@agentv/core 2.15.0 → 2.16.0

package/dist/index.d.cts

@@ -514,9 +514,34 @@ type RepoConfig = {
     readonly checkout?: RepoCheckout;
     readonly clone?: RepoClone;
 };
-type ResetConfig = {
-    readonly strategy?: 'none' | 'hard' | 'recreate';
-    readonly after_each?: boolean;
+type WorkspaceHookConfig = {
+    /** Optional command array to execute (e.g., ["bun", "run", "setup.ts"]) */
+    readonly command?: readonly string[];
+    /** @deprecated Use `command` instead */
+    readonly script?: readonly string[];
+    /** Optional timeout in milliseconds */
+    readonly timeout_ms?: number;
+    readonly timeoutMs?: number;
+    /** Optional working directory for command execution */
+    readonly cwd?: string;
+    /** Optional reset policy for this hook */
+    readonly reset?: 'none' | 'fast' | 'strict';
+    /** Optional cleanup policy for this hook */
+    readonly clean?: 'always' | 'on_success' | 'on_failure' | 'never';
+};
+type WorkspaceHooksConfig = {
+    /** Runs once before first test in the workspace lifecycle */
+    readonly before_all_tests?: WorkspaceHookConfig;
+    /** Runs before each test case */
+    readonly before_each_test?: WorkspaceHookConfig;
+    /** Runs after each test case */
+    readonly after_each_test?: WorkspaceHookConfig;
+    /** Runs once after final test in the workspace lifecycle */
+    readonly after_all_tests?: WorkspaceHookConfig;
+    /** Runs when reusing a pooled workspace slot */
+    readonly on_reuse?: WorkspaceHookConfig;
+    /** Runs/controls behavior when workspace lifecycle finishes */
+    readonly on_finish?: WorkspaceHookConfig;
 };
 type WorkspaceConfig = {
     /** Template directory or .code-workspace file. Directories are copied to temp workspace.
@@ -526,16 +551,14 @@ type WorkspaceConfig = {
     readonly isolation?: 'shared' | 'per_test';
     /** Repository definitions to clone/checkout into workspace */
     readonly repos?: readonly RepoConfig[];
-    /** Reset configuration for repos between test runs */
-    readonly reset?: ResetConfig;
-    /** Command to run once before first test (after workspace creation, before git baseline) */
-    readonly before_all?: WorkspaceScriptConfig;
-    /** Command to run once after last test (before workspace cleanup) */
-    readonly after_all?: WorkspaceScriptConfig;
-    /** Command to run before each test */
-    readonly before_each?: WorkspaceScriptConfig;
-    /** Command to run after each test (e.g., git reset for workspace reuse) */
-    readonly after_each?: WorkspaceScriptConfig;
+    /** Workspace lifecycle hooks */
+    readonly hooks?: WorkspaceHooksConfig;
+    /** Workspace materialization mode */
+    readonly mode?: 'pooled' | 'ephemeral' | 'static';
+    /** Required when mode=static: use this existing directory directly */
+    readonly static_path?: string;
+    /** @deprecated Use mode=pooled|ephemeral|static */
+    readonly pool?: boolean;
 };
 type CodeEvaluatorConfig = {
     readonly name: string;
@@ -2347,34 +2370,18 @@ declare class DeterministicAssertionEvaluator implements Evaluator {
 }
 
 declare class RepoManager {
-    private readonly cacheDir;
     private readonly verbose;
-    constructor(cacheDir?: string, verbose?: boolean);
+    constructor(verbose?: boolean);
     private runGit;
     /**
-     * Ensure a bare mirror cache exists for the given source.
-     * Creates on first access, fetches updates on subsequent calls.
-     * Returns the absolute path to the cache directory.
-     */
-    ensureCache(source: RepoSource, depth?: number, resolve?: 'remote' | 'local'): Promise<string>;
-    /**
-     * Clone a repo from cache into the workspace at the configured path.
+     * Clone a repo directly from source into the workspace at the configured path.
      * Handles checkout, ref resolution, ancestor walking, shallow clone, sparse checkout.
      */
     materialize(repo: RepoConfig, workspacePath: string): Promise<void>;
     /** Materialize all repos into the workspace. */
     materializeAll(repos: readonly RepoConfig[], workspacePath: string): Promise<void>;
     /** Reset repos in workspace to their checkout state. */
-    reset(repos: readonly RepoConfig[], workspacePath: string, strategy: 'hard' | 'recreate'): Promise<void>;
-    /**
-     * Seed the cache from a local repository, setting the remote to a given URL.
-     * Useful for avoiding slow network clones when a local clone already exists.
