eve 0.13.8 → 0.15.0

package/dist/src/compiled/json-schema/index.js

@@ -0,0 +1 @@
+export {};

package/dist/src/compiler/manifest.js

@@ -1 +1 @@
-import{z}from"#compiled/zod/index.js";import{discoverDiagnosticsSummarySchema}from"#discover/diagnostics.js";import{compiledRemoteAgentNodeSchema}from"#compiler/remote-agent-node.js";import{jsonObjectSchema}from"#shared/json-schemas.js";const COMPILED_AGENT_MANIFEST_KIND=`eve-agent-compiled-manifest`,ROOT_COMPILED_AGENT_NODE_ID=`__root__`,COMPILED_AGENT_MANIFEST_VERSION=30,moduleSourceRefSchema=z.object({exportName:z.string().optional(),sourceKind:z.literal(`module`),logicalPath:z.string(),sourceId:z.string()}).strict(),channelMethodSchema=z.union([z.literal(`GET`),z.literal(`POST`),z.literal(`PUT`),z.literal(`PATCH`),z.literal(`DELETE`),z.literal(`WEBSOCKET`)]),compiledChannelDefinitionSchema=z.object({kind:z.literal(`channel`),name:z.string(),logicalPath:z.string(),method:channelMethodSchema,urlPath:z.string(),sourceId:z.string(),sourceKind:z.literal(`module`),exportName:z.string().optional(),adapterKind:z.string().optional()}).strict(),disabledCompiledChannelEntrySchema=z.object({kind:z.literal(`disabled`),name:z.string(),logicalPath:z.string()}).strict(),compiledChannelEntrySchema=z.union([compiledChannelDefinitionSchema,disabledCompiledChannelEntrySchema]),modelRoutingSchema=z.union([z.object({kind:z.literal(`gateway`),target:z.string(),byok:z.string().optional()}).strict(),z.object({kind:z.literal(`external`),provider:z.string()}).strict()]),compiledRuntimeModelReferenceSchema=z.object({contextWindowTokens:z.number().int().positive().optional(),id:z.string(),source:moduleSourceRefSchema.optional(),providerOptions:z.record(z.string(),jsonObjectSchema).optional(),routing:modelRoutingSchema}).strict(),compiledAgentBuildDefinitionSchema=z.object({externalDependencies:z.array(z.string()).optional()}).strict(),compiledAgentWorkflowWorldDefinitionSchema=z.string(),compiledAgentWorkflowDefinitionSchema=z.object({world:compiledAgentWorkflowWorldDefinitionSchema.optional()}).strict(),compiledAgentCompactionDefinitionSchema=z.object({model:compiledRuntimeModelReferenceSchema.optional(),thresholdPercent:z.number().finite().min(0).max(1).optional()}).strict(),compiledAgentConfigSchema=z.object({build:compiledAgentBuildDefinitionSchema.optional(),compaction:compiledAgentCompactionDefinitionSchema.optional(),description:z.string().optional(),experimental:z.object({codeMode:z.boolean().optional(),workflow:compiledAgentWorkflowDefinitionSchema.optional()}).strict().optional(),model:compiledRuntimeModelReferenceSchema,name:z.string(),outputSchema:jsonObjectSchema.optional(),source:moduleSourceRefSchema.optional()}).strict(),compiledInstructionsSchema=z.object({name:z.string(),logicalPath:z.string(),markdown:z.string(),sourceId:z.string(),sourceKind:z.union([z.literal(`markdown`),z.literal(`module`)])}).strict(),compiledSkillBaseFields={name:z.string(),description:z.string(),license:z.string().optional(),markdown:z.string(),metadata:z.record(z.string(),z.string()).optional(),sourceId:z.string(),logicalPath:z.string()},compiledSkillSourceSchema=z.discriminatedUnion(`sourceKind`,[z.object({...compiledSkillBaseFields,sourceKind:z.literal(`markdown`)}).strict(),z.object({...compiledSkillBaseFields,sourceKind:z.literal(`module`),exportName:z.string().optional()}).strict(),z.object({...compiledSkillBaseFields,sourceKind:z.literal(`skill-package`),skillId:z.string(),skillFilePath:z.string(),rootPath:z.string(),assetsPath:z.string().optional(),referencesPath:z.string().optional(),scriptsPath:z.string().optional()}).strict()]),compiledScheduleDefinitionSchema=z.object({cron:z.string(),hasRun:z.boolean(),name:z.string(),logicalPath:z.string(),markdown:z.string().optional(),sourceId:z.string(),sourceKind:z.union([z.literal(`markdown`),z.literal(`module`)])}).strict(),compiledSandboxDefinitionSchema=z.object({backendName:z.string().optional(),description:z.string().optional(),exportName:z.string().optional(),logicalPath:z.string(),revalidationKey:z.string().optional(),sourceHash:z.string(),sourceId:z.string(),sourceKind:z.literal(`module`)}).strict(),compiledSandboxWorkspaceSchema=z.object({logicalPath:z.string(),rootEntries:z.array(z.string()).readonly(),sourceId:z.string(),sourcePath:z.string()}).strict(),compiledWorkspaceResourceRootSchema=z.object({contentHash:z.string().optional(),logicalPath:z.string(),rootEntries:z.array(z.string()).readonly()}).strict(),compiledConnectionDefinitionSchema=z.object({connectionName:z.string(),description:z.string(),exportName:z.string().optional(),logicalPath:z.string(),protocol:z.enum([`mcp`,`openapi`]).default(`mcp`),sourceId:z.string(),sourceKind:z.literal(`module`),url:z.string(),vercelConnect:z.object({connector:z.string()}).strict().optional()}).strict(),compiledToolDefinitionSchema=z.object({description:z.string(),exportName:z.string().optional(),inputSchema:jsonObjectSchema.nullable(),logicalPath:z.string(),name:z.string(),outputSchema:jsonObjectSchema.optional(),sourceId:z.string(),sourceKind:z.literal(`module`)}).strict(),compiledDynamicToolDefinitionSchema=z.object({eventNames:z.array(z.string()).readonly(),exportName:z.string().optional(),logicalPath:z.string(),slug:z.string(),sourceId:z.string(),sourceKind:z.literal(`module`)}).strict(),compiledDynamicSkillDefinitionSchema=z.object({eventNames:z.array(z.string()).readonly(),exportName:z.string().optional(),logicalPath:z.string(),slug:z.string(),sourceId:z.string(),sourceKind:z.literal(`module`)}).strict(),compiledDynamicInstructionsDefinitionSchema=z.object({eventNames:z.array(z.string()).readonly(),exportName:z.string().optional(),logicalPath:z.string(),slug:z.string(),sourceId:z.string(),sourceKind:z.literal(`module`)}).strict(),compiledHookDefinitionSchema=z.object({exportName:z.string().optional(),logicalPath:z.string(),slug:z.string(),sourceId:z.string(),sourceKind:z.literal(`module`)}).strict(),compiledAgentNodeManifestSchema=z.object({agentRoot:z.string(),appRoot:z.string(),channels:z.array(compiledChannelEntrySchema),config:compiledAgentConfigSchema,connections:z.array(compiledConnectionDefinitionSchema),diagnosticsSummary:discoverDiagnosticsSummarySchema,disabledFrameworkTools:z.array(z.string()).readonly(),workflowEnabled:z.boolean().default(!1),dynamicInstructions:z.array(compiledDynamicInstructionsDefinitionSchema).default([]),dynamicSkills:z.array(compiledDynamicSkillDefinitionSchema).default([]),dynamicTools:z.array(compiledDynamicToolDefinitionSchema).default([]),hooks:z.array(compiledHookDefinitionSchema),sandbox:compiledSandboxDefinitionSchema.nullable(),sandboxWorkspaces:z.array(compiledSandboxWorkspaceSchema),schedules:z.array(compiledScheduleDefinitionSchema),remoteAgents:z.array(compiledRemoteAgentNodeSchema),skills:z.array(compiledSkillSourceSchema).readonly(),instructions:compiledInstructionsSchema.optional(),tools:z.array(compiledToolDefinitionSchema),workspaceResourceRoot:compiledWorkspaceResourceRootSchema}).strict(),compiledSubagentNodeSchema=z.object({agent:compiledAgentNodeManifestSchema,description:z.string(),entryPath:z.string(),logicalPath:z.string(),name:z.string(),nodeId:z.string(),rootPath:z.string(),sourceId:z.string(),sourceKind:z.literal(`module`),exportName:z.string().optional()}).strict(),compiledSubagentEdgeSchema=z.object({childNodeId:z.string(),parentNodeId:z.string()}).strict(),compiledAgentManifestSchema=z.object({agentRoot:z.string(),appRoot:z.string(),channels:z.array(compiledChannelEntrySchema),config:compiledAgentConfigSchema,connections:z.array(compiledConnectionDefinitionSchema),diagnosticsSummary:discoverDiagnosticsSummarySchema,disabledFrameworkTools:z.array(z.string()).readonly(),workflowEnabled:z.boolean().default(!1),dynamicInstructions:z.array(compiledDynamicInstructionsDefinitionSchema).default([]),dynamicSkills:z.array(compiledDynamicSkillDefinitionSchema).default([]),dynamicTools:z.array(compiledDynamicToolDefinitionSchema).default([]),hooks:z.array(compiledHookDefinitionSchema),kind:z.literal(COMPILED_AGENT_MANIFEST_KIND),remoteAgents:z.array(compiledRemoteAgentNodeSchema),sandbox:compiledSandboxDefinitionSchema.nullable(),sandboxWorkspaces:z.array(compiledSandboxWorkspaceSchema),schedules:z.array(compiledScheduleDefinitionSchema),skills:z.array(compiledSkillSourceSchema).readonly(),subagentEdges:z.array(compiledSubagentEdgeSchema),subagents:z.array(compiledSubagentNodeSchema),instructions:compiledInstructionsSchema.optional(),tools:z.array(compiledToolDefinitionSchema),version:z.literal(30),workspaceResourceRoot:compiledWorkspaceResourceRootSchema}).strict();function createCompiledAgentNodeManifest(e){let t={agentRoot:e.agentRoot,appRoot:e.appRoot,channels:[...e.channels??