@agentv/core 4.21.0 → 4.22.0-next.1

package/dist/index.d.cts

@@ -1912,6 +1912,10 @@ type ResultsExportConfig = {
     readonly auto_push?: boolean;
     readonly branch_prefix?: string;
 };
+type HooksConfig = {
+    /** Shell command to run once at agentv startup. stdout is parsed for env var exports. */
+    readonly before_session?: string;
+};
 type AgentVConfig$1 = {
     readonly required_version?: string;
     readonly eval_patterns?: readonly string[];
@@ -1919,6 +1923,7 @@ type AgentVConfig$1 = {
     readonly results?: {
         readonly export?: ResultsExportConfig;
     };
+    readonly hooks?: HooksConfig;
 };
 /**
  * Load optional .agentv/config.yaml configuration file.
@@ -2937,6 +2942,36 @@ declare const rubricEvaluationSchema: z.ZodObject<{
     }[];
     overall_reasoning: string;
 }>;
+declare const scoreRangeEvaluationSchema: z.ZodObject<{
+    checks: z.ZodArray<z.ZodObject<{
+        id: z.ZodString;
+        score: z.ZodNumber;
+        reasoning: z.ZodOptional<z.ZodString>;
+    }, "strip", z.ZodTypeAny, {
+        id: string;
+        score: number;
+        reasoning?: string | undefined;
+    }, {
+        id: string;
+        score: number;
+        reasoning?: string | undefined;
+    }>, "many">;
+    overall_reasoning: z.ZodOptional<z.ZodString>;
+}, "strip", z.ZodTypeAny, {
+    checks: {
+        id: string;
+        score: number;
+        reasoning?: string | undefined;
+    }[];
+    overall_reasoning?: string | undefined;
+}, {
+    checks: {
+        id: string;
+        score: number;
+        reasoning?: string | undefined;
+    }[];
+    overall_reasoning?: string | undefined;
+}>;
 
 declare class LlmGrader implements Grader {
     readonly kind = "llm-grader";
@@ -3537,6 +3572,21 @@ declare const AgentVConfigSchema: z.ZodObject<{
         maxDurationMs?: number | undefined;
         maxCostUsd?: number | undefined;
     }>>;
+    /** Lifecycle hooks */
+    hooks: z.ZodOptional<z.ZodObject<{
+        /**
+         * Shell command to run once at agentv startup, before any command executes.
+         * stdout is parsed for env var exports (`KEY=value` or `export KEY="value"`)
+         * and injected into process.env. Keys already set in the environment are
+         * not overwritten — existing env always takes priority.
+         * stderr is forwarded to the user. Non-zero exit aborts with an error.
+         */
+        beforeSession: z.ZodOptional<z.ZodString>;
+    }, "strip", z.ZodTypeAny, {
+        beforeSession?: string | undefined;
+    }, {
+        beforeSession?: string | undefined;
+    }>>;
 }, "strip", z.ZodTypeAny, {
     execution?: {
         workers?: number | undefined;
@@ -3546,6 +3596,9 @@ declare const AgentVConfigSchema: z.ZodObject<{
         keepWorkspaces?: boolean | undefined;
         otelFile?: string | undefined;
     } | undefined;
+    hooks?: {
+        beforeSession?: string | undefined;
+    } | undefined;
     cache?: {
         enabled?: boolean | undefined;
         path?: string | undefined;
@@ -3567,6 +3620,9 @@ declare const AgentVConfigSchema: z.ZodObject<{
         keepWorkspaces?: boolean | undefined;
         otelFile?: string | undefined;
     } | undefined;
+    hooks?: {
+        beforeSession?: string | undefined;
+    } | undefined;
     cache?: {
         enabled?: boolean | undefined;
         path?: string | undefined;
@@ -4346,6 +4402,56 @@ declare function discoverAssertions(registry: GraderRegistry, baseDir: string):
  */
 declare function discoverGraders(registry: GraderRegistry, baseDir: string): Promise<string[]>;
 
+/**
+ * Session hook execution for AgentV.
+ *
+ * Runs a shell command once at agentv startup and injects exported environment
+ * variables into the current process. This lets projects fetch secrets at
+ * runtime (e.g. from a vault) without needing a wrapper script.
+ *
+ * ## How it works
+ *
+ * 1. The command is run via `sh -c` (or `cmd /c` on Windows).
+ * 2. stdout is captured and parsed for env var exports.
+ * 3. stderr is forwarded to the process stderr so the user sees output.
+ * 4. Non-zero exit aborts with a clear error.
+ * 5. Parsed keys are injected into `process.env` — only for keys not already
+ *    set, so existing env always wins.
+ *
+ * ## Supported output formats
+ *
+ * Both shell-export and dotenv formats are accepted:
+ *   export KEY="value"   (shell export — quotes optional)
+ *   KEY=value            (dotenv — no export prefix)
+ *
+ * Lines that don't match either pattern are silently ignored.
+ *
+ * @module
+ */
+/**
+ * Parse env var lines from hook stdout.
+ *
+ * Accepts:
+ *   export KEY="value"   → { KEY: "value" }
+ *   export KEY=value     → { KEY: "value" }
+ *   KEY=value            → { KEY: "value" }
+ *
+ * Strips surrounding single or double quotes from values.
+ * Skips lines with empty keys or values that look like shell syntax.
+ */
+declare function parseEnvOutput(stdout: string): Record<string, string>;
+/**
+ * Run the before_session hook command and inject exported env vars into process.env.
+ *
+ * - Runs via shell (`sh -c` on POSIX, `cmd /c` on Windows)
+ * - Captured stdout is parsed for env vars; stderr is forwarded to process.stderr
+ * - Non-zero exit throws an Error with the command and exit code
+ * - Keys already set in process.env are NOT overwritten
+ *
+ * @param command Shell command string to execute
+ */
+declare function runBeforeSessionHook(command: string): void;
+
 /**
  * Core types for the transcript import pipeline.