-     */
-    seedCache(localPath: string, remoteUrl: string, opts?: {
-        force?: boolean;
-    }): Promise<string>;
-    /** Remove the entire cache directory. */
-    cleanCache(): Promise<void>;
+    reset(repos: readonly RepoConfig[], workspacePath: string, reset: 'fast' | 'strict'): Promise<void>;
 }
 
 type MaybePromise<T> = T | Promise<T>;
@@ -2406,6 +2413,10 @@ interface RunEvalCaseOptions {
     readonly keepWorkspaces?: boolean;
     /** Force cleanup of workspaces even on failure */
     readonly cleanupWorkspaces?: boolean;
+    /** Retention policy for temp workspaces on successful cases */
+    readonly retainOnSuccess?: 'keep' | 'cleanup';
+    /** Retention policy for temp workspaces on failed cases */
+    readonly retainOnFailure?: 'keep' | 'cleanup';
     /** Pre-created shared workspace path (shared across tests in a suite) */
     readonly sharedWorkspacePath?: string;
     /** Pre-initialized baseline commit for shared workspace */
@@ -2461,12 +2472,22 @@ interface RunEvaluationOptions {
     readonly totalBudgetUsd?: number;
     /** Execution error tolerance: true halts on first error */
     readonly failOnError?: FailOnError;
-    /** Opt-in: reuse materialized workspaces across eval runs */
+    /** Workspace pooling: true (default) enables pool, false disables, undefined defaults to true */
     readonly poolWorkspaces?: boolean;
     /** Maximum number of pool slots on disk (default: 10, max: 50) */
     readonly poolMaxSlots?: number;
     /** Pre-existing workspace directory to use directly (skips clone/copy/pool) */
     readonly workspace?: string;
+    /** Workspace materialization mode override */
+    readonly workspaceMode?: 'pooled' | 'ephemeral' | 'static';
+    /** Static workspace path override (used when workspaceMode=static) */
+    readonly workspacePath?: string;
+    /** Workspace clean policy override for pooled reset */
+    readonly workspaceClean?: 'standard' | 'full';
+    /** Retention policy override for successful cases */
+    readonly retainOnSuccess?: 'keep' | 'cleanup';
+    /** Retention policy override for failed cases */
+    readonly retainOnFailure?: 'keep' | 'cleanup';
 }
 declare function runEvaluation(options: RunEvaluationOptions): Promise<readonly EvaluationResult[]>;
 declare function runEvalCase(options: RunEvalCaseOptions): Promise<EvaluationResult>;
@@ -2978,6 +2999,7 @@ interface AcquireWorkspaceOptions {
     repos: readonly RepoConfig[];
     maxSlots: number;
     repoManager: RepoManager;
+    poolReset?: 'none' | 'fast' | 'strict';
 }
 interface PoolSlot {
     readonly index: number;
@@ -3106,7 +3128,6 @@ declare function toCamelCaseDeep(obj: unknown): unknown;
 
 declare function getAgentvHome(): string;
 declare function getWorkspacesRoot(): string;
-declare function getGitCacheRoot(): string;
 declare function getSubagentsRoot(): string;
 declare function getTraceStateRoot(): string;
 declare function getWorkspacePoolRoot(): string;
@@ -3265,4 +3286,4 @@ type AgentKernel = {
 };
 declare function createAgentKernel(): AgentKernel;
 
-export { type AcquireWorkspaceOptions, AgentJudgeEvaluator, type AgentJudgeEvaluatorConfig, type AgentJudgeEvaluatorOptions, type AgentKernel, type AgentVConfig$1 as AgentVConfig, type AgentVConfig as AgentVTsConfig, type AnthropicResolvedConfig, type ArgsMatchMode, type AssertionResult, type AssistantTestMessage, type AzureResolvedConfig, type CacheConfig, type ChildEvaluatorResult, type ClaudeResolvedConfig, type CliResolvedConfig, CodeEvaluator, type CodeEvaluatorConfig, type CodeEvaluatorOptions, type CompositeAggregatorConfig, CompositeEvaluator, type CompositeEvaluatorConfig, type CompositeEvaluatorOptions, type ConfidenceIntervalAggregation, type ContainsAllEvaluatorConfig, type ContainsAnyEvaluatorConfig, type ContainsEvaluatorConfig, type CopilotCliResolvedConfig, type CopilotSdkResolvedConfig, CostEvaluator, type CostEvaluatorConfig, type CostEvaluatorOptions, DEFAULT_EVALUATOR_TEMPLATE, DEFAULT_EVAL_PATTERNS, DEFAULT_EXPLORATION_TOOLS, DeterministicAssertionEvaluator, type EndsWithEvaluatorConfig, type EnsureSubagentsOptions, type EnsureSubagentsResult, type EnvLookup, type EqualsEvaluatorConfig, type EvalAssertionInput, type EvalCase, type EvalConfig, type EvalMetadata, type EvalRunResult, type EvalSuiteResult, type EvalSummary, type EvalTest, type EvalTestInput, type EvaluationCache, type EvaluationContext, type EvaluationResult, type EvaluationScore, type EvaluationVerdict, type Evaluator, type EvaluatorConfig, type EvaluatorDispatchContext, type EvaluatorFactory, type EvaluatorFactoryFn, type EvaluatorKind, EvaluatorRegistry, type EvaluatorResult, type ExecutionDefaults, type ExecutionError, type ExecutionMetrics, ExecutionMetricsEvaluator, type ExecutionMetricsEvaluatorConfig, type ExecutionMetricsEvaluatorOptions, type ExecutionStatus, type FailOnError, type FailureStage, FieldAccuracyEvaluator, type FieldAccuracyEvaluatorConfig, type FieldAccuracyEvaluatorOptions, type FieldAggregationType, type FieldConfig, type