[]],connections:[...e.connections??[]],config:{build:e.config.build===void 0?void 0:{externalDependencies:e.config.build.externalDependencies===void 0?void 0:[...e.config.build.externalDependencies]},compaction:{model:e.config.compaction?.model===void 0?void 0:cloneCompiledRuntimeModelReference(e.config.compaction.model),thresholdPercent:e.config.compaction?.thresholdPercent},description:e.config.description,experimental:e.config.experimental===void 0?void 0:{codeMode:e.config.experimental.codeMode,workflow:e.config.experimental.workflow===void 0?void 0:{world:e.config.experimental.workflow.world}},model:cloneCompiledRuntimeModelReference(e.config.model),name:e.config.name,outputSchema:e.config.outputSchema,source:e.config.source===void 0?void 0:{...e.config.source}},diagnosticsSummary:e.diagnosticsSummary??{errors:0,warnings:0},disabledFrameworkTools:[...e.disabledFrameworkTools??[]],workflowEnabled:e.workflowEnabled??!1,dynamicInstructions:[...e.dynamicInstructions??[]],dynamicSkills:[...e.dynamicSkills??[]],dynamicTools:[...e.dynamicTools??[]],hooks:[...e.hooks??[]],remoteAgents:[...e.remoteAgents??[]],sandbox:e.sandbox??null,sandboxWorkspaces:[...e.sandboxWorkspaces??[]],schedules:[...e.schedules??[]],skills:[...e.skills??[]],tools:[...e.tools??[]],workspaceResourceRoot:e.workspaceResourceRoot??{logicalPath:``,rootEntries:deriveResourceRootEntries({sandboxWorkspaces:e.sandboxWorkspaces,skills:e.skills})}};return e.instructions!==void 0&&(t.instructions=e.instructions),t}function deriveResourceRootEntries(e){let t=new Set;(e.skills??[]).length>0&&t.add(`skills/`);for(let n of e.sandboxWorkspaces??[])for(let e of n.rootEntries)t.add(e);return[...t].sort((e,t)=>e.localeCompare(t))}function createCompiledSubagentNodeId(e,t){return e===`__root__`?t:`${e}::${t}`}function createCompiledAgentManifest(e){return{...createCompiledAgentNodeManifest(e),kind:COMPILED_AGENT_MANIFEST_KIND,subagentEdges:[...e.subagentEdges??[]],subagents:[...e.subagents??[]],version:30}}function cloneCompiledRuntimeModelReference(e){let t={id:e.id,routing:cloneModelRouting(e.routing)};return e.contextWindowTokens!==void 0&&(t.contextWindowTokens=e.contextWindowTokens),e.providerOptions!==void 0&&(t.providerOptions={...e.providerOptions}),e.source!==void 0&&(t.source={...e.source}),t}function cloneModelRouting(e){return e.kind===`external`?{kind:`external`,provider:e.provider}:e.byok===void 0?{kind:`gateway`,target:e.target}:{kind:`gateway`,target:e.target,byok:e.byok}}export{COMPILED_AGENT_MANIFEST_KIND,COMPILED_AGENT_MANIFEST_VERSION,ROOT_COMPILED_AGENT_NODE_ID,compiledAgentManifestSchema,createCompiledAgentManifest,createCompiledAgentNodeManifest,createCompiledSubagentNodeId,deriveResourceRootEntries};
+import{z}from"#compiled/zod/index.js";import{discoverDiagnosticsSummarySchema}from"#discover/diagnostics.js";import{compiledRemoteAgentNodeSchema}from"#compiler/remote-agent-node.js";import{jsonObjectSchema}from"#shared/json-schemas.js";const COMPILED_AGENT_MANIFEST_KIND=`eve-agent-compiled-manifest`,ROOT_COMPILED_AGENT_NODE_ID=`__root__`,COMPILED_AGENT_MANIFEST_VERSION=30,moduleSourceRefSchema=z.object({exportName:z.string().optional(),sourceKind:z.literal(`module`),logicalPath:z.string(),sourceId:z.string()}).strict(),channelMethodSchema=z.union([z.literal(`GET`),z.literal(`POST`),z.literal(`PUT`),z.literal(`PATCH`),z.literal(`DELETE`),z.literal(`WEBSOCKET`)]),compiledChannelDefinitionSchema=z.object({kind:z.literal(`channel`),name:z.string(),logicalPath:z.string(),method:channelMethodSchema,urlPath:z.string(),sourceId:z.string(),sourceKind:z.literal(`module`),exportName:z.string().optional(),adapterKind:z.string().optional()}).strict(),disabledCompiledChannelEntrySchema=z.object({kind:z.literal(`disabled`),name:z.string(),logicalPath:z.string()}).strict(),compiledChannelEntrySchema=z.union([compiledChannelDefinitionSchema,disabledCompiledChannelEntrySchema]),modelRoutingSchema=z.union([z.object({kind:z.literal(`gateway`),target:z.string(),byok:z.string().optional()}).strict(),z.object({kind:z.literal(`external`),provider:z.string()}).strict()]),compiledRuntimeModelReferenceSchema=z.object({contextWindowTokens:z.number().int().positive().optional(),id:z.string(),source:moduleSourceRefSchema.optional(),providerOptions:z.record(z.string(),jsonObjectSchema).optional(),routing:modelRoutingSchema}).strict(),compiledAgentBuildDefinitionSchema=z.object({externalDependencies:z.array(z.string()).optional()}).strict(),compiledAgentWorkflowWorldDefinitionSchema=z.string(),compiledAgentWorkflowDefinitionSchema=z.object({world:compiledAgentWorkflowWorldDefinitionSchema.optional()}).strict(),compiledAgentCompactionDefinitionSchema=z.object({model:compiledRuntimeModelReferenceSchema.optional(),thresholdPercent:z.number().finite().min(0).max(1).optional()}).strict(),compiledAgentConfigSchema=z.object({build:compiledAgentBuildDefinitionSchema.optional(),compaction:compiledAgentCompactionDefinitionSchema.optional(),description:z.string().optional(),experimental:z.object({workflow:compiledAgentWorkflowDefinitionSchema.optional()}).strict().optional(),model:compiledRuntimeModelReferenceSchema,name:z.string(),outputSchema:jsonObjectSchema.optional(),reasoning:z.enum([`provider-default`,`none`,`minimal`,`low`,`medium`,`high`,`xhigh`]).optional(),source:moduleSourceRefSchema.optional()}).strict(),compiledInstructionsSchema=z.object({name:z.string(),logicalPath:z.string(),markdown:z.string(),sourceId:z.string(),sourceKind:z.union([z.literal(`markdown`),z.literal(`module`)])}).strict(),compiledSkillBaseFields={name:z.string(),description:z.string(),license:z.string().optional(),markdown:z.string(),metadata:z.record(z.string(),z.string()).optional(),sourceId:z.string(),logicalPath:z.string()},compiledSkillSourceSchema=z.discriminatedUnion(`sourceKind`,[z.object({...compiledSkillBaseFields,sourceKind:z.literal(`markdown`)}).strict(),z.object({...compiledSkillBaseFields,sourceKind:z.literal(`module`),exportName:z.string().optional()}).strict(),z.object({...compiledSkillBaseFields,sourceKind:z.literal(`skill-package`),skillId:z.string(),skillFilePath:z.string(),rootPath:z.string(),assetsPath:z.string().optional(),referencesPath:z.string().optional(),scriptsPath:z.string().optional()}).strict()]),compiledScheduleDefinitionSchema=z.object({cron:z.string(),hasRun:z.boolean(),name:z.string(),logicalPath:z.string(),markdown:z.string().optional(),sourceId:z.string(),sourceKind:z.union([z.literal(`markdown`),z.literal