  *
@@ -4663,4 +4769,4 @@ type AgentKernel = {
 };
 declare function createAgentKernel(): AgentKernel;
 
-export { type AcquireWorkspaceOptions, type AgentKernel, type AgentVConfig$1 as AgentVConfig, type AgentVResolvedConfig, type AgentVConfig as AgentVTsConfig, type AgentVConfig$1 as AgentVYamlConfig, type AnthropicResolvedConfig, type ApiFormat, type ArgsMatchMode, type AssertContext, type AssertEntry, type AssertFn, type AssertResult, type AssertionEntry, type AssertionResult, type AssistantTestMessage, type AzureResolvedConfig, type BenchmarkEntry, type BenchmarkRegistry, COMMON_TARGET_SETTINGS, type CacheConfig, type CheckedOutResultsRepoBranch, type ChildGraderResult, type ClaudeDiscoverOptions, type ClaudeResolvedConfig, type ClaudeSession, type CliResolvedConfig, CodeGrader, type CodeGraderConfig, type CodeGraderOptions, type CodexDiscoverOptions, type CodexSession, type CommandExecutor, type CompositeAggregatorConfig, CompositeGrader, type CompositeGraderConfig, type CompositeGraderOptions, type ConfidenceIntervalAggregation, type ContainsAllGraderConfig, type ContainsAnyGraderConfig, type ContainsGraderConfig, type Content, type ContentFile, type ContentImage, type ContentPreprocessorConfig, type ContentText, type ConversationAggregation, type ConversationMode, type ConversationTurn, type ConversationTurnInput, type CopilotCliResolvedConfig, type DiscoverOptions as CopilotDiscoverOptions, type CopilotLogResolvedConfig, type CopilotSdkResolvedConfig, type CopilotSession, type CopilotSessionMeta, CostGrader, type CostGraderConfig, type CostGraderOptions, type CreateContainerOptions, DEFAULT_CATEGORY, DEFAULT_EVAL_PATTERNS, DEFAULT_EXPLORATION_TOOLS, DEFAULT_GRADER_TEMPLATE, DEFAULT_THRESHOLD, type DependencyFailurePolicy, type DependencyResult, type DepsScanResult, DeterministicAssertionGrader, type DockerWorkspaceConfig, DockerWorkspaceProvider, type EndsWithGraderConfig, type EnsureSubagentsOptions, type EnsureSubagentsResult, type EnvLookup, type EqualsGraderConfig, type EvalAssertionInput, type EvalCase, type EvalConfig, type EvalMetadata, type EvalRunResult, type EvalSuiteResult, type EvalSummary, type EvalTargetRef, type EvalTest, type EvalTestInput, type EvalsJsonCase, type EvalsJsonFile, type EvaluationCache, type EvaluationContext, type EvaluationResult, type EvaluationScore, type EvaluationVerdict, type ExecInContainerOptions, type ExecResult, type ExecutionDefaults, type ExecutionError, type ExecutionMetrics, ExecutionMetricsGrader, type ExecutionMetricsGraderConfig, type ExecutionMetricsGraderOptions, type ExecutionStatus, type FailOnError, type FailureStage, FieldAccuracyGrader, type FieldAccuracyGraderConfig, type FieldAccuracyGraderOptions, type FieldAggregationType, type FieldConfig, type FieldMatchType, type GeminiResolvedConfig, type GenerateRubricsOptions, type Grader, type GraderConfig, type GraderDispatchContext, type GraderFactory, type GraderFactoryFn, type GraderKind, GraderRegistry, type GraderResult, type IcontainsAllGraderConfig, type IcontainsAnyGraderConfig, type IcontainsGraderConfig, type InlineAssertEvaluatorConfig, type IsJsonGraderConfig, type JsonObject, type JsonPrimitive, type JsonValue, LatencyGrader, type LatencyGraderConfig, type LatencyGraderOptions, LlmGrader, type LlmGraderConfig, type LlmGraderOptions, type LlmGraderPromptAssembly, type LocalPathValidationError, type MeanAggregation, type Message, type MockResolvedConfig, OTEL_BACKEND_PRESETS, type OpenAIResolvedConfig, type OpenRouterResolvedConfig, type OtelBackendPreset, type OtelExportOptions, OtelStreamingObserver, OtelTraceExporter, OtlpJsonFileExporter, type OutputMessage, PASS_THRESHOLD, type ParsedCopilotSession, type PassAtKAggregation, type PiCliResolvedConfig, type PiCodingAgentResolvedConfig, type PoolSlot, type PreparedResultsRepoBranch, type ProgressEvent, type PromptInputs, type PromptScriptConfig, type Provider, type ProviderFactoryFn, type ProviderKind, ProviderRegistry, type ProviderRequest, type ProviderResponse, type ProviderStreamCallbacks, type ProviderTokenUsage, type RegexGraderConfig, type RepoCheckout, type RepoClone, type RepoConfig, type RepoDep, RepoManager, type RepoSource, type ResolvedTarget, type ResolvedWorkspaceTemplate, ResponseCache, type ResultsExportConfig, type ResultsRepoCachePaths, type ResultsRepoStatus, type RubricItem, type RubricsEvaluatorConfig, RunBudgetTracker, type RunEvalCaseOptions, type RunEvaluationOptions, type ScoreRange, type ScriptExecutionContext, SkillTriggerGrader, type SkillTriggerGraderConfig, type StartsWithGraderConfig, type SystemTestMessage, TEST_MESSAGE_ROLES, type TargetAccessConfig, type TargetDefinition, type TargetHooksConfig, TemplateNotDirectoryError, TemplateNotFoundError, type TestMessage, type TestMessageContent, type TestMessageRole, type TokenUsage, TokenUsageGrader, type TokenUsageGraderConfig, type TokenUsageGraderOptions, type ToolCall, type ToolTestMessage, type ToolTrajectoryExpectedItem, ToolTrajectoryGrader, type ToolTrajectoryGraderConfig, type ToolTrajectoryGraderOptions, type TraceComputeResult, type TraceSummary, type TranscriptEntry, type TranscriptJsonLine, TranscriptProvider, type TranscriptReplayEntry, type TranscriptSource, type TranspileResult, type TrialAggregation, type TrialResult, type TrialStrategy, type TrialsConfig, type TsEvalResult, type TurnFailurePolicy, type UserTestMessage, type VSCodeResolvedConfig, type WorkspaceConfig, WorkspaceCreationError, type WorkspaceHookConfig, type WorkspaceHooksConfig, WorkspacePoolManager, type WorkspaceScriptConfig, addBenchmark, assembleLlmGraderPrompt, avgToolDurationMs, buildDirectoryChain, buildOutputSchema, buildPromptInputs, buildRubricOutputSchema, buildScoreRangeOutputSchema, buildSearchRoots, calculateRubricScore, captureFileChanges, checkoutResultsRepoBranch, clampScore, cleanupEvalWorkspaces, cleanupWorkspace, commitAndPushResultsBranch, computeTraceSummary, computeWorkspaceFingerprint, consumeClaudeLogEntries, consumeCodexLogEntries, consumeCopilotCliLogEntries, consumeCopilotSdkLogEntries, consumePiLogEntries, createAgentKernel, createBuiltinProviderRegistry, createBuiltinRegistry, createDraftResultsPr, createProvider, createTempWorkspace, deepEqual, defineConfig, deriveBenchmarkId, deriveCategory, detectFormat, directorySizeBytes, discoverAssertions, discoverBenchmarks, discoverClaudeSessions, discoverCodexSessions, discoverCopilotSessions, discoverGraders, discoverProviders, ensureResultsRepoClone, ensureVSCodeSubagents, evaluate, executeScript, executeWorkspaceScript, explorationRatio, extractCacheConfig, extractFailOnError, extractImageBlocks, extractJsonBlob, extractLastAssistantContent, extractTargetFromSuite, extractTargetRefsFromSuite, extractTargetsFromSuite, extractTargetsFromTestCase, extractThreshold, extractTrialsConfig, extractWorkersFromSuite, fileExists, findGitRoot, formatToolCalls, freeformEvaluationSchema, generateRubrics, getAgentvConfigDir, getAgentvHome, getBenchmark, getBenchmarksRegistryPath, getOutputFilenames, getResultsRepoCachePaths, getResultsRepoStatus, getSubagentsRoot, getTextContent, getTraceStateRoot, getWorkspacePath, getWorkspacePoolRoot, getWorkspacesRoot, groupTranscriptJsonLines, initializeBaseline, isAgentSkillsFormat, isContent, isContentArray, isGraderKind, isJsonObject, isJsonValue, isNonEmptyString, isTestMessage, isTestMessageRole, listTargetNames, loadBenchmarkRegistry, loadConfig, loadEvalCaseById, loadEvalCases, loadEvalSuite, loadTestById, loadTestSuite, loadTests, loadTsConfig, loadTsEvalFile, mergeExecutionMetrics, negateScore, normalizeLineEndings, normalizeResultsExportConfig, parseAgentSkillsEvals, parseClaudeSession, parseCodexSession, parseCopilotEvents, parseJsonFromText, parseJsonSafe, prepareResultsRepoBranch, pushResultsRepoBranch, readJsonFile, readTargetDefinitions, readTestSuiteMetadata, readTextFile, readTranscriptFile, readTranscriptJsonl, removeBenchmark, resolveAndCreateProvider, resolveDelegatedTargetDefinition, resolveFileReference, resolveResultsRepoRunsDir, resolveResultsRepoUrl, resolveTargetDefinition, resolveWorkspaceTemplate, rubricEvaluationSchema, runContainsAllAssertion, runContainsAnyAssertion, runContainsAssertion, runEndsWithAssertion, runEqualsAssertion, runEvalCase, runEvaluation, runIcontainsAllAssertion, runIcontainsAnyAssertion, runIcontainsAssertion, runIsJsonAssertion, runRegexAssertion, runStartsWithAssertion, saveBenchmarkRegistry, scanRepoDeps, scoreToVerdict, shouldEnableCache, shouldSkipCacheForTemperature, stageResultsArtifacts, subscribeToClaudeLogEntries, subscribeToCodexLogEntries, subscribeToCopilotCliLogEntries, subscribeToCopilotSdkLogEntries, subscribeToPiLogEntries, substituteVariables, syncResultsRepo, toCamelCaseDeep, toSnakeCaseDeep, toTranscriptJsonLines, tokensPerTool, touchBenchmark, transpileEvalYaml, transpileEvalYamlFile, trimBaselineResult };
+export { type AcquireWorkspaceOptions, type AgentKernel, type AgentVConfig$1 as AgentVConfig, type AgentVResolvedConfig, type AgentVConfig as AgentVTsConfig, type AgentVConfig$1 as AgentVYamlConfig, type AnthropicResolvedConfig, type ApiFormat, type ArgsMatchMode, type AssertContext, type AssertEntry, type AssertFn, type AssertResult, type AssertionEntry, type AssertionResult, type AssistantTestMessage, type AzureResolvedConfig, type BenchmarkEntry, type BenchmarkRegistry, COMMON_TARGET_SETTINGS, type CacheConfig, type CheckedOutResultsRepoBranch, type ChildGraderResult, type ClaudeDiscoverOptions, type ClaudeResolvedConfig, type ClaudeSession, type CliResolvedConfig, CodeGrader, type CodeGraderConfig, type CodeGraderOptions, type CodexDiscoverOptions, type CodexSession, type CommandExecutor, type CompositeAggregatorConfig, CompositeGrader, type CompositeGraderConfig, type CompositeGraderOptions, type ConfidenceIntervalAggregation, type ContainsAllGraderConfig, type ContainsAnyGraderConfig, type ContainsGraderConfig, type Content, type ContentFile, type ContentImage, type ContentPreprocessorConfig, type ContentText, type ConversationAggregation, type ConversationMode, type ConversationTurn, type ConversationTurnInput, type CopilotCliResolvedConfig, type DiscoverOptions as CopilotDiscoverOptions, type CopilotLogResolvedConfig, type CopilotSdkResolvedConfig, type CopilotSession, type CopilotSessionMeta, CostGrader, type CostGraderConfig, type CostGraderOptions, type CreateContainerOptions, DEFAULT_CATEGORY, DEFAULT_EVAL_PATTERNS, DEFAULT_EXPLORATION_TOOLS, DEFAULT_GRADER_TEMPLATE, DEFAULT_THRESHOLD, type DependencyFailurePolicy, type DependencyResult, type DepsScanResult, DeterministicAssertionGrader, type DockerWorkspaceConfig, DockerWorkspaceProvider, type EndsWithGraderConfig, type EnsureSubagentsOptions, type EnsureSubagentsResult, type EnvLookup, type EqualsGraderConfig, type EvalAssertionInput, type EvalCase, type EvalConfig, type EvalMetadata, type EvalRunResult, type EvalSuiteResult, type EvalSummary, type EvalTargetRef, type EvalTest, type EvalTestInput, type EvalsJsonCase, type EvalsJsonFile, type EvaluationCache, type EvaluationContext, type EvaluationResult, type EvaluationScore, type EvaluationVerdict, type ExecInContainerOptions, type ExecResult, type ExecutionDefaults, type ExecutionError, type ExecutionMetrics, ExecutionMetricsGrader, type ExecutionMetricsGraderConfig, type ExecutionMetricsGraderOptions, type ExecutionStatus, type FailOnError, type FailureStage, FieldAccuracyGrader, type FieldAccuracyGraderConfig, type FieldAccuracyGraderOptions, type FieldAggregationType, type FieldConfig, type FieldMatchType, type GeminiResolvedConfig, type GenerateRubricsOptions, type Grader, type GraderConfig, type GraderDispatchContext, type GraderFactory, type GraderFactoryFn, type GraderKind, GraderRegistry, type GraderResult, type IcontainsAllGraderConfig, type IcontainsAnyGraderConfig, type IcontainsGraderConfig, type InlineAssertEvaluatorConfig, type IsJsonGraderConfig, type JsonObject, type JsonPrimitive, type JsonValue, LatencyGrader, type LatencyGraderConfig, type LatencyGraderOptions, LlmGrader, type LlmGraderConfig, type LlmGraderOptions, type LlmGraderPromptAssembly, type LocalPathValidationError, type MeanAggregation, type Message, type MockResolvedConfig, OTEL_BACKEND_PRESETS, type OpenAIResolvedConfig, type OpenRouterResolvedConfig, type OtelBackendPreset, type OtelExportOptions, OtelStreamingObserver, OtelTraceExporter, OtlpJsonFileExporter, type OutputMessage, PASS_THRESHOLD, type ParsedCopilotSession, type PassAtKAggregation, type PiCliResolvedConfig, type PiCodingAgentResolvedConfig, type PoolSlot, type PreparedResultsRepoBranch, type ProgressEvent, type PromptInputs, type PromptScriptConfig, type Provider, type ProviderFactoryFn, type ProviderKind, ProviderRegistry, type ProviderRequest, type ProviderResponse, type ProviderStreamCallbacks, type ProviderTokenUsage, type RegexGraderConfig, type RepoCheckout, type RepoClone, type RepoConfig, type RepoDep, RepoManager, type RepoSource, type ResolvedTarget, type ResolvedWorkspaceTemplate, ResponseCache, type ResultsExportConfig, type ResultsRepoCachePaths, type ResultsRepoStatus, type RubricItem, type RubricsEvaluatorConfig, RunBudgetTracker, type