FieldMatchType, type GeminiResolvedConfig, type GenerateRubricsOptions, type IcontainsAllEvaluatorConfig, type IcontainsAnyEvaluatorConfig, type IcontainsEvaluatorConfig, type IsJsonEvaluatorConfig, type JsonObject, type JsonPrimitive, type JsonValue, LatencyEvaluator, type LatencyEvaluatorConfig, type LatencyEvaluatorOptions, LlmJudgeEvaluator, type LlmJudgeEvaluatorConfig, type LlmJudgeEvaluatorOptions, type LlmJudgePromptAssembly, type MeanAggregation, type Message, type MockResolvedConfig, OTEL_BACKEND_PRESETS, type OtelBackendPreset, type OtelExportOptions, OtelStreamingObserver, OtelTraceExporter, OtlpJsonFileExporter, type OutputMessage, type PassAtKAggregation, type PiAgentSdkResolvedConfig, type PiCodingAgentResolvedConfig, type PoolSlot, type ProgressEvent, type PromptInputs, type PromptScriptConfig, type Provider, type ProviderFactoryFn, type ProviderKind, ProviderRegistry, type ProviderRequest, type ProviderResponse, type ProviderStreamCallbacks, type ProviderTokenUsage, type RegexEvaluatorConfig, type RepoCheckout, type RepoClone, type RepoConfig, RepoManager, type RepoSource, type ResetConfig, type ResolvedTarget, type ResolvedWorkspaceTemplate, ResponseCache, type RubricItem, type RubricsEvaluatorConfig, type RunEvalCaseOptions, type RunEvaluationOptions, type ScoreRange, type ScriptExecutionContext, SimpleTraceFileExporter, type StartsWithEvaluatorConfig, type SystemTestMessage, TEST_MESSAGE_ROLES, type TargetAccessConfig, type TargetDefinition, TemplateNotDirectoryError, TemplateNotFoundError, type TestMessage, type TestMessageContent, type TestMessageRole, type TokenUsage, TokenUsageEvaluator, type TokenUsageEvaluatorConfig, type TokenUsageEvaluatorOptions, type ToolCall, type ToolTestMessage, ToolTrajectoryEvaluator, type ToolTrajectoryEvaluatorConfig, type ToolTrajectoryEvaluatorOptions, type ToolTrajectoryExpectedItem, type TraceComputeResult, type TraceSummary, type TrialAggregation, type TrialResult, type TrialStrategy, type TrialsConfig, type UserTestMessage, type VSCodeResolvedConfig, type WorkspaceConfig, WorkspaceCreationError, WorkspacePoolManager, type WorkspaceScriptConfig, assembleLlmJudgePrompt, avgToolDurationMs, buildDirectoryChain, buildOutputSchema, buildPromptInputs, buildRubricOutputSchema, buildScoreRangeOutputSchema, buildSearchRoots, calculateRubricScore, captureFileChanges, clampScore, cleanupEvalWorkspaces, cleanupWorkspace, computeTraceSummary, computeWorkspaceFingerprint, consumeClaudeLogEntries, consumeCodexLogEntries, consumeCopilotCliLogEntries, consumeCopilotSdkLogEntries, consumePiLogEntries, createAgentKernel, createBuiltinProviderRegistry, createBuiltinRegistry, createProvider, createTempWorkspace, deepEqual, defineConfig, detectFormat, discoverAssertions, discoverProviders, ensureVSCodeSubagents, evaluate, executeScript, executeWorkspaceScript, explorationRatio, extractCacheConfig, extractFailOnError, extractJsonBlob, extractTargetFromSuite, extractTargetsFromSuite, extractTargetsFromTestCase, extractTrialsConfig, fileExists, findGitRoot, freeformEvaluationSchema, generateRubrics, getAgentvHome, getGitCacheRoot, getHitCount, getSubagentsRoot, getTraceStateRoot, getWorkspacePath, getWorkspacePoolRoot, getWorkspacesRoot, initializeBaseline, isEvaluatorKind, isGuidelineFile, isJsonObject, isJsonValue, isNonEmptyString, isTestMessage, isTestMessageRole, listTargetNames, loadConfig, loadEvalCaseById, loadEvalCases, loadEvalSuite, loadTestById, loadTestSuite, loadTests, loadTsConfig, mergeExecutionMetrics, negateScore, normalizeLineEndings, parseJsonFromText, parseJsonSafe, readJsonFile, readTargetDefinitions, readTestSuiteMetadata, readTextFile, resolveAndCreateProvider, resolveFileReference, resolveTargetDefinition, resolveWorkspaceTemplate, rubricEvaluationSchema, runContainsAllAssertion, runContainsAnyAssertion, runContainsAssertion, runEndsWithAssertion, runEqualsAssertion, runEvalCase, runEvaluation, runIcontainsAllAssertion, runIcontainsAnyAssertion, runIcontainsAssertion, runIsJsonAssertion, runRegexAssertion, runStartsWithAssertion, scoreToVerdict, shouldEnableCache, shouldSkipCacheForTemperature, subscribeToClaudeLogEntries, subscribeToCodexLogEntries, subscribeToCopilotCliLogEntries, subscribeToCopilotSdkLogEntries, subscribeToPiLogEntries, substituteVariables, toCamelCaseDeep, toSnakeCaseDeep, tokensPerTool, trimBaselineResult };
+export { type AcquireWorkspaceOptions, AgentJudgeEvaluator, type AgentJudgeEvaluatorConfig, type AgentJudgeEvaluatorOptions, type AgentKernel, type AgentVConfig$1 as AgentVConfig, type AgentVConfig as AgentVTsConfig, type AnthropicResolvedConfig, type ArgsMatchMode, type AssertionResult, type AssistantTestMessage, type AzureResolvedConfig, type CacheConfig, type ChildEvaluatorResult, type ClaudeResolvedConfig, type CliResolvedConfig, CodeEvaluator, type CodeEvaluatorConfig, type CodeEvaluatorOptions, type CompositeAggregatorConfig, CompositeEvaluator, type CompositeEvaluatorConfig, type CompositeEvaluatorOptions, type ConfidenceIntervalAggregation, type ContainsAllEvaluatorConfig, type ContainsAnyEvaluatorConfig, type ContainsEvaluatorConfig, type CopilotCliResolvedConfig, type CopilotSdkResolvedConfig, CostEvaluator, type CostEvaluatorConfig, type CostEvaluatorOptions, DEFAULT_EVALUATOR_TEMPLATE, DEFAULT_EVAL_PATTERNS, DEFAULT_EXPLORATION_TOOLS, DeterministicAssertionEvaluator, type EndsWithEvaluatorConfig, type EnsureSubagentsOptions, type EnsureSubagentsResult, type EnvLookup, type EqualsEvaluatorConfig, type EvalAssertionInput, type EvalCase, type EvalConfig, type EvalMetadata, type EvalRunResult, type EvalSuiteResult, type EvalSummary, type EvalTest, type EvalTestInput, type EvaluationCache, type EvaluationContext, type EvaluationResult, type EvaluationScore, type EvaluationVerdict, type Evaluator, type EvaluatorConfig, type EvaluatorDispatchContext, type EvaluatorFactory, type EvaluatorFactoryFn, type EvaluatorKind, EvaluatorRegistry, type EvaluatorResult, type ExecutionDefaults, type ExecutionError, type ExecutionMetrics, ExecutionMetricsEvaluator, type ExecutionMetricsEvaluatorConfig, type ExecutionMetricsEvaluatorOptions, type ExecutionStatus, type FailOnError, type FailureStage, FieldAccuracyEvaluator, type FieldAccuracyEvaluatorConfig, type FieldAccuracyEvaluatorOptions, type FieldAggregationType, type FieldConfig, type FieldMatchType, type GeminiResolvedConfig, type GenerateRubricsOptions, type IcontainsAllEvaluatorConfig, type IcontainsAnyEvaluatorConfig, type IcontainsEvaluatorConfig, type IsJsonEvaluatorConfig, type JsonObject, type JsonPrimitive, type JsonValue, LatencyEvaluator, type LatencyEvaluatorConfig, type LatencyEvaluatorOptions, LlmJudgeEvaluator, type LlmJudgeEvaluatorConfig, type LlmJudgeEvaluatorOptions, type LlmJudgePromptAssembly, type MeanAggregation, type Message, type MockResolvedConfig, OTEL_BACKEND_PRESETS, type OtelBackendPreset, type OtelExportOptions, OtelStreamingObserver, OtelTraceExporter, OtlpJsonFileExporter, type OutputMessage, type PassAtKAggregation, type PiAgentSdkResolvedConfig, type PiCodingAgentResolvedConfig, type PoolSlot, type ProgressEvent, type PromptInputs, type PromptScriptConfig, type Provider, type ProviderFactoryFn, type ProviderKind, ProviderRegistry, type ProviderRequest, type ProviderResponse, type ProviderStreamCallbacks, type ProviderTokenUsage, type RegexEvaluatorConfig, type RepoCheckout, type RepoClone, type RepoConfig, RepoManager, type RepoSource, type ResolvedTarget, type ResolvedWorkspaceTemplate, ResponseCache, type RubricItem, type RubricsEvaluatorConfig, type RunEvalCaseOptions, type RunEvaluationOptions, type ScoreRange, type ScriptExecutionContext, SimpleTraceFileExporter, type StartsWithEvaluatorConfig, type SystemTestMessage, TEST_MESSAGE_ROLES, type TargetAccessConfig, type TargetDefinition, TemplateNotDirectoryError, TemplateNotFoundError, type TestMessage, type TestMessageContent, type TestMessageRole, type TokenUsage, TokenUsageEvaluator, type TokenUsageEvaluatorConfig, type TokenUsageEvaluatorOptions, type ToolCall, type ToolTestMessage, ToolTrajectoryEvaluator, type ToolTrajectoryEvaluatorConfig, type ToolTrajectoryEvaluatorOptions, type ToolTrajectoryExpectedItem, type TraceComputeResult, type TraceSummary, type TrialAggregation, type TrialResult, type TrialStrategy, type TrialsConfig, type UserTestMessage, type VSCodeResolvedConfig, type WorkspaceConfig, WorkspaceCreationError, type WorkspaceHookConfig, type WorkspaceHooksConfig, WorkspacePoolManager, type WorkspaceScriptConfig, assembleLlmJudgePrompt, avgToolDurationMs, buildDirectoryChain, buildOutputSchema, buildPromptInputs, buildRubricOutputSchema, buildScoreRangeOutputSchema, buildSearchRoots, calculateRubricScore, captureFileChanges, clampScore, cleanupEvalWorkspaces, cleanupWorkspace, computeTraceSummary, computeWorkspaceFingerprint, consumeClaudeLogEntries, consumeCodexLogEntries, consumeCopilotCliLogEntries, consumeCopilotSdkLogEntries, consumePiLogEntries, createAgentKernel, createBuiltinProviderRegistry, createBuiltinRegistry, createProvider, createTempWorkspace, deepEqual, defineConfig, detectFormat, discoverAssertions, discoverProviders, ensureVSCodeSubagents, evaluate, executeScript, executeWorkspaceScript, explorationRatio, extractCacheConfig, extractFailOnError, extractJsonBlob, extractTargetFromSuite, extractTargetsFromSuite, extractTargetsFromTestCase, extractTrialsConfig, fileExists, findGitRoot, freeformEvaluationSchema, generateRubrics, getAgentvHome, getHitCount, getSubagentsRoot, getTraceStateRoot, getWorkspacePath, getWorkspacePoolRoot, getWorkspacesRoot, initializeBaseline, isEvaluatorKind, isGuidelineFile, isJsonObject, isJsonValue, isNonEmptyString, isTestMessage, isTestMessageRole, listTargetNames, loadConfig, loadEvalCaseById, loadEvalCases, loadEvalSuite, loadTestById, loadTestSuite, loadTests, loadTsConfig, mergeExecutionMetrics, negateScore, normalizeLineEndings, parseJsonFromText, parseJsonSafe, readJsonFile, readTargetDefinitions, readTestSuiteMetadata, readTextFile, resolveAndCreateProvider, resolveFileReference, resolveTargetDefinition, resolveWorkspaceTemplate, rubricEvaluationSchema, runContainsAllAssertion, runContainsAnyAssertion, runContainsAssertion, runEndsWithAssertion, runEqualsAssertion, runEvalCase, runEvaluation, runIcontainsAllAssertion, runIcontainsAnyAssertion, runIcontainsAssertion, runIsJsonAssertion, runRegexAssertion, runStartsWithAssertion, scoreToVerdict, shouldEnableCache, shouldSkipCacheForTemperature, subscribeToClaudeLogEntries, subscribeToCodexLogEntries, subscribeToCopilotCliLogEntries, subscribeToCopilotSdkLogEntries, subscribeToPiLogEntries, substituteVariables, toCamelCaseDeep, toSnakeCaseDeep, tokensPerTool, trimBaselineResult };