(`module`)])}).strict(),compiledSandboxDefinitionSchema=z.object({backendName:z.string().optional(),description:z.string().optional(),exportName:z.string().optional(),logicalPath:z.string(),revalidationKey:z.string().optional(),sourceHash:z.string(),sourceId:z.string(),sourceKind:z.literal(`module`)}).strict(),compiledSandboxWorkspaceSchema=z.object({logicalPath:z.string(),rootEntries:z.array(z.string()).readonly(),sourceId:z.string(),sourcePath:z.string()}).strict(),compiledWorkspaceResourceRootSchema=z.object({contentHash:z.string().optional(),logicalPath:z.string(),rootEntries:z.array(z.string()).readonly()}).strict(),compiledConnectionDefinitionSchema=z.object({connectionName:z.string(),description:z.string(),exportName:z.string().optional(),logicalPath:z.string(),protocol:z.enum([`mcp`,`openapi`]).default(`mcp`),sourceId:z.string(),sourceKind:z.literal(`module`),url:z.string(),vercelConnect:z.object({connector:z.string()}).strict().optional()}).strict(),compiledToolDefinitionSchema=z.object({description:z.string(),exportName:z.string().optional(),inputSchema:jsonObjectSchema.nullable(),logicalPath:z.string(),name:z.string(),outputSchema:jsonObjectSchema.optional(),sourceId:z.string(),sourceKind:z.literal(`module`)}).strict(),compiledDynamicToolDefinitionSchema=z.object({eventNames:z.array(z.string()).readonly(),exportName:z.string().optional(),logicalPath:z.string(),slug:z.string(),sourceId:z.string(),sourceKind:z.literal(`module`)}).strict(),compiledDynamicSkillDefinitionSchema=z.object({eventNames:z.array(z.string()).readonly(),exportName:z.string().optional(),logicalPath:z.string(),slug:z.string(),sourceId:z.string(),sourceKind:z.literal(`module`)}).strict(),compiledDynamicInstructionsDefinitionSchema=z.object({eventNames:z.array(z.string()).readonly(),exportName:z.string().optional(),logicalPath:z.string(),slug:z.string(),sourceId:z.string(),sourceKind:z.literal(`module`)}).strict(),compiledHookDefinitionSchema=z.object({exportName:z.string().optional(),logicalPath:z.string(),slug:z.string(),sourceId:z.string(),sourceKind:z.literal(`module`)}).strict(),compiledAgentNodeManifestSchema=z.object({agentRoot:z.string(),appRoot:z.string(),channels:z.array(compiledChannelEntrySchema),config:compiledAgentConfigSchema,connections:z.array(compiledConnectionDefinitionSchema),diagnosticsSummary:discoverDiagnosticsSummarySchema,disabledFrameworkTools:z.array(z.string()).readonly(),workflowEnabled:z.boolean().default(!1),dynamicInstructions:z.array(compiledDynamicInstructionsDefinitionSchema).default([]),dynamicSkills:z.array(compiledDynamicSkillDefinitionSchema).default([]),dynamicTools:z.array(compiledDynamicToolDefinitionSchema).default([]),hooks:z.array(compiledHookDefinitionSchema),sandbox:compiledSandboxDefinitionSchema.nullable(),sandboxWorkspaces:z.array(compiledSandboxWorkspaceSchema),schedules:z.array(compiledScheduleDefinitionSchema),remoteAgents:z.array(compiledRemoteAgentNodeSchema),skills:z.array(compiledSkillSourceSchema).readonly(),instructions:compiledInstructionsSchema.optional(),tools:z.array(compiledToolDefinitionSchema),workspaceResourceRoot:compiledWorkspaceResourceRootSchema}).strict(),compiledSubagentNodeSchema=z.object({agent:compiledAgentNodeManifestSchema,description:z.string(),entryPath:z.string(),logicalPath:z.string(),name:z.string(),nodeId:z.string(),rootPath:z.string(),sourceId:z.string(),sourceKind:z.literal(`module`),exportName:z.string().optional()}).strict(),compiledSubagentEdgeSchema=z.object({childNodeId:z.string(),parentNodeId:z.string()}).strict(),compiledAgentManifestSchema=z.object({agentRoot:z.string(),appRoot:z.string(),channels:z.array(compiledChannelEntrySchema),config:compiledAgentConfigSchema,connections:z.array(compiledConnectionDefinitionSchema),diagnosticsSummary:discoverDiagnosticsSummarySchema,disabledFrameworkTools:z.array(z.string()).readonly(),workflowEnabled:z.boolean().default(!1),dynamicInstructions:z.array(compiledDynamicInstructionsDefinitionSchema).default([]),dynamicSkills:z.array(compiledDynamicSkillDefinitionSchema).default([]),dynamicTools:z.array(compiledDynamicToolDefinitionSchema).default([]),hooks:z.array(compiledHookDefinitionSchema),kind:z.literal(COMPILED_AGENT_MANIFEST_KIND),remoteAgents:z.array(compiledRemoteAgentNodeSchema),sandbox:compiledSandboxDefinitionSchema.nullable(),sandboxWorkspaces:z.array(compiledSandboxWorkspaceSchema),schedules:z.array(compiledScheduleDefinitionSchema),skills:z.array(compiledSkillSourceSchema).readonly(),subagentEdges:z.array(compiledSubagentEdgeSchema),subagents:z.array(compiledSubagentNodeSchema),instructions:compiledInstructionsSchema.optional(),tools:z.array(compiledToolDefinitionSchema),version:z.literal(30),workspaceResourceRoot:compiledWorkspaceResourceRootSchema}).strict();function createCompiledAgentNodeManifest(e){let t={agentRoot:e.agentRoot,appRoot:e.appRoot,channels:[...e.channels??[]],connections:[...e.connections??[]],config:{build:e.config.build===void 0?void 0:{externalDependencies:e.config.build.externalDependencies===void 0?void 0:[...e.config.build.externalDependencies]},compaction:{model:e.config.compaction?.model===void 0?void 0:cloneCompiledRuntimeModelReference(e.config.compaction.model),thresholdPercent:e.config.compaction?.thresholdPercent},description:e.config.description,experimental:e.config.experimental===void 0?void 0:{workflow:e.config.experimental.workflow===void 0?void 0:{world:e.config.experimental.workflow.world}},model:cloneCompiledRuntimeModelReference(e.config.model),name:e.config.name,outputSchema:e.config.outputSchema,reasoning:e.config.reasoning,source:e.config.source===void 0?void 0:{...e.config.source}},diagnosticsSummary:e.diagnosticsSummary??{errors:0,warnings:0},disabledFrameworkTools:[...e.disabledFrameworkTools??[]],workflowEnabled:e.workflowEnabled??!1,dynamicInstructions:[...e.dynamicInstructions??[]],dynamicSkills:[...e.dynamicSkills??[]],dynamicTools:[...e.dynamicTools??[]],hooks:[...e.hooks??[]],remoteAgents:[...e.remoteAgents??[]],sandbox:e.sandbox??null,sandboxWorkspaces:[...e.sandboxWorkspaces??[]],schedules:[...e.schedules??[]],skills:[...e.skills??[]],tools:[...e.tools??[]],workspaceResourceRoot:e.workspaceResourceRoot??{logicalPath:``,rootEntries:deriveResourceRootEntries({sandboxWorkspaces:e.sandboxWorkspaces,skills:e.skills})}};return e.instructions!==void 0&&(t.instructions=e.instructions),t}function deriveResourceRootEntries(e){let t=new Set;(e.skills??[]).length>0&&t.add(`skills/`);for(let n of e.sandboxWorkspaces??[])for(let e of n.rootEntries)t.add(e);return[...t].sort((e,t)=>e.localeCompare(t))}function createCompiledSubagentNodeId(e,t){return e===`__root__`?t:`${e}::${t}`}function createCompiledAgentManifest(e){return{...createCompiledAgentNodeManifest(e),kind:COMPILED_AGENT_MANIFEST_KIND,subagentEdges:[...e.subagentEdges??[]],subagents:[...e.subagents??[]],version:30}}function cloneCompiledRuntimeModelReference(e){let t={id:e.id,routing:cloneModelRouting(e.routing)};return e.contextWindowTokens!==void 0&&(t.contextWindowTokens=e.contextWindowTokens),e.providerOptions!==void 0&&(t.providerOptions={...e.providerOptions}),e.source!==void 0&&(t.source={...e.source}),t}function cloneModelRouting(e){return e.kind===`external`?{kind:`external`,provider:e.provider}:e.byok===void 0?{kind:`gateway`,target:e.target}:{kind:`gateway`,target:e.target,byok:e.byok}}export{COMPILED_AGENT_MANIFEST_KIND,COMPILED_AGENT_MANIFEST_VERSION,ROOT_COMPILED_AGENT_NODE_ID,compiledAgentManifestSchema,createCompiledAgentManifest,createCompiledAgentNodeManifest,createCompiledSubagentNodeId,deriveResourceRootEntries};