RunEvalCaseOptions, type RunEvaluationOptions, type ScoreRange, type ScriptExecutionContext, SkillTriggerGrader, type SkillTriggerGraderConfig, type StartsWithGraderConfig, type SystemTestMessage, TEST_MESSAGE_ROLES, type TargetAccessConfig, type TargetDefinition, type TargetHooksConfig, TemplateNotDirectoryError, TemplateNotFoundError, type TestMessage, type TestMessageContent, type TestMessageRole, type TokenUsage, TokenUsageGrader, type TokenUsageGraderConfig, type TokenUsageGraderOptions, type ToolCall, type ToolTestMessage, type ToolTrajectoryExpectedItem, ToolTrajectoryGrader, type ToolTrajectoryGraderConfig, type ToolTrajectoryGraderOptions, type TraceComputeResult, type TraceSummary, type TranscriptEntry, type TranscriptJsonLine, TranscriptProvider, type TranscriptReplayEntry, type TranscriptSource, type TranspileResult, type TrialAggregation, type TrialResult, type TrialStrategy, type TrialsConfig, type TsEvalResult, type TurnFailurePolicy, type UserTestMessage, type VSCodeResolvedConfig, type WorkspaceConfig, WorkspaceCreationError, type WorkspaceHookConfig, type WorkspaceHooksConfig, WorkspacePoolManager, type WorkspaceScriptConfig, addBenchmark, assembleLlmGraderPrompt, avgToolDurationMs, buildDirectoryChain, buildOutputSchema, buildPromptInputs, buildRubricOutputSchema, buildScoreRangeOutputSchema, buildSearchRoots, calculateRubricScore, captureFileChanges, checkoutResultsRepoBranch, clampScore, cleanupEvalWorkspaces, cleanupWorkspace, commitAndPushResultsBranch, computeTraceSummary, computeWorkspaceFingerprint, consumeClaudeLogEntries, consumeCodexLogEntries, consumeCopilotCliLogEntries, consumeCopilotSdkLogEntries, consumePiLogEntries, createAgentKernel, createBuiltinProviderRegistry, createBuiltinRegistry, createDraftResultsPr, createProvider, createTempWorkspace, deepEqual, defineConfig, deriveBenchmarkId, deriveCategory, detectFormat, directorySizeBytes, discoverAssertions, discoverBenchmarks, discoverClaudeSessions, discoverCodexSessions, discoverCopilotSessions, discoverGraders, discoverProviders, ensureResultsRepoClone, ensureVSCodeSubagents, evaluate, executeScript, executeWorkspaceScript, explorationRatio, extractCacheConfig, extractFailOnError, extractImageBlocks, extractJsonBlob, extractLastAssistantContent, extractTargetFromSuite, extractTargetRefsFromSuite, extractTargetsFromSuite, extractTargetsFromTestCase, extractThreshold, extractTrialsConfig, extractWorkersFromSuite, fileExists, findGitRoot, formatToolCalls, freeformEvaluationSchema, generateRubrics, getAgentvConfigDir, getAgentvHome, getBenchmark, getBenchmarksRegistryPath, getOutputFilenames, getResultsRepoCachePaths, getResultsRepoStatus, getSubagentsRoot, getTextContent, getTraceStateRoot, getWorkspacePath, getWorkspacePoolRoot, getWorkspacesRoot, groupTranscriptJsonLines, initializeBaseline, isAgentSkillsFormat, isContent, isContentArray, isGraderKind, isJsonObject, isJsonValue, isNonEmptyString, isTestMessage, isTestMessageRole, listTargetNames, loadBenchmarkRegistry, loadConfig, loadEvalCaseById, loadEvalCases, loadEvalSuite, loadTestById, loadTestSuite, loadTests, loadTsConfig, loadTsEvalFile, mergeExecutionMetrics, negateScore, normalizeLineEndings, normalizeResultsExportConfig, parseAgentSkillsEvals, parseClaudeSession, parseCodexSession, parseCopilotEvents, parseEnvOutput, parseJsonFromText, parseJsonSafe, prepareResultsRepoBranch, pushResultsRepoBranch, readJsonFile, readTargetDefinitions, readTestSuiteMetadata, readTextFile, readTranscriptFile, readTranscriptJsonl, removeBenchmark, resolveAndCreateProvider, resolveDelegatedTargetDefinition, resolveFileReference, resolveResultsRepoRunsDir, resolveResultsRepoUrl, resolveTargetDefinition, resolveWorkspaceTemplate, rubricEvaluationSchema, runBeforeSessionHook, runContainsAllAssertion, runContainsAnyAssertion, runContainsAssertion, runEndsWithAssertion, runEqualsAssertion, runEvalCase, runEvaluation, runIcontainsAllAssertion, runIcontainsAnyAssertion, runIcontainsAssertion, runIsJsonAssertion, runRegexAssertion, runStartsWithAssertion, saveBenchmarkRegistry, scanRepoDeps, scoreRangeEvaluationSchema, scoreToVerdict, shouldEnableCache, shouldSkipCacheForTemperature, stageResultsArtifacts, subscribeToClaudeLogEntries, subscribeToCodexLogEntries, subscribeToCopilotCliLogEntries, subscribeToCopilotSdkLogEntries, subscribeToPiLogEntries, substituteVariables, syncResultsRepo, toCamelCaseDeep, toSnakeCaseDeep, toTranscriptJsonLines, tokensPerTool, touchBenchmark, transpileEvalYaml, transpileEvalYamlFile, trimBaselineResult };

package/dist/index.d.ts

@@ -1912,6 +1912,10 @@ type ResultsExportConfig = {
     readonly auto_push?: boolean;
     readonly branch_prefix?: string;
 };
+type HooksConfig = {
+    /** Shell command to run once at agentv startup. stdout is parsed for env var exports. */
+    readonly before_session?: string;
+};
 type AgentVConfig$1 = {
     readonly required_version?: string;
     readonly eval_patterns?: readonly string[];
@@ -1919,6 +1923,7 @@ type AgentVConfig$1 = {
     readonly results?: {
         readonly export?: ResultsExportConfig;
     };
+    readonly hooks?: HooksConfig;
 };
 /**
  * Load optional .agentv/config.yaml configuration file.
@@ -2937,6 +2942,36 @@ declare const rubricEvaluationSchema: z.ZodObject<{
     }[];
     overall_reasoning: string;
 }>;
+declare const scoreRangeEvaluationSchema: z.ZodObject<{
+    checks: z.ZodArray<z.ZodObject<{
+        id: z.ZodString;
+        score: z.ZodNumber;
+        reasoning: z.ZodOptional<z.ZodString>;
+    }, "strip", z.ZodTypeAny, {
+        id: string;
+        score: number;
+        reasoning?: string | undefined;
+    }, {
+        id: string;
+        score: number;
+        reasoning?: string | undefined;
+    }>, "many">;
+    overall_reasoning: z.ZodOptional<z.ZodString>;
+}, "strip", z.ZodTypeAny, {
+    checks: {
+        id: string;
+        score: number;
+        reasoning?: string | undefined;
+    }[];
+    overall_reasoning?: string | undefined;
+}, {
+    checks: {
+        id: string;
+        score: number;
+        reasoning?: string | undefined;
+    }[];
+    overall_reasoning?: string | undefined;
+}>;
 
 declare class LlmGrader implements Grader {
     readonly kind = "llm-grader";
@@ -3537,6 +3572,21 @@ declare const AgentVConfigSchema: z.ZodObject<{
         maxDurationMs?: number | undefined;
         maxCostUsd?: number | undefined;
     }>>;
+    /** Lifecycle hooks */
+    hooks: z.ZodOptional<z.ZodObject<{
+        /**
+         * Shell command to run once at agentv startup, before any command executes.