package/dist/index.d.ts

@@ -514,9 +514,34 @@ type RepoConfig = {
     readonly checkout?: RepoCheckout;
     readonly clone?: RepoClone;
 };
-type ResetConfig = {
-    readonly strategy?: 'none' | 'hard' | 'recreate';
-    readonly after_each?: boolean;
+type WorkspaceHookConfig = {
+    /** Optional command array to execute (e.g., ["bun", "run", "setup.ts"]) */
+    readonly command?: readonly string[];
+    /** @deprecated Use `command` instead */
+    readonly script?: readonly string[];
+    /** Optional timeout in milliseconds */
+    readonly timeout_ms?: number;
+    readonly timeoutMs?: number;
+    /** Optional working directory for command execution */
+    readonly cwd?: string;
+    /** Optional reset policy for this hook */
+    readonly reset?: 'none' | 'fast' | 'strict';
+    /** Optional cleanup policy for this hook */
+    readonly clean?: 'always' | 'on_success' | 'on_failure' | 'never';
+};
+type WorkspaceHooksConfig = {
+    /** Runs once before first test in the workspace lifecycle */
+    readonly before_all_tests?: WorkspaceHookConfig;
+    /** Runs before each test case */
+    readonly before_each_test?: WorkspaceHookConfig;
+    /** Runs after each test case */
+    readonly after_each_test?: WorkspaceHookConfig;
+    /** Runs once after final test in the workspace lifecycle */
+    readonly after_all_tests?: WorkspaceHookConfig;
+    /** Runs when reusing a pooled workspace slot */
+    readonly on_reuse?: WorkspaceHookConfig;
+    /** Runs/controls behavior when workspace lifecycle finishes */
+    readonly on_finish?: WorkspaceHookConfig;
 };
 type WorkspaceConfig = {
     /** Template directory or .code-workspace file. Directories are copied to temp workspace.
@@ -526,16 +551,14 @@ type WorkspaceConfig = {
     readonly isolation?: 'shared' | 'per_test';
     /** Repository definitions to clone/checkout into workspace */
     readonly repos?: readonly RepoConfig[];
-    /** Reset configuration for repos between test runs */
-    readonly reset?: ResetConfig;
-    /** Command to run once before first test (after workspace creation, before git baseline) */
-    readonly before_all?: WorkspaceScriptConfig;
-    /** Command to run once after last test (before workspace cleanup) */
-    readonly after_all?: WorkspaceScriptConfig;
-    /** Command to run before each test */
-    readonly before_each?: WorkspaceScriptConfig;
-    /** Command to run after each test (e.g., git reset for workspace reuse) */
-    readonly after_each?: WorkspaceScriptConfig;
+    /** Workspace lifecycle hooks */
+    readonly hooks?: WorkspaceHooksConfig;
+    /** Workspace materialization mode */
+    readonly mode?: 'pooled' | 'ephemeral' | 'static';
+    /** Required when mode=static: use this existing directory directly */
+    readonly static_path?: string;
+    /** @deprecated Use mode=pooled|ephemeral|static */
+    readonly pool?: boolean;
 };
 type CodeEvaluatorConfig = {
     readonly name: string;
@@ -2347,34 +2370,18 @@ declare class DeterministicAssertionEvaluator implements Evaluator {
 }
 
 declare class RepoManager {
-    private readonly cacheDir;
     private readonly verbose;
-    constructor(cacheDir?: string, verbose?: boolean);
+    constructor(verbose?: boolean);
     private runGit;
     /**
-     * Ensure a bare mirror cache exists for the given source.
-     * Creates on first access, fetches updates on subsequent calls.
-     * Returns the absolute path to the cache directory.
-     */
-    ensureCache(source: RepoSource, depth?: number, resolve?: 'remote' | 'local'): Promise<string>;
-    /**
-     * Clone a repo from cache into the workspace at the configured path.
+     * Clone a repo directly from source into the workspace at the configured path.
      * Handles checkout, ref resolution, ancestor walking, shallow clone, sparse checkout.
      */
     materialize(repo: RepoConfig, workspacePath: string): Promise<void>;
     /** Materialize all repos into the workspace. */
     materializeAll(repos: readonly RepoConfig[], workspacePath: string): Promise<void>;
     /** Reset repos in workspace to their checkout state. */
-    reset(repos: readonly RepoConfig[], workspacePath: string, strategy: 'hard' | 'recreate'): Promise<void>;
-    /**
-     * Seed the cache from a local repository, setting the remote to a given URL.
-     * Useful for avoiding slow network clones when a local clone already exists.