package/dist/src/compiler/normalize-agent-config.js

@@ -1 +1 @@
-import{isAbsolute,join,relative,resolve}from"node:path";import{DEFAULT_AGENT_MODEL_ID}from"#shared/default-agent-model.js";import{toErrorMessage}from"#shared/errors.js";import{normalizeJsonSchemaDefinition}from"#shared/json-schema.js";import{normalizeLogicalPath}from"#discover/filesystem.js";import{classifyModelRouting}from"#internal/classify-model-routing.js";import{normalizeAgentDefinition}from"#internal/authored-definition/core.js";import{formatLanguageModelGatewayId}from"#internal/runtime-model.js";import{parseJsonObject}from"#shared/json.js";import{loadModuleBackedDefinition}from"#compiler/normalize-helpers.js";async function compileAgentConfig(e,t){let n=e.configModule,r=n===void 0?void 0:formatAgentConfigModulePath(e,n),a=normalizeAgentDefinition(n===void 0?{model:DEFAULT_AGENT_MODEL_ID}:await loadModuleBackedDefinition({agentRoot:e.agentRoot,displayPath:r,kind:`agent config`,source:n}),n===void 0?`Expected the default agent config to match the public eve shape.`:`Expected the agent config export "${n.exportName??`default`}" from "${r}" to match the public eve shape.`),o=await normalizeAuthoredModelReference({modelCatalog:t.modelCatalog,purpose:`the primary compaction trigger model`,contextWindowTokens:a.modelContextWindowTokens,providerOptions:a.modelOptions?.providerOptions,source:n,sourcePath:r,value:a.model}),s={},c={compaction:s,model:o,name:e.agentId};a.description!==void 0&&(c.description=a.description);let l=normalizeExperimentalDefinition(a.experimental);return l!==void 0&&(c.experimental=l),a.build!==void 0&&(c.build={externalDependencies:a.build.externalDependencies===void 0?void 0:[...a.build.externalDependencies]}),a.outputSchema!==void 0&&(c.outputSchema=normalizeJsonSchemaDefinition(a.outputSchema,`output`)),n!==void 0&&(c.source={exportName:n.exportName,sourceKind:`module`,logicalPath:n.logicalPath,sourceId:n.sourceId}),a.compaction?.model!==void 0&&(s.model=await normalizeAuthoredModelReference({modelCatalog:t.modelCatalog,purpose:`the compaction summary model`,contextWindowTokens:a.compaction.modelContextWindowTokens,providerOptions:a.modelOptions?.providerOptions,source:n,sourcePath:r,value:a.compaction.model})),a.compaction?.thresholdPercent!==void 0&&(s.thresholdPercent=a.compaction.thresholdPercent),c}function normalizeExperimentalDefinition(e){if(e===void 0)return;let t={};return e.codeMode!==void 0&&(t.codeMode=e.codeMode),e.workflow!==void 0&&(t.workflow={world:e.workflow.world}),t}async function normalizeAuthoredModelReference(e){if(typeof e.value==`string`)return await withCompiledRuntimeModelLimits({id:formatLanguageModelGatewayId(e.value),providerOptions:parseProviderOptionsRecord(e.providerOptions),routing:classifyModelRouting(e.value,e.providerOptions)},e);let t=e.source;if(t===void 0)throw Error(`Expected ${e.purpose} to provide a valid AI SDK language model reference.`);let n=e.value,r=n.specificationVersion;if(r!==`v2`&&r!==`v3`&&r!==`v4`||typeof n.provider!=`string`||typeof n.modelId!=`string`||typeof n.doGenerate!=`function`||typeof n.doStream!=`function`)throw Error(`Expected the authored agent config export "${t.exportName??`default`}" from "${e.sourcePath??t.logicalPath}" to provide a valid AI SDK language model.`);let i={id:formatLanguageModelGatewayId(n),source:{exportName:t.exportName,sourceKind:`module`,logicalPath:t.logicalPath,sourceId:t.sourceId},providerOptions:parseProviderOptionsRecord(e.providerOptions),routing:classifyModelRouting(n,e.providerOptions)};if(e.contextWindowTokens===void 0){let t=await e.modelCatalog.getByProviderModelId(n.provider,n.modelId);if(t)return{...i,id:t.slug,contextWindowTokens:t.limits.contextWindowTokens}}return await withCompiledRuntimeModelLimits(i,e)}function formatAgentConfigModulePath(e,r){let i=join(e.agentRoot,r.logicalPath);return normalizeLogicalPath(relative(resolveTopLevelAgentRoot(e),i))}function resolveTopLevelAgentRoot(e){let t=resolve(e.appRoot),n=resolve(t,`agent`);return isPathInsideOrEqual(n,resolve(e.agentRoot))?n:t}function isPathInsideOrEqual(t,r){let i=relative(t,r);return i===``||!i.startsWith(`..`)&&!isAbsolute(i)}async function withCompiledRuntimeModelLimits(e,t){if(t.contextWindowTokens!==void 0)return{...e,contextWindowTokens:t.contextWindowTokens};let n;try{n=await t.modelCatalog.getModelLimits(e.id)}catch(n){throw Error(`Failed to load AI Gateway model metadata for ${t.purpose} "${e.id}". ${toErrorMessage(n)}`)}if(n===null)throw Error(`Cannot compile agent compaction because ${t.purpose} "${e.id}" does not have known AI Gateway context window metadata.`);return{...e,contextWindowTokens:n.contextWindowTokens}}function parseProviderOptionsRecord(e){if(e===void 0)return;let t={};for(let[n,r]of Object.entries(e))t[n]=parseJsonObject(r);return t}export{compileAgentConfig};
+import{isAbsolute,join,relative,resolve}from"node:path";import{DEFAULT_AGENT_MODEL_ID}from"#shared/default-agent-model.js";import{toErrorMessage}from"#shared/errors.js";import{normalizeJsonSchemaDefinition}from"#shared/json-schema.js";import{normalizeLogicalPath}from"#discover/filesystem.js";import{classifyModelRouting}from"#internal/classify-model-routing.js";import{normalizeAgentDefinition}from"#internal/authored-definition/core.js";import{formatLanguageModelGatewayId}from"#internal/runtime-model.js";import{parseJsonObject}from"#shared/json.js";import{loadModuleBackedDefinition}from"#compiler/normalize-helpers.js";async function compileAgentConfig(e,t){let n=e.configModule,r=n===void 0?void 0:formatAgentConfigModulePath(e,n),a=normalizeAgentDefinition(n===void 0?{model:DEFAULT_AGENT_MODEL_ID}:await loadModuleBackedDefinition({agentRoot:e.agentRoot,displayPath:r,kind:`agent config`,source:n}),n===void 0?`Expected the default agent config to match the public eve shape.`:`Expected the agent config export "${n.exportName??`default`}" from "${r}" to match the public eve shape.`),o=await normalizeAuthoredModelReference({modelCatalog:t.modelCatalog,purpose:`the primary compaction trigger model`,contextWindowTokens:a.modelContextWindowTokens,providerOptions:a.modelOptions?.providerOptions,source:n,sourcePath:r,value:a.model}),s={},c={compaction:s,model:o,name:e.agentId};a.description!==void 0&&(c.description=a.description);let l=normalizeExperimentalDefinition(a.experimental);return l!==void 0&&(c.experimental=l),a.build!==void 0&&(c.build={externalDependencies:a.build.externalDependencies===void 0?void 0:[...a.build.externalDependencies]}),a.outputSchema!==void 0&&(c.outputSchema=normalizeJsonSchemaDefinition(a.outputSchema,`output`)),a.reasoning!==void 0&&(c.reasoning=a.reasoning),n!==void 0&&(c.source={exportName:n.exportName,sourceKind:`module`,logicalPath:n.logicalPath,sourceId:n.sourceId}),a.compaction?.model!==void 0&&(s.model=await normalizeAuthoredModelReference({modelCatalog:t.modelCatalog,purpose:`the compaction summary model`,contextWindowTokens:a.compaction.modelContextWindowTokens,providerOptions:a.modelOptions?.providerOptions,source:n,sourcePath:r,value:a.compaction.model})),a.compaction?.thresholdPercent!==void 0&&(s.thresholdPercent=a.compaction.thresholdPercent),c}function normalizeExperimentalDefinition(e){if(e===void 0)return;let t={};return e.workflow!==void 0&&(t.workflow={world:e.workflow.world}),t}async function normalizeAuthoredModelReference(e){if(typeof e.value==`string`)return await withCompiledRuntimeModelLimits({id:formatLanguageModelGatewayId(e.value),providerOptions:parseProviderOptionsRecord(e.providerOptions),routing:classifyModelRouting(e.value,e.providerOptions)},e);let t=e.source;if(t===void 0)throw Error(`Expected ${e.purpose} to provide a valid AI SDK language model reference.`);let n=e.value,r=n.specificationVersion;if(r!==`v2`&&r!==`v3`&&r!==`v4`||typeof n.provider!=`string`||typeof n.modelId!=`string`||typeof n.doGenerate!=`function`||typeof n.doStream!=`function`)throw Error(`Expected the authored agent config export "${t.exportName??`default`}" from "${e.sourcePath??t.logicalPath}" to provide a valid AI SDK language model.`);let i={id:formatLanguageModelGatewayId(n),source:{exportName:t.exportName,sourceKind:`module`,logicalPath:t.logicalPath,sourceId:t.sourceId},providerOptions:parseProviderOptionsRecord(e.providerOptions),routing:classifyModelRouting(n,e.providerOptions)};if(e.contextWindowTokens===void 0){let t=await e.modelCatalog.getByProviderModelId(n.provider,n.modelId);if(t)return{...i,id:t.slug,contextWindowTokens:t.limits.contextWindowTokens}}return await withCompiledRuntimeModelLimits(i,e)}function formatAgentConfigModulePath(e,r){let i=join(e.agentRoot,r.logicalPath);return normalizeLogicalPath(relative(resolveTopLevelAgentRoot(e),i))}function resolveTopLevelAgentRoot(e){let t=resolve(e.appRoot),n=resolve(t,`agent`);return isPathInsideOrEqual(n,resolve(e.agentRoot))?n:t}function isPathInsideOrEqual(t,r){let i=relative(t,r);return i===``||!i.startsWith(`..`)&&!isAbsolute(i)}async function withCompiledRuntimeModelLimits(e,t){if(t.contextWindowTokens!==void 0)return{...e,contextWindowTokens:t.contextWindowTokens};let n;try{n=await t.modelCatalog.getModelLimits(e.id)}catch(n){throw Error(`Failed to load AI Gateway model metadata for ${t.purpose} "${e.id}". ${toErrorMessage(n)}`)}if(n===null)throw Error(`Cannot compile agent compaction because ${t.purpose} "${e.id}" does not have known AI Gateway context window metadata.`);return{...e,contextWindowTokens:n.contextWindowTokens}}function parseProviderOptionsRecord(e){if(e===void 0)return;let t={};for(let[n,r]of Object.entries(e))t[n]=parseJsonObject(r);return t}export{compileAgentConfig};