+         * stdout is parsed for env var exports (`KEY=value` or `export KEY="value"`)
+         * and injected into process.env. Keys already set in the environment are
+         * not overwritten — existing env always takes priority.
+         * stderr is forwarded to the user. Non-zero exit aborts with an error.
+         */
+        beforeSession: z.ZodOptional<z.ZodString>;
+    }, "strip", z.ZodTypeAny, {
+        beforeSession?: string | undefined;
+    }, {
+        beforeSession?: string | undefined;
+    }>>;
 }, "strip", z.ZodTypeAny, {
     execution?: {
         workers?: number | undefined;
@@ -3546,6 +3596,9 @@ declare const AgentVConfigSchema: z.ZodObject<{
         keepWorkspaces?: boolean | undefined;
         otelFile?: string | undefined;
     } | undefined;
+    hooks?: {
+        beforeSession?: string | undefined;
+    } | undefined;
     cache?: {
         enabled?: boolean | undefined;
         path?: string | undefined;
@@ -3567,6 +3620,9 @@ declare const AgentVConfigSchema: z.ZodObject<{
         keepWorkspaces?: boolean | undefined;
         otelFile?: string | undefined;
     } | undefined;
+    hooks?: {
+        beforeSession?: string | undefined;
+    } | undefined;
     cache?: {
         enabled?: boolean | undefined;
         path?: string | undefined;
@@ -4346,6 +4402,56 @@ declare function discoverAssertions(registry: GraderRegistry, baseDir: string):
  */
 declare function discoverGraders(registry: GraderRegistry, baseDir: string): Promise<string[]>;
 
+/**
+ * Session hook execution for AgentV.
+ *
+ * Runs a shell command once at agentv startup and injects exported environment
+ * variables into the current process. This lets projects fetch secrets at
+ * runtime (e.g. from a vault) without needing a wrapper script.
+ *
+ * ## How it works
+ *
+ * 1. The command is run via `sh -c` (or `cmd /c` on Windows).
+ * 2. stdout is captured and parsed for env var exports.
+ * 3. stderr is forwarded to the process stderr so the user sees output.
+ * 4. Non-zero exit aborts with a clear error.
+ * 5. Parsed keys are injected into `process.env` — only for keys not already
+ *    set, so existing env always wins.
+ *
+ * ## Supported output formats
+ *
+ * Both shell-export and dotenv formats are accepted:
+ *   export KEY="value"   (shell export — quotes optional)
+ *   KEY=value            (dotenv — no export prefix)
+ *
+ * Lines that don't match either pattern are silently ignored.
+ *
+ * @module
+ */
+/**
+ * Parse env var lines from hook stdout.
+ *
+ * Accepts:
+ *   export KEY="value"   → { KEY: "value" }
+ *   export KEY=value     → { KEY: "value" }
+ *   KEY=value            → { KEY: "value" }
+ *
+ * Strips surrounding single or double quotes from values.
+ * Skips lines with empty keys or values that look like shell syntax.
+ */
+declare function parseEnvOutput(stdout: string): Record<string, string>;
+/**
+ * Run the before_session hook command and inject exported env vars into process.env.
+ *
+ * - Runs via shell (`sh -c` on POSIX, `cmd /c` on Windows)
+ * - Captured stdout is parsed for env vars; stderr is forwarded to process.stderr
+ * - Non-zero exit throws an Error with the command and exit code
+ * - Keys already set in process.env are NOT overwritten
+ *
+ * @param command Shell command string to execute
+ */
+declare function runBeforeSessionHook(command: string): void;
+
 /**
  * Core types for the transcript import pipeline.
  *
@@ -4663,4 +4769,4 @@ type AgentKernel = {
 };
 declare function createAgentKernel(): AgentKernel;
 
-export { type AcquireWorkspaceOptions, type AgentKernel, type AgentVConfig$1 as AgentVConfig, type AgentVResolvedConfig, type AgentVConfig as AgentVTsConfig, type AgentVConfig$1 as AgentVYamlConfig, type AnthropicResolvedConfig, type ApiFormat, type ArgsMatchMode, type AssertContext, type AssertEntry, type AssertFn, type AssertResult, type AssertionEntry, type AssertionResult, type AssistantTestMessage, type AzureResolvedConfig, type BenchmarkEntry, type BenchmarkRegistry, COMMON_TARGET_SETTINGS, type CacheConfig, type CheckedOutResultsRepoBranch, type ChildGraderResult, type ClaudeDiscoverOptions, type ClaudeResolvedConfig, type ClaudeSession, type CliResolvedConfig, CodeGrader, type CodeGraderConfig, type CodeGraderOptions, type CodexDiscoverOptions, type CodexSession, type CommandExecutor, type CompositeAggregatorConfig, CompositeGrader, type CompositeGraderConfig, type CompositeGraderOptions, type ConfidenceIntervalAggregation, type ContainsAllGraderConfig, type ContainsAnyGraderConfig, type ContainsGraderConfig, type Content, type ContentFile, type ContentImage, type ContentPreprocessorConfig, type ContentText, type ConversationAggregation, type ConversationMode, type ConversationTurn, type ConversationTurnInput, type CopilotCliResolvedConfig, type DiscoverOptions as CopilotDiscoverOptions, type CopilotLogResolvedConfig, type CopilotSdkResolvedConfig, type CopilotSession, type CopilotSessionMeta, CostGrader, type CostGraderConfig, type CostGraderOptions, type CreateContainerOptions, DEFAULT_CATEGORY, DEFAULT_EVAL_PATTERNS, DEFAULT_EXPLORATION_TOOLS, DEFAULT_GRADER_TEMPLATE, DEFAULT_THRESHOLD, type DependencyFailurePolicy, type DependencyResult, type DepsScanResult, DeterministicAssertionGrader, type DockerWorkspaceConfig, DockerWorkspaceProvider, type EndsWithGraderConfig, type EnsureSubagentsOptions, type EnsureSubagentsResult, type EnvLookup, type EqualsGraderConfig, type EvalAssertionInput, type EvalCase, type EvalConfig, type EvalMetadata, type EvalRunResult, type EvalSuiteResult, type EvalSummary, type EvalTargetRef, type EvalTest, type EvalTestInput, type EvalsJsonCase, type EvalsJsonFile, type EvaluationCache, type EvaluationContext, type EvaluationResult, type EvaluationScore, type EvaluationVerdict, type ExecInContainerOptions, type ExecResult, type ExecutionDefaults, type ExecutionError, type ExecutionMetrics, ExecutionMetricsGrader, type ExecutionMetricsGraderConfig, type ExecutionMetricsGraderOptions, type ExecutionStatus, type FailOnError, type FailureStage, FieldAccuracyGrader, type FieldAccuracyGraderConfig, type FieldAccuracyGraderOptions, type FieldAggregationType, type FieldConfig, type FieldMatchType, type GeminiResolvedConfig, type GenerateRubricsOptions, type