-     */
-    seedCache(localPath: string, remoteUrl: string, opts?: {
-        force?: boolean;
-    }): Promise<string>;
-    /** Remove the entire cache directory. */
-    cleanCache(): Promise<void>;
+    reset(repos: readonly RepoConfig[], workspacePath: string, reset: 'fast' | 'strict'): Promise<void>;
 }
 
 type MaybePromise<T> = T | Promise<T>;
@@ -2406,6 +2413,10 @@ interface RunEvalCaseOptions {
     readonly keepWorkspaces?: boolean;
     /** Force cleanup of workspaces even on failure */
     readonly cleanupWorkspaces?: boolean;
+    /** Retention policy for temp workspaces on successful cases */
+    readonly retainOnSuccess?: 'keep' | 'cleanup';
+    /** Retention policy for temp workspaces on failed cases */
+    readonly retainOnFailure?: 'keep' | 'cleanup';
     /** Pre-created shared workspace path (shared across tests in a suite) */
     readonly sharedWorkspacePath?: string;
     /** Pre-initialized baseline commit for shared workspace */
@@ -2461,12 +2472,22 @@ interface RunEvaluationOptions {
     readonly totalBudgetUsd?: number;
     /** Execution error tolerance: true halts on first error */
     readonly failOnError?: FailOnError;
-    /** Opt-in: reuse materialized workspaces across eval runs */
+    /** Workspace pooling: true (default) enables pool, false disables, undefined defaults to true */
     readonly poolWorkspaces?: boolean;
     /** Maximum number of pool slots on disk (default: 10, max: 50) */
     readonly poolMaxSlots?: number;
     /** Pre-existing workspace directory to use directly (skips clone/copy/pool) */
     readonly workspace?: string;
+    /** Workspace materialization mode override */
+    readonly workspaceMode?: 'pooled' | 'ephemeral' | 'static';
+    /** Static workspace path override (used when workspaceMode=static) */
+    readonly workspacePath?: string;
+    /** Workspace clean policy override for pooled reset */
+    readonly workspaceClean?: 'standard' | 'full';
+    /** Retention policy override for successful cases */
+    readonly retainOnSuccess?: 'keep' | 'cleanup';
+    /** Retention policy override for failed cases */
+    readonly retainOnFailure?: 'keep' | 'cleanup';
 }
 declare function runEvaluation(options: RunEvaluationOptions): Promise<readonly EvaluationResult[]>;
 declare function runEvalCase(options: RunEvalCaseOptions): Promise<EvaluationResult>;
@@ -2978,6 +2999,7 @@ interface AcquireWorkspaceOptions {
     repos: readonly RepoConfig[];
     maxSlots: number;
     repoManager: RepoManager;
+    poolReset?: 'none' | 'fast' | 'strict';
 }
 interface PoolSlot {
     readonly index: number;
@@ -3106,7 +3128,6 @@ declare function toCamelCaseDeep(obj: unknown): unknown;
 
 declare function getAgentvHome(): string;
 declare function getWorkspacesRoot(): string;
-declare function getGitCacheRoot(): string;
 declare function getSubagentsRoot(): string;
 declare function getTraceStateRoot(): string;
 declare function getWorkspacePoolRoot(): string;
@@ -3265,4 +3286,4 @@ type AgentKernel = {
 };
 declare function createAgentKernel(): AgentKernel;
 
-export { type AcquireWorkspaceOptions, AgentJudgeEvaluator, type AgentJudgeEvaluatorConfig, type AgentJudgeEvaluatorOptions, type AgentKernel, type AgentVConfig$1 as AgentVConfig, type AgentVConfig as AgentVTsConfig, type AnthropicResolvedConfig, type ArgsMatchMode, type AssertionResult, type AssistantTestMessage, type AzureResolvedConfig, type CacheConfig, type ChildEvaluatorResult, type ClaudeResolvedConfig, type CliResolvedConfig, CodeEvaluator, type CodeEvaluatorConfig, type CodeEvaluatorOptions, type CompositeAggregatorConfig, CompositeEvaluator, type CompositeEvaluatorConfig, type CompositeEvaluatorOptions, type ConfidenceIntervalAggregation, type ContainsAllEvaluatorConfig, type ContainsAnyEvaluatorConfig, type ContainsEvaluatorConfig, type CopilotCliResolvedConfig, type CopilotSdkResolvedConfig, CostEvaluator, type CostEvaluatorConfig, type CostEvaluatorOptions, DEFAULT_EVALUATOR_TEMPLATE, DEFAULT_EVAL_PATTERNS, DEFAULT_EXPLORATION_TOOLS, DeterministicAssertionEvaluator, type EndsWithEvaluatorConfig, type EnsureSubagentsOptions, type EnsureSubagentsResult, type EnvLookup, type EqualsEvaluatorConfig, type EvalAssertionInput, type EvalCase, type EvalConfig, type EvalMetadata, type EvalRunResult, type EvalSuiteResult, type EvalSummary, type EvalTest, type EvalTestInput, type EvaluationCache, type EvaluationContext, type EvaluationResult, type EvaluationScore, type EvaluationVerdict, type Evaluator, type EvaluatorConfig, type EvaluatorDispatchContext, type EvaluatorFactory, type EvaluatorFactoryFn, type EvaluatorKind, EvaluatorRegistry, type EvaluatorResult, type ExecutionDefaults, type ExecutionError, type ExecutionMetrics, ExecutionMetricsEvaluator, type ExecutionMetricsEvaluatorConfig, type ExecutionMetricsEvaluatorOptions, type ExecutionStatus, type FailOnError, type FailureStage, FieldAccuracyEvaluator, type FieldAccuracyEvaluatorConfig, type FieldAccuracyEvaluatorOptions, type FieldAggregationType, type FieldConfig, type FieldMatchType, type GeminiResolvedConfig, type GenerateRubricsOptions, type IcontainsAllEvaluatorConfig, type IcontainsAnyEvaluatorConfig, type IcontainsEvaluatorConfig, type IsJsonEvaluatorConfig, type JsonObject, type