package/dist/src/context/build-dynamic-tools.js

@@ -1 +1 @@
-import{createLogger}from"#internal/logging.js";import{LiveStepToolsKey,SessionDynamicToolMetadataKey,TurnDynamicToolMetadataKey}from"#context/keys.js";import{jsonSchema}from"ai";import{buildCallbackContext}from"#context/build-callback-context.js";const log=createLogger(`dynamic-tools`);function lookupStepFunction(e){try{let t=globalThis[Symbol.for(`@workflow/core//registeredSteps`)];return t===void 0?null:t.get(e)||null}catch{return null}}function replayTools(e){let t=[];for(let n of e){if(!n.executeStepFnName||!n.closureVars){log.warn(`Dynamic tool "${n.name}" has no registered step function — skipping on this step. The bundler transform may not have processed this tool file.`);continue}let e=lookupStepFunction(n.executeStepFnName);if(!e){log.warn(`Dynamic tool "${n.name}" references step function "${n.executeStepFnName}" which is not registered — skipping on this step.`);continue}t.push({description:n.description,execute:t=>e(n.closureVars,t,buildCallbackContext()),inputSchema:jsonSchema(n.inputSchema),name:n.name,needsApproval:buildReplayedNeedsApproval(n),outputSchema:n.outputSchema===void 0?void 0:jsonSchema(n.outputSchema)})}return t}function buildReplayedNeedsApproval(e){if(e.needsApprovalStepFnName===void 0)return;let t=lookupStepFunction(e.needsApprovalStepFnName);return t===null?(log.warn(`Dynamic tool "${e.name}" references approval function "${e.needsApprovalStepFnName}" which is not registered — requiring approval by default.`),()=>!0):n=>!!t(e.closureVars??{},n)}function buildDynamicTools(e){let r=e.get(LiveStepToolsKey)??[],i=replayTools(e.get(TurnDynamicToolMetadataKey)??[]),a=replayTools(e.get(SessionDynamicToolMetadataKey)??[]);return[...r,...i,...a]}export{buildDynamicTools};
+import{createLogger}from"#internal/logging.js";import{LiveStepToolsKey,SessionDynamicToolMetadataKey,TurnDynamicToolMetadataKey}from"#context/keys.js";import{jsonSchema}from"ai";import{buildCallbackContext}from"#context/build-callback-context.js";const log=createLogger(`dynamic-tools`);function lookupStepFunction(e){try{let t=globalThis[Symbol.for(`@workflow/core//registeredSteps`)];return t===void 0?null:t.get(e)||null}catch{return null}}function replayTools(e){let t=[];for(let n of e){if(!n.executeStepFnName||!n.closureVars){log.warn(`Dynamic tool "${n.name}" has no registered step function — skipping on this step. The bundler transform may not have processed this tool file.`);continue}let e=lookupStepFunction(n.executeStepFnName);if(!e){log.warn(`Dynamic tool "${n.name}" references step function "${n.executeStepFnName}" which is not registered — skipping on this step.`);continue}t.push({description:n.description,execute:t=>e(n.closureVars,t,buildCallbackContext()),inputSchema:jsonSchema(n.inputSchema),name:n.name,approval:buildReplayedApproval(n),outputSchema:n.outputSchema===void 0?void 0:jsonSchema(n.outputSchema)})}return t}function buildReplayedApproval(e){if(e.approvalStepFnName===void 0)return;let t=lookupStepFunction(e.approvalStepFnName);return t===null?(log.warn(`Dynamic tool "${e.name}" references approval function "${e.approvalStepFnName}" which is not registered — requiring approval by default.`),()=>`user-approval`):async n=>await t(e.closureVars??{},n)}function buildDynamicTools(e){let r=e.get(LiveStepToolsKey)??[],i=replayTools(e.get(TurnDynamicToolMetadataKey)??[]),a=replayTools(e.get(SessionDynamicToolMetadataKey)??[]);return[...r,...i,...a]}export{buildDynamicTools};