Grader, type GraderConfig, type GraderDispatchContext, type GraderFactory, type GraderFactoryFn, type GraderKind, GraderRegistry, type GraderResult, type IcontainsAllGraderConfig, type IcontainsAnyGraderConfig, type IcontainsGraderConfig, type InlineAssertEvaluatorConfig, type IsJsonGraderConfig, type JsonObject, type JsonPrimitive, type JsonValue, LatencyGrader, type LatencyGraderConfig, type LatencyGraderOptions, LlmGrader, type LlmGraderConfig, type LlmGraderOptions, type LlmGraderPromptAssembly, type LocalPathValidationError, type MeanAggregation, type Message, type MockResolvedConfig, OTEL_BACKEND_PRESETS, type OpenAIResolvedConfig, type OpenRouterResolvedConfig, type OtelBackendPreset, type OtelExportOptions, OtelStreamingObserver, OtelTraceExporter, OtlpJsonFileExporter, type OutputMessage, PASS_THRESHOLD, type ParsedCopilotSession, type PassAtKAggregation, type PiCliResolvedConfig, type PiCodingAgentResolvedConfig, type PoolSlot, type PreparedResultsRepoBranch, type ProgressEvent, type PromptInputs, type PromptScriptConfig, type Provider, type ProviderFactoryFn, type ProviderKind, ProviderRegistry, type ProviderRequest, type ProviderResponse, type ProviderStreamCallbacks, type ProviderTokenUsage, type RegexGraderConfig, type RepoCheckout, type RepoClone, type RepoConfig, type RepoDep, RepoManager, type RepoSource, type ResolvedTarget, type ResolvedWorkspaceTemplate, ResponseCache, type ResultsExportConfig, type ResultsRepoCachePaths, type ResultsRepoStatus, type RubricItem, type RubricsEvaluatorConfig, RunBudgetTracker, type RunEvalCaseOptions, type RunEvaluationOptions, type ScoreRange, type ScriptExecutionContext, SkillTriggerGrader, type SkillTriggerGraderConfig, type StartsWithGraderConfig, type SystemTestMessage, TEST_MESSAGE_ROLES, type TargetAccessConfig, type TargetDefinition, type TargetHooksConfig, TemplateNotDirectoryError, TemplateNotFoundError, type TestMessage, type TestMessageContent, type TestMessageRole, type TokenUsage, TokenUsageGrader, type TokenUsageGraderConfig, type TokenUsageGraderOptions, type ToolCall, type ToolTestMessage, type ToolTrajectoryExpectedItem, ToolTrajectoryGrader, type ToolTrajectoryGraderConfig, type ToolTrajectoryGraderOptions, type TraceComputeResult, type TraceSummary, type TranscriptEntry, type TranscriptJsonLine, TranscriptProvider, type TranscriptReplayEntry, type TranscriptSource, type TranspileResult, type TrialAggregation, type TrialResult, type TrialStrategy, type TrialsConfig, type TsEvalResult, type TurnFailurePolicy, type UserTestMessage, type VSCodeResolvedConfig, type WorkspaceConfig, WorkspaceCreationError, type WorkspaceHookConfig, type WorkspaceHooksConfig, WorkspacePoolManager, type WorkspaceScriptConfig, addBenchmark, assembleLlmGraderPrompt, avgToolDurationMs, buildDirectoryChain, buildOutputSchema, buildPromptInputs, buildRubricOutputSchema, buildScoreRangeOutputSchema, buildSearchRoots, calculateRubricScore, captureFileChanges, checkoutResultsRepoBranch, clampScore, cleanupEvalWorkspaces, cleanupWorkspace, commitAndPushResultsBranch, computeTraceSummary, computeWorkspaceFingerprint, consumeClaudeLogEntries, consumeCodexLogEntries, consumeCopilotCliLogEntries, consumeCopilotSdkLogEntries, consumePiLogEntries, createAgentKernel, createBuiltinProviderRegistry, createBuiltinRegistry, createDraftResultsPr, createProvider, createTempWorkspace, deepEqual, defineConfig, deriveBenchmarkId, deriveCategory, detectFormat, directorySizeBytes, discoverAssertions, discoverBenchmarks, discoverClaudeSessions, discoverCodexSessions, discoverCopilotSessions, discoverGraders, discoverProviders, ensureResultsRepoClone, ensureVSCodeSubagents, evaluate, executeScript, executeWorkspaceScript, explorationRatio, extractCacheConfig, extractFailOnError, extractImageBlocks, extractJsonBlob, extractLastAssistantContent, extractTargetFromSuite, extractTargetRefsFromSuite, extractTargetsFromSuite, extractTargetsFromTestCase, extractThreshold, extractTrialsConfig, extractWorkersFromSuite, fileExists, findGitRoot, formatToolCalls, freeformEvaluationSchema, generateRubrics, getAgentvConfigDir, getAgentvHome, getBenchmark, getBenchmarksRegistryPath, getOutputFilenames, getResultsRepoCachePaths, getResultsRepoStatus, getSubagentsRoot, getTextContent, getTraceStateRoot, getWorkspacePath, getWorkspacePoolRoot, getWorkspacesRoot, groupTranscriptJsonLines, initializeBaseline, isAgentSkillsFormat, isContent, isContentArray, isGraderKind, isJsonObject, isJsonValue, isNonEmptyString, isTestMessage, isTestMessageRole, listTargetNames, loadBenchmarkRegistry, loadConfig, loadEvalCaseById, loadEvalCases, loadEvalSuite, loadTestById, loadTestSuite, loadTests, loadTsConfig, loadTsEvalFile, mergeExecutionMetrics, negateScore, normalizeLineEndings, normalizeResultsExportConfig, parseAgentSkillsEvals, parseClaudeSession, parseCodexSession, parseCopilotEvents, parseJsonFromText, parseJsonSafe, prepareResultsRepoBranch, pushResultsRepoBranch, readJsonFile, readTargetDefinitions, readTestSuiteMetadata, readTextFile, readTranscriptFile, readTranscriptJsonl, removeBenchmark, resolveAndCreateProvider, resolveDelegatedTargetDefinition, resolveFileReference, resolveResultsRepoRunsDir, resolveResultsRepoUrl, resolveTargetDefinition, resolveWorkspaceTemplate, rubricEvaluationSchema, runContainsAllAssertion, runContainsAnyAssertion, runContainsAssertion, runEndsWithAssertion, runEqualsAssertion, runEvalCase, runEvaluation, runIcontainsAllAssertion, runIcontainsAnyAssertion, runIcontainsAssertion, runIsJsonAssertion, runRegexAssertion, runStartsWithAssertion, saveBenchmarkRegistry, scanRepoDeps, scoreToVerdict, shouldEnableCache, shouldSkipCacheForTemperature, stageResultsArtifacts, subscribeToClaudeLogEntries, subscribeToCodexLogEntries, subscribeToCopilotCliLogEntries, subscribeToCopilotSdkLogEntries, subscribeToPiLogEntries, substituteVariables, syncResultsRepo, toCamelCaseDeep, toSnakeCaseDeep, toTranscriptJsonLines, tokensPerTool, touchBenchmark, transpileEvalYaml, transpileEvalYamlFile, trimBaselineResult };
