleJobOutput(value, fallbackFiles = []) {
         fileChanges: Array.isArray(record.fileChanges)
             ? record.fileChanges.filter((entry) => typeof entry === "string")
             : [],
-        files: sampleFiles,
+        files,
         traces: Array.isArray(record.traces)
             ? record.traces.filter((entry) => typeof entry === "string")
-            : traceFilePaths(sampleFiles),
+            : traceFilePaths(files),
     };
 }
 function normalizeEvaluationSampleOutputs(args) {
@@ -2498,8 +3171,16 @@ function compareSampleOutputs(left, right) {
     }
     return left.sample.id.localeCompare(right.sample.id);
 }
-function createEvaluationRecord(subjectId, rawSamples) {
-    const samples = mergeEvaluationSampleRecords(rawSamples);
+function createEvaluationRecord(subjectId, subjectName, rawSamples) {
+    const samples = mergeEvaluationSampleRecords(rawSamples).map((sample) => subjectName
+        ? {
+            ...sample,
+            subject: {
+                ...sample.subject,
+                label: subjectName,
+            },
+        }
+        : sample);
     const startedAt = minTimestamp(samples.flatMap((sample) => (sample.startedAt ? [sample.startedAt] : [])));
     const finishedAt = maxTimestamp(samples.flatMap((sample) => (sample.finishedAt ? [sample.finishedAt] : [])));
     const durationValues = samples.flatMap((sample) => typeof sample.durationMs === "number" ? [sample.durationMs] : []);
@@ -2513,6 +3194,7 @@ function createEvaluationRecord(subjectId, rawSamples) {
         subject: {
             id: subjectId,
             kind: "subject",
+            ...(subjectName ? { label: subjectName } : {}),
         },
         status: samples.length > 0 && completedSampleCount === samples.length
             ? "completed"
@@ -2533,6 +3215,10 @@ function createEvaluationRecord(subjectId, rawSamples) {
         samples,
     };
 }
+function normalizedSubjectDisplayName(value) {
+    const normalized = value?.trim();
+    return normalized ? normalized : null;
+}
 function aggregateSampleMetrics(samples) {
     const metricNames = new Set(samples.flatMap((sample) => Object.keys(sample.metrics ?? {})));
     if (metricNames.size === 0) {
@@ -2563,14 +3249,14 @@ function mergeEvaluationSampleRecords(samples) {
 function mergeEvaluationSampleGroup(group) {
     const first = group[0];
     if (group.length === 1) {
-        return normalizeSingleCaseDurations(first);
+        return first;
     }
     const startedAt = minTimestamp(group.flatMap((sample) => (sample.startedAt ? [sample.startedAt] : [])));
     const finishedAt = maxTimestamp(group.flatMap((sample) => (sample.finishedAt ? [sample.finishedAt] : [])));
     const durationMs = startedAt && finishedAt
         ? Math.max(0, Date.parse(finishedAt) - Date.parse(startedAt))
         : undefined;
-    const cases = group.flatMap((sample) => normalizeCaseDurations(sample));
+    const cases = group.flatMap((sample) => sample.cases ?? []);
     const metrics = aggregateSampleGroupMetrics(group);
     const usage = mergeUsageSummaries(group.map((sample) => sample.usage));
     const errors = group.flatMap((sample) => sample.error ? [sample.error] : []);
@@ -2588,22 +3274,6 @@ function mergeEvaluationSampleGroup(group) {
         ...(cases.length > 0 ? { cases } : {}),
     };
 }
-function normalizeSingleCaseDurations(sample) {
-    if (!sample.cases) {
-        return sample;
-    }
-    const cases = normalizeCaseDurations(sample);
-    return cases.length === sample.cases.length
-        ? { ...sample, cases }
-        : sample;
-}
-function normalizeCaseDurations(sample) {
-    return (sample.cases ?? []).map((caseResult) => (typeof caseResult.durationMs === "number" ||
-        sample.cases?.length !== 1 ||
-        typeof sample.durationMs !== "number"
-        ? caseResult
-        : { ...caseResult, durationMs: sample.durationMs }));
-}
 function aggregateSampleGroupMetrics(group) {
     const metricNames = new Set(group.flatMap((sample) => Object.keys(sample.metrics ?? {})));
     if (metricNames.size === 0) {