package/dist/src/context/dynamic-tool-lifecycle.js

@@ -1 +1 @@
-import{createLogger}from"#internal/logging.js";import{LiveStepToolsKey,SessionDynamicToolMetadataKey,TurnDynamicToolMetadataKey}from"#context/keys.js";import{toErrorMessage}from"#shared/errors.js";import{ALLOWED_DYNAMIC_TOOL_EVENTS,isBrandedToolEntry}from"#shared/dynamic-tool-definition.js";import{jsonSchema,zodSchema}from"ai";import{buildCallbackContext}from"#context/build-callback-context.js";import{buildResolveContext}from"#context/dynamic-resolve-context.js";import{normalizeJsonSchemaDefinition}from"#internal/json-schema.js";const log=createLogger(`dynamic-tools`);function toHarnessToolDefinition(e,t){return{description:t.description,execute:e=>t.execute(e,buildCallbackContext()),inputSchema:convertInputSchema(t.inputSchema),name:e,needsApproval:t.needsApproval,outputSchema:convertOptionalOutputSchema(t.outputSchema),...t.toModelOutput===void 0?{}:{toModelOutput:t.toModelOutput}}}function convertInputSchema(e){return typeof e==`object`&&e&&`~standard`in e?zodSchema(e):jsonSchema(e)}function convertOptionalOutputSchema(e){if(e!==void 0)return typeof e==`object`&&e&&`~standard`in e?zodSchema(e):jsonSchema(e)}function qualifyDynamicToolNames(e,t,n){let r=Object.keys(n),i=[];if(r.length===0)return i;if(t)return i.push({name:e,entryKey:r[0],entry:n[r[0]]}),i;for(let e of r)i.push({name:e,entryKey:e,entry:n[e]});return i}function replayDynamicSessionTools(e,t){let n=[];for(let t of e){if(!t.executeStepFnName||!t.closureVars){log.warn(`Dynamic tool "${t.name}" has no registered step function — skipping on this step. The bundler transform may not have processed this tool file.`);continue}let e=lookupStepFunction(t.executeStepFnName);if(!e){log.warn(`Dynamic tool "${t.name}" references step function "${t.executeStepFnName}" which is not registered — skipping on this step.`);continue}n.push({description:t.description,execute:n=>e(t.closureVars,n,buildCallbackContext()),inputSchema:jsonSchema(t.inputSchema),name:t.name,outputSchema:t.outputSchema===void 0?void 0:jsonSchema(t.outputSchema)})}return n}function getStepRegistry(){let e=Symbol.for(`@workflow/core//registeredSteps`),t=globalThis,n=t[e];return n===void 0&&(n=new Map,t[e]=n),n}function lookupStepFunction(e){try{return getStepRegistry().get(e)||null}catch{return null}}function registerStepFunction(e,t){getStepRegistry().set(e,t)}function safeSerialize(e){try{return JSON.parse(JSON.stringify(e))}catch{return{}}}function durableKeyForEvent(e){switch(e){case`session.started`:return SessionDynamicToolMetadataKey;case`turn.started`:return TurnDynamicToolMetadataKey;default:return}}async function resolveToolsFromEvent(e,t,n,r){let a=await Promise.allSettled(t.map(async t=>{let i=t.events[n.type];if(i===void 0)return null;let a=await i(n,buildResolveContext(e,r));if(a==null)return null;let s,c;return isBrandedToolEntry(a)?(s={_single:a},c=!0):(s=a,c=!1),{resolver:t,entries:s,isSingle:c}})),s=[],c=[],l=new Map;for(let e of a){if(e.status===`rejected`){log.error(`Dynamic tool resolver (${n.type}) threw — skipping.`,{error:toErrorMessage(e.reason)});continue}if(e.value===null)continue;let{resolver:t,entries:r,isSingle:a}=e.value,o=qualifyDynamicToolNames(t.slug,a,r);for(let{name:e,entryKey:r,entry:i}of o){let a=l.get(e);if(a!==void 0&&a!==t.slug)throw Error(`Dynamic tool "${e}" from resolver "${t.slug}" collides with dynamic resolver "${a}". Namespace the map key manually, e.g. "${t.slug}__${e}".`);if(l.set(e,t.slug),c.push(toHarnessToolDefinition(e,i)),n.type===`step.started`)continue;let o=`__executeStepFn`in i?i.__executeStepFn:void 0,u=`__closureVars`in i?i.__closureVars:void 0,f=o?.stepId,p=u===void 0?void 0:safeSerialize(u);if(f===void 0){let e=`eve:framework-dynamic:${t.slug}:${r}`,n=i.execute.bind(i);registerStepFunction(e,(e,t,r)=>n(t,r)),f=e,p={}}let m;if(i.needsApproval!==void 0){m=`eve:dynamic-tool-approval:${t.slug}:${r}`;let e=i.needsApproval.bind(i);registerStepFunction(m,(t,n)=>e(n))}s.push({name:e,description:i.description,inputSchema:normalizeJsonSchemaDefinition(i.inputSchema),outputSchema:i.outputSchema===void 0?void 0:normalizeJsonSchemaDefinition(i.outputSchema,`output`),resolverSlug:t.slug,entryKey:r,executeStepFnName:f,needsApprovalStepFnName:m,closureVars:p})}}return{metadata:s,liveTools:c}}async function dispatchDynamicToolEvent(e){let{ctx:n,resolvers:r,event:i,messages:o}=e;if(!ALLOWED_DYNAMIC_TOOL_EVENTS.has(i.type))return;let s=r.filter(e=>e.eventNames.includes(i.type));if(s.length===0)return;let{metadata:c,liveTools:l}=await resolveToolsFromEvent(n,s,i,o);if(i.type===`step.started`){n.setVirtualContext(LiveStepToolsKey,l);return}let u=durableKeyForEvent(i.type);if(u===void 0)return;let d=new Set(s.map(e=>e.slug)),f=(n.get(u)??[]).filter(e=>!d.has(e.resolverSlug));n.set(u,[...f,...c])}export{dispatchDynamicToolEvent,replayDynamicSessionTools};
+import{createLogger}from"#internal/logging.js";import{LiveStepToolsKey,SessionDynamicToolMetadataKey,TurnDynamicToolMetadataKey}from"#context/keys.js";import{toErrorMessage}from"#shared/errors.js";import{ALLOWED_DYNAMIC_TOOL_EVENTS,isBrandedToolEntry}from"#shared/dynamic-tool-definition.js";import{jsonSchema,zodSchema}from"ai";import{buildCallbackContext}from"#context/build-callback-context.js";import{buildResolveContext}from"#context/dynamic-resolve-context.js";import{normalizeJsonSchemaDefinition}from"#internal/json-schema.js";const log=createLogger(`dynamic-tools`);function toHarnessToolDefinition(e,t){return{description:t.description,execute:e=>t.execute(e,buildCallbackContext()),inputSchema:convertInputSchema(t.inputSchema),name:e,approval:t.approval,outputSchema:convertOptionalOutputSchema(t.outputSchema),...t.toModelOutput===void 0?{}:{toModelOutput:t.toModelOutput}}}function convertInputSchema(e){return typeof e==`object`&&e&&`~standard`in e?zodSchema(e):jsonSchema(e)}function convertOptionalOutputSchema(e){if(e!==void 0)return typeof e==`object`&&e&&`~standard`in e?zodSchema(e):jsonSchema(e)}function qualifyDynamicToolNames(e,t,n){let r=Object.keys(n),i=[];if(r.length===0)return i;if(t)return i.push({name:e,entryKey:r[0],entry:n[r[0]]}),i;for(let e of r)i.push({name:e,entryKey:e,entry:n[e]});return i}function replayDynamicSessionTools(e,t){let n=[];for(let t of e){if(!t.executeStepFnName||!t.closureVars){log.warn(`Dynamic tool "${t.name}" has no registered step function — skipping on this step. The bundler transform may not have processed this tool file.`);continue}let e=lookupStepFunction(t.executeStepFnName);if(!e){log.warn(`Dynamic tool "${t.name}" references step function "${t.executeStepFnName}" which is not registered — skipping on this step.`);continue}n.push({description:t.description,execute:n=>e(t.closureVars,n,buildCallbackContext()),inputSchema:jsonSchema(t.inputSchema),name:t.name,outputSchema:t.outputSchema===void 0?void 0:jsonSchema(t.outputSchema)})}return n}function getStepRegistry(){let e=Symbol.for(`@workflow/core//registeredSteps`),t=globalThis,n=t[e];return n===void 0&&(n=new Map,t[e]=n),n}function lookupStepFunction(e){try{return getStepRegistry().get(e)||null}catch{return null}}function registerStepFunction(e,t){getStepRegistry().set(e,t)}function safeSerialize(e){try{return JSON.parse(JSON.stringify(e))}catch{return{}}}function durableKeyForEvent(e){switch(e){case`session.started`:return SessionDynamicToolMetadataKey;case`turn.started`:return TurnDynamicToolMetadataKey;default:return}}async function resolveToolsFromEvent(e,t,n,r){let a=await Promise.allSettled(t.map(async t=>{let i=t.events[n.type];if(i===void 0)return null;let a=await i(n,buildResolveContext(e,r));if(a==null)return null;let s,c;return isBrandedToolEntry(a)?(s={_single:a},c=!0):(s=a,c=!1),{resolver:t,entries:s,isSingle:c}})),s=[],c=[],l=new Map;for(let e of a){if(e.status===`rejected`){log.error(`Dynamic tool resolver (${n.type}) threw — skipping.`,{error:toErrorMessage(e.reason)});continue}if(e.value===null)continue;let{resolver:t,entries:r,isSingle:a}=e.value,o=qualifyDynamicToolNames(t.slug,a,r);for(let{name:e,entryKey:r,entry:i}of o){let a=l.get(e);if(a!==void 0&&a!==t.slug)throw Error(`Dynamic tool "${e}" from resolver "${t.slug}" collides with dynamic resolver "${a}". Namespace the map key manually, e.g. "${t.slug}__${e}".`);if(l.set(e,t.slug),c.push(toHarnessToolDefinition(e,i)),n.type===`step.started`)continue;let o=`__executeStepFn`in i?i.__executeStepFn:void 0,u=`__closureVars`in i?i.__closureVars:void 0,f=o?.stepId,p=u===void 0?void 0:safeSerialize(u);if(f===void 0){let e=`eve:framework-dynamic:${t.slug}:${r}`,n=i.execute.bind(i);registerStepFunction(e,(e,t,r)=>n(t,r)),f=e,p={}}let m;if(i.approval!==void 0){m=`eve:dynamic-tool-approval:${t.slug}:${r}`;let e=i.approval.bind(i);registerStepFunction(m,(t,n)=>e(n))}s.push({name:e,description:i.description,inputSchema:normalizeJsonSchemaDefinition(i.inputSchema),outputSchema:i.outputSchema===void 0?void 0:normalizeJsonSchemaDefinition(i.outputSchema,`output`),resolverSlug:t.slug,entryKey:r,executeStepFnName:f,approvalStepFnName:m,closureVars:p})}}return{metadata:s,liveTools:c}}async function dispatchDynamicToolEvent(e){let{ctx:n,resolvers:r,event:i,messages:o}=e;if(!ALLOWED_DYNAMIC_TOOL_EVENTS.has(i.type))return;let s=r.filter(e=>e.eventNames.includes(i.type));if(s.length===0)return;let{metadata:c,liveTools:l}=await resolveToolsFromEvent(n,s,i,o);if(i.type===`step.started`){n.setVirtualContext(LiveStepToolsKey,l);return}let u=durableKeyForEvent(i.type);if(u===void 0)return;let d=new Set(s.map(e=>e.slug)),f=(n.get(u)??[]).filter(e=>!d.has(e.resolverSlug));n.set(u,[...f,...c])}export{dispatchDynamicToolEvent,replayDynamicSessionTools};