JsonPrimitive, type JsonValue, LatencyEvaluator, type LatencyEvaluatorConfig, type LatencyEvaluatorOptions, LlmJudgeEvaluator, type LlmJudgeEvaluatorConfig, type LlmJudgeEvaluatorOptions, type LlmJudgePromptAssembly, type MeanAggregation, type Message, type MockResolvedConfig, OTEL_BACKEND_PRESETS, type OtelBackendPreset, type OtelExportOptions, OtelStreamingObserver, OtelTraceExporter, OtlpJsonFileExporter, type OutputMessage, type PassAtKAggregation, type PiAgentSdkResolvedConfig, type PiCodingAgentResolvedConfig, type PoolSlot, type ProgressEvent, type PromptInputs, type PromptScriptConfig, type Provider, type ProviderFactoryFn, type ProviderKind, ProviderRegistry, type ProviderRequest, type ProviderResponse, type ProviderStreamCallbacks, type ProviderTokenUsage, type RegexEvaluatorConfig, type RepoCheckout, type RepoClone, type RepoConfig, RepoManager, type RepoSource, type ResetConfig, type ResolvedTarget, type ResolvedWorkspaceTemplate, ResponseCache, type RubricItem, type RubricsEvaluatorConfig, type RunEvalCaseOptions, type RunEvaluationOptions, type ScoreRange, type ScriptExecutionContext, SimpleTraceFileExporter, type StartsWithEvaluatorConfig, type SystemTestMessage, TEST_MESSAGE_ROLES, type TargetAccessConfig, type TargetDefinition, TemplateNotDirectoryError, TemplateNotFoundError, type TestMessage, type TestMessageContent, type TestMessageRole, type TokenUsage, TokenUsageEvaluator, type TokenUsageEvaluatorConfig, type TokenUsageEvaluatorOptions, type ToolCall, type ToolTestMessage, ToolTrajectoryEvaluator, type ToolTrajectoryEvaluatorConfig, type ToolTrajectoryEvaluatorOptions, type ToolTrajectoryExpectedItem, type TraceComputeResult, type TraceSummary, type TrialAggregation, type TrialResult, type TrialStrategy, type TrialsConfig, type UserTestMessage, type VSCodeResolvedConfig, type WorkspaceConfig, WorkspaceCreationError, WorkspacePoolManager, type WorkspaceScriptConfig, assembleLlmJudgePrompt, avgToolDurationMs, buildDirectoryChain, buildOutputSchema, buildPromptInputs, buildRubricOutputSchema, buildScoreRangeOutputSchema, buildSearchRoots, calculateRubricScore, captureFileChanges, clampScore, cleanupEvalWorkspaces, cleanupWorkspace, computeTraceSummary, computeWorkspaceFingerprint, consumeClaudeLogEntries, consumeCodexLogEntries, consumeCopilotCliLogEntries, consumeCopilotSdkLogEntries, consumePiLogEntries, createAgentKernel, createBuiltinProviderRegistry, createBuiltinRegistry, createProvider, createTempWorkspace, deepEqual, defineConfig, detectFormat, discoverAssertions, discoverProviders, ensureVSCodeSubagents, evaluate, executeScript, executeWorkspaceScript, explorationRatio, extractCacheConfig, extractFailOnError, extractJsonBlob, extractTargetFromSuite, extractTargetsFromSuite, extractTargetsFromTestCase, extractTrialsConfig, fileExists, findGitRoot, freeformEvaluationSchema, generateRubrics, getAgentvHome, getGitCacheRoot, getHitCount, getSubagentsRoot, getTraceStateRoot, getWorkspacePath, getWorkspacePoolRoot, getWorkspacesRoot, initializeBaseline, isEvaluatorKind, isGuidelineFile, isJsonObject, isJsonValue, isNonEmptyString, isTestMessage, isTestMessageRole, listTargetNames, loadConfig, loadEvalCaseById, loadEvalCases, loadEvalSuite, loadTestById, loadTestSuite, loadTests, loadTsConfig, mergeExecutionMetrics, negateScore, normalizeLineEndings, parseJsonFromText, parseJsonSafe, readJsonFile, readTargetDefinitions, readTestSuiteMetadata, readTextFile, resolveAndCreateProvider, resolveFileReference, resolveTargetDefinition, resolveWorkspaceTemplate, rubricEvaluationSchema, runContainsAllAssertion, runContainsAnyAssertion, runContainsAssertion, runEndsWithAssertion, runEqualsAssertion, runEvalCase, runEvaluation, runIcontainsAllAssertion, runIcontainsAnyAssertion, runIcontainsAssertion, runIsJsonAssertion, runRegexAssertion, runStartsWithAssertion, scoreToVerdict, shouldEnableCache, shouldSkipCacheForTemperature, subscribeToClaudeLogEntries, subscribeToCodexLogEntries, subscribeToCopilotCliLogEntries, subscribeToCopilotSdkLogEntries, subscribeToPiLogEntries, substituteVariables, toCamelCaseDeep, toSnakeCaseDeep, tokensPerTool, trimBaselineResult };
+export { type AcquireWorkspaceOptions, AgentJudgeEvaluator, type AgentJudgeEvaluatorConfig, type AgentJudgeEvaluatorOptions, type AgentKernel, type AgentVConfig$1 as AgentVConfig, type AgentVConfig as AgentVTsConfig, type AnthropicResolvedConfig, type ArgsMatchMode, type AssertionResult, type AssistantTestMessage, type AzureResolvedConfig, type CacheConfig, type ChildEvaluatorResult, type ClaudeResolvedConfig, type CliResolvedConfig, CodeEvaluator, type CodeEvaluatorConfig, type CodeEvaluatorOptions, type CompositeAggregatorConfig, CompositeEvaluator, type CompositeEvaluatorConfig, type CompositeEvaluatorOptions, type ConfidenceIntervalAggregation, type ContainsAllEvaluatorConfig, type ContainsAnyEvaluatorConfig, type ContainsEvaluatorConfig, type CopilotCliResolvedConfig, type CopilotSdkResolvedConfig, CostEvaluator, type CostEvaluatorConfig, type CostEvaluatorOptions, DEFAULT_EVALUATOR_TEMPLATE, DEFAULT_EVAL_PATTERNS, DEFAULT_EXPLORATION_TOOLS, DeterministicAssertionEvaluator, type EndsWithEvaluatorConfig, type EnsureSubagentsOptions, type EnsureSubagentsResult, type EnvLookup, type