+export { type AcquireWorkspaceOptions, type AgentKernel, type AgentVConfig$1 as AgentVConfig, type AgentVResolvedConfig, type AgentVConfig as AgentVTsConfig, type AgentVConfig$1 as AgentVYamlConfig, type AnthropicResolvedConfig, type ApiFormat, type ArgsMatchMode, type AssertContext, type AssertEntry, type AssertFn, type AssertResult, type AssertionEntry, type AssertionResult, type AssistantTestMessage, type AzureResolvedConfig, type BenchmarkEntry, type BenchmarkRegistry, COMMON_TARGET_SETTINGS, type CacheConfig, type CheckedOutResultsRepoBranch, type ChildGraderResult, type ClaudeDiscoverOptions, type ClaudeResolvedConfig, type ClaudeSession, type CliResolvedConfig, CodeGrader, type CodeGraderConfig, type CodeGraderOptions, type CodexDiscoverOptions, type CodexSession, type CommandExecutor, type CompositeAggregatorConfig, CompositeGrader, type CompositeGraderConfig, type CompositeGraderOptions, type ConfidenceIntervalAggregation, type ContainsAllGraderConfig, type ContainsAnyGraderConfig, type ContainsGraderConfig, type Content, type ContentFile, type ContentImage, type ContentPreprocessorConfig, type ContentText, type ConversationAggregation, type ConversationMode, type ConversationTurn, type ConversationTurnInput, type CopilotCliResolvedConfig, type DiscoverOptions as CopilotDiscoverOptions, type CopilotLogResolvedConfig, type CopilotSdkResolvedConfig, type CopilotSession, type CopilotSessionMeta, CostGrader, type CostGraderConfig, type CostGraderOptions, type CreateContainerOptions, DEFAULT_CATEGORY, DEFAULT_EVAL_PATTERNS, DEFAULT_EXPLORATION_TOOLS, DEFAULT_GRADER_TEMPLATE, DEFAULT_THRESHOLD, type DependencyFailurePolicy, type DependencyResult, type DepsScanResult, DeterministicAssertionGrader, type DockerWorkspaceConfig, DockerWorkspaceProvider, type EndsWithGraderConfig, type EnsureSubagentsOptions, type EnsureSubagentsResult, type EnvLookup, type EqualsGraderConfig, type EvalAssertionInput, type EvalCase, type EvalConfig, type EvalMetadata, type EvalRunResult, type EvalSuiteResult, type EvalSummary, type EvalTargetRef, type EvalTest, type EvalTestInput, type EvalsJsonCase, type EvalsJsonFile, type EvaluationCache, type EvaluationContext, type EvaluationResult, type EvaluationScore, type EvaluationVerdict, type ExecInContainerOptions, type ExecResult, type ExecutionDefaults, type ExecutionError, type ExecutionMetrics, ExecutionMetricsGrader, type ExecutionMetricsGraderConfig, type ExecutionMetricsGraderOptions, type ExecutionStatus, type FailOnError, type FailureStage, FieldAccuracyGrader, type FieldAccuracyGraderConfig, type FieldAccuracyGraderOptions, type FieldAggregationType, type FieldConfig, type FieldMatchType, type GeminiResolvedConfig, type GenerateRubricsOptions, type Grader, type GraderConfig, type GraderDispatchContext, type GraderFactory, type GraderFactoryFn, type GraderKind, GraderRegistry, type GraderResult, type IcontainsAllGraderConfig, type IcontainsAnyGraderConfig, type IcontainsGraderConfig, type InlineAssertEvaluatorConfig, type IsJsonGraderConfig, type JsonObject, type JsonPrimitive, type JsonValue, LatencyGrader, type LatencyGraderConfig, type LatencyGraderOptions, LlmGrader, type LlmGraderConfig, type LlmGraderOptions, type LlmGraderPromptAssembly, type LocalPathValidationError, type MeanAggregation, type Message, type MockResolvedConfig, OTEL_BACKEND_PRESETS, type OpenAIResolvedConfig, type OpenRouterResolvedConfig, type OtelBackendPreset, type OtelExportOptions, OtelStreamingObserver, OtelTraceExporter, OtlpJsonFileExporter, type OutputMessage, PASS_THRESHOLD, type ParsedCopilotSession, type PassAtKAggregation, type PiCliResolvedConfig, type PiCodingAgentResolvedConfig, type PoolSlot, type PreparedResultsRepoBranch, type ProgressEvent, type PromptInputs, type PromptScriptConfig, type Provider, type ProviderFactoryFn, type ProviderKind, ProviderRegistry, type ProviderRequest, type ProviderResponse, type ProviderStreamCallbacks, type ProviderTokenUsage, type RegexGraderConfig, type RepoCheckout, type RepoClone, type RepoConfig, type RepoDep, RepoManager, type RepoSource, type ResolvedTarget, type ResolvedWorkspaceTemplate, ResponseCache, type ResultsExportConfig, type ResultsRepoCachePaths, type ResultsRepoStatus, type RubricItem, type RubricsEvaluatorConfig, RunBudgetTracker, type RunEvalCaseOptions, type RunEvaluationOptions, type ScoreRange, type ScriptExecutionContext, SkillTriggerGrader, type SkillTriggerGraderConfig, type StartsWithGraderConfig, type SystemTestMessage, TEST_MESSAGE_ROLES, type TargetAccessConfig, type TargetDefinition, type TargetHooksConfig, TemplateNotDirectoryError, TemplateNotFoundError, type TestMessage, type TestMessageContent, type TestMessageRole, type TokenUsage, TokenUsageGrader, type TokenUsageGraderConfig, type TokenUsageGraderOptions, type ToolCall, type ToolTestMessage, type ToolTrajectoryExpectedItem, ToolTrajectoryGrader, type ToolTrajectoryGraderConfig, type ToolTrajectoryGraderOptions, type TraceComputeResult, type TraceSummary, type TranscriptEntry, type TranscriptJsonLine, TranscriptProvider, type TranscriptReplayEntry, type TranscriptSource, type TranspileResult, type TrialAggregation, type TrialResult, type TrialStrategy, type TrialsConfig, type TsEvalResult, type TurnFailurePolicy, type UserTestMessage, type VSCodeResolvedConfig, type WorkspaceConfig, WorkspaceCreationError, type WorkspaceHookConfig, type WorkspaceHooksConfig, WorkspacePoolManager, type WorkspaceScriptConfig, addBenchmark, assembleLlmGraderPrompt, avgToolDurationMs, buildDirectoryChain, buildOutputSchema, buildPromptInputs, buildRubricOutputSchema, buildScoreRangeOutputSchema, buildSearchRoots, calculateRubricScore, captureFileChanges, checkoutResultsRepoBranch, clampScore, cleanupEvalWorkspaces, cleanupWorkspace, commitAndPushResultsBranch, computeTraceSummary, computeWorkspaceFingerprint, consumeClaudeLogEntries, consumeCodexLogEntries, consumeCopilotCliLogEntries, consumeCopilotSdkLogEntries, consumePiLogEntries, createAgentKernel, createBuiltinProviderRegistry, createBuiltinRegistry, createDraftResultsPr, createProvider, createTempWorkspace, deepEqual, defineConfig, deriveBenchmarkId, deriveCategory, detectFormat, directorySizeBytes, discoverAssertions, discoverBenchmarks, discoverClaudeSessions, discoverCodexSessions, discoverCopilotSessions, discoverGraders, discoverProviders, ensureResultsRepoClone, ensureVSCodeSubagents, evaluate, executeScript, executeWorkspaceScript, explorationRatio, extractCacheConfig, extractFailOnError, extractImageBlocks, extractJsonBlob, extractLastAssistantContent, extractTargetFromSuite, extractTargetRefsFromSuite, extractTargetsFromSuite, extractTargetsFromTestCase, extractThreshold, extractTrialsConfig, extractWorkersFromSuite, fileExists, findGitRoot, formatToolCalls, freeformEvaluationSchema, generateRubrics, getAgentvConfigDir, getAgentvHome, getBenchmark, getBenchmarksRegistryPath, getOutputFilenames, getResultsRepoCachePaths, getResultsRepoStatus, getSubagentsRoot, getTextContent, getTraceStateRoot, getWorkspacePath, getWorkspacePoolRoot, getWorkspacesRoot, groupTranscriptJsonLines, initializeBaseline, isAgentSkillsFormat, isContent, isContentArray, isGraderKind, isJsonObject, isJsonValue, isNonEmptyString, isTestMessage, isTestMessageRole, listTargetNames, loadBenchmarkRegistry, loadConfig, loadEvalCaseById, loadEvalCases, loadEvalSuite, loadTestById, loadTestSuite, loadTests, loadTsConfig, loadTsEvalFile, mergeExecutionMetrics, negateScore, normalizeLineEndings, normalizeResultsExportConfig, parseAgentSkillsEvals, parseClaudeSession, parseCodexSession, parseCopilotEvents, parseEnvOutput, parseJsonFromText, parseJsonSafe, prepareResultsRepoBranch, pushResultsRepoBranch, readJsonFile, readTargetDefinitions, readTestSuiteMetadata, readTextFile, readTranscriptFile, readTranscriptJsonl, removeBenchmark, resolveAndCreateProvider, resolveDelegatedTargetDefinition, resolveFileReference, resolveResultsRepoRunsDir, resolveResultsRepoUrl, resolveTargetDefinition, resolveWorkspaceTemplate, rubricEvaluationSchema, runBeforeSessionHook, runContainsAllAssertion, runContainsAnyAssertion, runContainsAssertion, runEndsWithAssertion, runEqualsAssertion, runEvalCase, runEvaluation, runIcontainsAllAssertion, runIcontainsAnyAssertion, runIcontainsAssertion, runIsJsonAssertion, runRegexAssertion, runStartsWithAssertion, saveBenchmarkRegistry, scanRepoDeps, scoreRangeEvaluationSchema, scoreToVerdict, shouldEnableCache, shouldSkipCacheForTemperature, stageResultsArtifacts, subscribeToClaudeLogEntries, subscribeToCodexLogEntries, subscribeToCopilotCliLogEntries, subscribeToCopilotSdkLogEntries, subscribeToPiLogEntries, substituteVariables, syncResultsRepo, toCamelCaseDeep, toSnakeCaseDeep, toTranscriptJsonLines, tokensPerTool, touchBenchmark, transpileEvalYaml, transpileEvalYamlFile, trimBaselineResult };

package/dist/index.js

@@ -118,6 +118,7 @@ import {
   runIsJsonAssertion,
   runRegexAssertion,
   runStartsWithAssertion,
+  scoreRangeEvaluationSchema,
   scoreToVerdict,
   subscribeToClaudeLogEntries,
   subscribeToCodexLogEntries,
@@ -128,7 +129,7 @@ import {
   toCamelCaseDeep,
   toSnakeCaseDeep,
   tokensPerTool
-} from "./chunk-WCW3V6QJ.js";
+} from "./chunk-B3BLJRYI.js";
 import {
   COMMON_TARGET_SETTINGS,
   TEST_MESSAGE_ROLES,
@@ -457,6 +458,17 @@ var AgentVConfigSchema = z.object({
     maxCostUsd: z.number().min(0).optional(),
     /** Maximum duration per run in milliseconds */
     maxDurationMs: z.number().int().min(0).optional()
+  }).optional(),
+  /** Lifecycle hooks */
+  hooks: z.object({
+    /**
+     * Shell command to run once at agentv startup, before any command executes.
+     * stdout is parsed for env var exports (`KEY=value` or `export KEY="value"`)
+     * and injected into process.env. Keys already set in the environment are
+     * not overwritten — existing env always takes priority.
+     * stderr is forwarded to the user. Non-zero exit aborts with an error.
+     */
+    beforeSession: z.string().optional()
   }).optional()
 });
 function defineConfig(config) {
@@ -1709,6 +1721,62 @@ var RunBudgetTracker = class {
   }
 };
 
+// src/evaluation/hooks.ts
+import { spawnSync } from "node:child_process";
+var ANSI_YELLOW = "\x1B[33m";
+var ANSI_RESET = "\x1B[0m";
+function parseEnvOutput(stdout) {
+  const result = {};
+  for (const line of stdout.split("\n")) {
+    const trimmed = line.trim();
+    if (!trimmed || trimmed.startsWith("#")) continue;
+    const match = trimmed.match(/^(?:export\s+)?([A-Za-z_][A-Za-z0-9_]*)=(.*)$/);
+    if (!match) continue;
+    const key = match[1];
+    let value = match[2];
+    if (value.startsWith('"') && value.endsWith('"') || value.startsWith("'") && value.endsWith("'")) {
+      value = value.slice(1, -1);
+    }
+    if (key) {
+      result[key] = value;
+    }
+  }
+  return result;
+}
+function runBeforeSessionHook(command) {
+  const isWindows = process.platform === "win32";
+  const shell = isWindows ? "cmd" : "sh";
+  const shellFlag = isWindows ? "/c" : "-c";
+  console.log(`${ANSI_YELLOW}Running before_session hook: ${command}${ANSI_RESET}`);
+  const result = spawnSync(shell, [shellFlag, command], {
+    encoding: "utf8",
+    // Do not inherit stdio — capture stdout for parsing, forward stderr manually
+    stdio: ["ignore", "pipe", "pipe"]
+  });
+  if (result.stderr) {
+    process.stderr.write(result.stderr);
+  }
+  if (result.error) {
+    throw new Error(`before_session hook failed to start: ${result.error.message}`);
+  }
+  if (result.status !== 0) {
+    throw new Error(
+      `before_session hook exited with code ${result.status ?? "unknown"}: ${command}`
+    );
+  }
+  const vars = parseEnvOutput(result.stdout ?? "");
+  let injected = 0;
+  for (const [key, value] of Object.entries(vars)) {
+    if (process.env[key] === void 0) {
+      process.env[key] = value;
+      injected++;
+    }
+  }
+  if (injected > 0) {
+    console.log(`before_session hook injected ${injected} environment variable(s).`);
+  }
+}
+
 // src/import/claude-parser.ts
 var SKIPPED_TYPES = /* @__PURE__ */ new Set(["progress", "system", "file-history-snapshot"]);
 function parseClaudeSession(jsonl) {
@@ -2476,6 +2544,7 @@ export {
   parseClaudeSession,
   parseCodexSession,
   parseCopilotEvents,
+  parseEnvOutput,
   parseJsonFromText,
   parseJsonSafe,
   prepareResultsRepoBranch,
@@ -2495,6 +2564,7 @@ export {
   resolveTargetDefinition,
   resolveWorkspaceTemplate,
   rubricEvaluationSchema,
+  runBeforeSessionHook,
   runContainsAllAssertion,
   runContainsAnyAssertion,
   runContainsAssertion,
@@ -2510,6 +2580,7 @@ export {
   runStartsWithAssertion,
   saveBenchmarkRegistry,
   scanRepoDeps,
+  scoreRangeEvaluationSchema,
   scoreToVerdict,
   shouldEnableCache,
   shouldSkipCacheForTemperature,