package/dist/src/context/keys.d.ts

@@ -62,7 +62,7 @@ export interface DurableDynamicToolMetadata {
     readonly resolverSlug: string;
     readonly entryKey: string;
     readonly executeStepFnName?: string;
-    readonly needsApprovalStepFnName?: string;
+    readonly approvalStepFnName?: string;
     readonly closureVars?: Record<string, unknown>;
 }
 /**

package/dist/src/evals/assertions/collector.d.ts

@@ -1,4 +1,5 @@
 import type { AssertionHandle, AssertionResult, AssertionSeverity, EveEvalTaskResult } from "#evals/types.js";
+import type { EveEvalAssertionSubject } from "#evals/assertions/run.js";
 /**
  * Outcome of evaluating one assertion: a 0–1 score (boolean assertions use
  * exactly 0 or 1) with optional human-readable detail and metadata.
@@ -9,13 +10,12 @@ export interface AssertionOutcome {
     readonly metadata?: Readonly<Record<string, unknown>>;
 }
 /**
- * A run-level assertion (e.g. `t.completed()`), evaluated lazily against the
- * final task result after `test(t)` returns. The evaluation is deferred so
- * the assertion always sees the complete run regardless of call order.
+ * A scoped assertion evaluated lazily after `test(t)` returns. The selected
+ * subject may be the aggregate run, one session, or one immutable turn.
  */
 export interface RunAssertion {
     readonly name: string;
-    evaluate(result: EveEvalTaskResult): AssertionOutcome | Promise<AssertionOutcome>;
+    evaluate(result: EveEvalAssertionSubject): AssertionOutcome | Promise<AssertionOutcome>;
 }
 /**
  * Collects the assertions recorded by an eval's `test(t)`. Run-level
@@ -26,8 +26,10 @@ export interface RunAssertion {
  */
 export declare class AssertionCollector {
     #private;
-    /** Register a run-level assertion evaluated against the final result. */
-    recordRun(spec: RunAssertion, severity?: AssertionSeverity): AssertionHandle;
+    /** Whether the eval has already recorded an assertion. */
+    get hasEntries(): boolean;
+    /** Register a deferred assertion against a turn or session scope. */
+    recordScoped(spec: RunAssertion, selectSubject: (result: EveEvalTaskResult) => EveEvalAssertionSubject, severity?: AssertionSeverity): AssertionHandle;
     /** Register a value/judge assertion, evaluating the captured value now. */
     recordValue(input: {
         readonly name: string;
@@ -35,6 +37,17 @@ export declare class AssertionCollector {
         readonly threshold?: number;
         readonly score: () => Promise<AssertionOutcome>;
     }): AssertionHandle;
+    /** Record an already-computed assertion outcome and return whether it passed. */
+    recordOutcome(input: {
+        readonly name: string;
+        readonly outcome: AssertionOutcome;
+    }): boolean;
+    /** Record and await a required value assertion, returning whether it passed. */
+    recordRequirement(input: {
+        readonly name: string;
+        readonly threshold?: number;
+        readonly score: () => Promise<AssertionOutcome>;
+    }): Promise<boolean>;
     /**
      * Awaits every pending value/judge assertion, evaluates the deferred
      * run-level assertions against `result`, and returns the recorded results.