EqualsEvaluatorConfig, type EvalAssertionInput, type EvalCase, type EvalConfig, type EvalMetadata, type EvalRunResult, type EvalSuiteResult, type EvalSummary, type EvalTest, type EvalTestInput, type EvaluationCache, type EvaluationContext, type EvaluationResult, type EvaluationScore, type EvaluationVerdict, type Evaluator, type EvaluatorConfig, type EvaluatorDispatchContext, type EvaluatorFactory, type EvaluatorFactoryFn, type EvaluatorKind, EvaluatorRegistry, type EvaluatorResult, type ExecutionDefaults, type ExecutionError, type ExecutionMetrics, ExecutionMetricsEvaluator, type ExecutionMetricsEvaluatorConfig, type ExecutionMetricsEvaluatorOptions, type ExecutionStatus, type FailOnError, type FailureStage, FieldAccuracyEvaluator, type FieldAccuracyEvaluatorConfig, type FieldAccuracyEvaluatorOptions, type FieldAggregationType, type FieldConfig, type FieldMatchType, type GeminiResolvedConfig, type GenerateRubricsOptions, type IcontainsAllEvaluatorConfig, type IcontainsAnyEvaluatorConfig, type IcontainsEvaluatorConfig, type IsJsonEvaluatorConfig, type JsonObject, type JsonPrimitive, type JsonValue, LatencyEvaluator, type LatencyEvaluatorConfig, type LatencyEvaluatorOptions, LlmJudgeEvaluator, type LlmJudgeEvaluatorConfig, type LlmJudgeEvaluatorOptions, type LlmJudgePromptAssembly, type MeanAggregation, type Message, type MockResolvedConfig, OTEL_BACKEND_PRESETS, type OtelBackendPreset, type OtelExportOptions, OtelStreamingObserver, OtelTraceExporter, OtlpJsonFileExporter, type OutputMessage, type PassAtKAggregation, type PiAgentSdkResolvedConfig, type PiCodingAgentResolvedConfig, type PoolSlot, type ProgressEvent, type PromptInputs, type PromptScriptConfig, type Provider, type ProviderFactoryFn, type ProviderKind, ProviderRegistry, type ProviderRequest, type ProviderResponse, type ProviderStreamCallbacks, type ProviderTokenUsage, type RegexEvaluatorConfig, type RepoCheckout, type RepoClone, type RepoConfig, RepoManager, type RepoSource, type ResolvedTarget, type ResolvedWorkspaceTemplate, ResponseCache, type RubricItem, type RubricsEvaluatorConfig, type RunEvalCaseOptions, type RunEvaluationOptions, type ScoreRange, type ScriptExecutionContext, SimpleTraceFileExporter, type StartsWithEvaluatorConfig, type SystemTestMessage, TEST_MESSAGE_ROLES, type TargetAccessConfig, type TargetDefinition, TemplateNotDirectoryError, TemplateNotFoundError, type TestMessage, type TestMessageContent, type TestMessageRole, type TokenUsage, TokenUsageEvaluator, type TokenUsageEvaluatorConfig, type TokenUsageEvaluatorOptions, type ToolCall, type ToolTestMessage, ToolTrajectoryEvaluator, type ToolTrajectoryEvaluatorConfig, type ToolTrajectoryEvaluatorOptions, type ToolTrajectoryExpectedItem, type TraceComputeResult, type TraceSummary, type TrialAggregation, type TrialResult, type TrialStrategy, type TrialsConfig, type UserTestMessage, type VSCodeResolvedConfig, type WorkspaceConfig, WorkspaceCreationError, type WorkspaceHookConfig, type WorkspaceHooksConfig, WorkspacePoolManager, type WorkspaceScriptConfig, assembleLlmJudgePrompt, avgToolDurationMs, buildDirectoryChain, buildOutputSchema, buildPromptInputs, buildRubricOutputSchema, buildScoreRangeOutputSchema, buildSearchRoots, calculateRubricScore, captureFileChanges, clampScore, cleanupEvalWorkspaces, cleanupWorkspace, computeTraceSummary, computeWorkspaceFingerprint, consumeClaudeLogEntries, consumeCodexLogEntries, consumeCopilotCliLogEntries, consumeCopilotSdkLogEntries, consumePiLogEntries, createAgentKernel, createBuiltinProviderRegistry, createBuiltinRegistry, createProvider, createTempWorkspace, deepEqual, defineConfig, detectFormat, discoverAssertions, discoverProviders, ensureVSCodeSubagents, evaluate, executeScript, executeWorkspaceScript, explorationRatio, extractCacheConfig, extractFailOnError, extractJsonBlob, extractTargetFromSuite, extractTargetsFromSuite, extractTargetsFromTestCase, extractTrialsConfig, fileExists, findGitRoot, freeformEvaluationSchema, generateRubrics, getAgentvHome, getHitCount, getSubagentsRoot, getTraceStateRoot, getWorkspacePath, getWorkspacePoolRoot, getWorkspacesRoot, initializeBaseline, isEvaluatorKind, isGuidelineFile, isJsonObject, isJsonValue, isNonEmptyString, isTestMessage, isTestMessageRole, listTargetNames, loadConfig, loadEvalCaseById, loadEvalCases, loadEvalSuite, loadTestById, loadTestSuite, loadTests, loadTsConfig, mergeExecutionMetrics, negateScore, normalizeLineEndings, parseJsonFromText, parseJsonSafe, readJsonFile, readTargetDefinitions, readTestSuiteMetadata, readTextFile, resolveAndCreateProvider, resolveFileReference, resolveTargetDefinition, resolveWorkspaceTemplate, rubricEvaluationSchema, runContainsAllAssertion, runContainsAnyAssertion, runContainsAssertion, runEndsWithAssertion, runEqualsAssertion, runEvalCase, runEvaluation, runIcontainsAllAssertion, runIcontainsAnyAssertion, runIcontainsAssertion, runIsJsonAssertion, runRegexAssertion, runStartsWithAssertion, scoreToVerdict, shouldEnableCache, shouldSkipCacheForTemperature, subscribeToClaudeLogEntries, subscribeToCodexLogEntries, subscribeToCopilotCliLogEntries, subscribeToCopilotSdkLogEntries, subscribeToPiLogEntries, substituteVariables, toCamelCaseDeep, toSnakeCaseDeep, tokensPerTool, trimBaselineResult };