package/dist/src/evals/assertions/collector.js

@@ -1 +1 @@
-import{toErrorMessage}from"#shared/errors.js";var AssertionCollector=class{#e=[];#t=[];recordRun(e,t=`gate`){let n={name:e.name,severity:t,threshold:void 0,kind:`deferred`,spec:e,score:0,failed:!1};return this.#e.push(n),makeHandle(n,Promise.resolve())}recordValue(t){let n={name:t.name,severity:t.severity,threshold:t.threshold,kind:`resolved`,score:0,failed:!1};this.#e.push(n);let r=t.score().then(e=>{n.score=e.score,n.message=e.message,n.metadata=e.metadata}).catch(t=>{n.score=0,n.severity=`gate`,n.threshold=void 0,n.message=toErrorMessage(t),n.failed=!0});return this.#t.push(r),makeHandle(n,r)}async finalize(e){await Promise.all(this.#t);let t=[];for(let n of this.#e){if(n.kind===`deferred`&&n.spec!==void 0){let t=await n.spec.evaluate(e);n.score=t.score,n.message=t.message,n.metadata=t.metadata}t.push({name:n.name,score:n.score,severity:n.severity,threshold:n.threshold,passed:computePassed(n.severity,n.threshold,n.score,n.failed),message:n.message,metadata:n.metadata})}return t}};function computePassed(e,t,n,r){if(r)return!1;let i=t??(e===`gate`?1:void 0);return i===void 0||n>=i}function makeHandle(e,t){let n={gate(t){return e.severity=`gate`,e.threshold=t,n},soft(t){return e.severity=`soft`,e.threshold=t,n},atLeast(t){return e.severity=`soft`,e.threshold=t,n},then(e,n){return t.then(e,n)}};return n}export{AssertionCollector};
+import{toErrorMessage}from"#shared/errors.js";var AssertionCollector=class{#e=[];#t=[];get hasEntries(){return this.#e.length>0}recordScoped(e,t,n=`gate`){let r={name:e.name,severity:n,threshold:void 0,kind:`deferred`,spec:e,selectSubject:t,score:0,failed:!1};return this.#e.push(r),makeHandle(r)}recordValue(e){let t={name:e.name,severity:e.severity,threshold:e.threshold,kind:`resolved`,score:0,failed:!1};this.#e.push(t);let n=settleEntry(t,e.score);return this.#t.push(n),makeHandle(t)}recordOutcome(e){let t={name:e.name,severity:`gate`,threshold:void 0,kind:`resolved`,score:e.outcome.score,message:e.outcome.message,metadata:e.outcome.metadata,failed:!1};return this.#e.push(t),computePassed(t.severity,t.threshold,t.score,t.failed)}async recordRequirement(e){let t={name:e.name,severity:`gate`,threshold:e.threshold,kind:`resolved`,score:0,failed:!1};return this.#e.push(t),await settleEntry(t,e.score),computePassed(t.severity,t.threshold,t.score,t.failed)}async finalize(e){await Promise.all(this.#t);let t=[];for(let n of this.#e){if(n.kind===`deferred`&&n.spec!==void 0){let t=await n.spec.evaluate(n.selectSubject?.(e)??e);n.score=t.score,n.message=t.message,n.metadata=t.metadata}t.push({name:n.name,score:n.score,severity:n.severity,threshold:n.threshold,passed:computePassed(n.severity,n.threshold,n.score,n.failed),message:n.message,metadata:n.metadata})}return t}};function computePassed(e,t,n,r){if(r)return!1;let i=t??(e===`gate`?1:void 0);return i===void 0||n>=i}async function settleEntry(t,n){try{let e=await n();t.score=e.score,t.message=e.message,t.metadata=e.metadata}catch(n){t.score=0,t.severity=`gate`,t.threshold=void 0,t.message=toErrorMessage(n),t.failed=!0}}function makeHandle(e){let t={gate(n){return e.severity=`gate`,e.threshold=n,t},soft(n){return e.severity=`soft`,e.threshold=n,t},atLeast(n){return e.severity=`soft`,e.threshold=n,t}};return t}export{AssertionCollector};

package/dist/src/evals/assertions/run.d.ts

@@ -1,37 +1,38 @@
 import type { StandardSchemaV1 } from "#compiled/@standard-schema/spec/index.js";
 import type { HandleMessageStreamEvent } from "#protocol/message.js";
 import { type EveEvalSkillLoadMatchOptions, type EveEvalSubagentCallMatchOptions, type EveEvalToolCallMatchOptions } from "#evals/match.js";
+import type { EveEvalEventMatch } from "#evals/match.js";
 import type { RunAssertion } from "#evals/assertions/collector.js";
+/** Minimal captured scope consumed by deterministic eval assertions. */
+export interface EveEvalAssertionSubject {
+    readonly derived: import("#evals/types.js").EveEvalDerivedFacts;
+    readonly events: readonly HandleMessageStreamEvent[];
+    readonly output: unknown;
+    readonly status: "completed" | "failed" | "waiting";
+}
 /**
- * Asserts the run ran to completion: it did not fail and did not park on an
+ * Asserts the run finished successfully: it did not fail and did not park on an
  * unanswered HITL input request.
  */
-export declare function completed(): RunAssertion;
+export declare function succeeded(): RunAssertion;
 /**
- * Asserts the run ended parked on HITL input. For approval gates and
- * ask-question flows.
+ * Asserts the run ended cleanly parked on HITL input.
  */
-export declare function waiting(): RunAssertion;
-/**
- * Asserts the run did not fail: terminal status is not `"failed"` and no
- * `turn.failed` / `step.failed` events were emitted. Parked runs pass; use
- * {@link completed} to also reject parking.
- */
-export declare function didNotFail(): RunAssertion;
+export declare function parked(): RunAssertion;
 /**
  * Asserts the joined assistant message text contains `token` (substring for
  * strings, `test` for RegExps).
  */
 export declare function messageIncludes(token: string | RegExp): RunAssertion;
 /**
- * Asserts a tool call with `name` happened. Options constrain the call
- * further: `input` partial-deep-matches, `output` matches the result,
- * `isError` constrains error state, and `times` requires an exact count.
+ * Asserts a completed tool call with `name` happened. Options constrain the
+ * call further: `input` partial-deep-matches, `output` matches the result,
+ * `status` overrides the lifecycle state, and `count` requires an exact count.
  */
 export declare function calledTool(name: string, options?: EveEvalToolCallMatchOptions): RunAssertion;
 /**
  * Sugar over {@link calledTool} for the framework `load_skill` tool: asserts a
- * skill with id `skill` was loaded. `output`/`isError`/`times` constrain the
+ * skill with id `skill` was loaded. `output`/`status`/`count` constrain the
  * matching call exactly as for `calledTool`.
  */
 export declare function loadedSkill(skill: string, options?: EveEvalSkillLoadMatchOptions): RunAssertion;
@@ -40,7 +41,7 @@ export declare function loadedSkill(skill: string, options?: EveEvalSkillLoadMat
  */
 export declare function notCalledTool(name: string): RunAssertion;
 /**
- * Asserts the named tools were called in the given order (subsequence match:
+ * Asserts the named tools were requested in the given order (subsequence match:
  * other calls may interleave).
  */
 export declare function toolOrder(names: readonly string[]): RunAssertion;
@@ -66,7 +67,13 @@ export declare function calledSubagent(name: string, options?: EveEvalSubagentCa
  * Escape hatch: asserts an arbitrary predicate over the full typed event
  * stream. `label` names the assertion in reports.
  */
-export declare function event(predicate: (events: readonly HandleMessageStreamEvent[]) => boolean, label: string): RunAssertion;
+export declare function eventsSatisfy(label: string, predicate: (events: readonly HandleMessageStreamEvent[]) => boolean): RunAssertion;
+/** Asserts a typed stream event exists, optionally with an exact count. */
+export declare function typedEvent(matcher: EveEvalEventMatch): RunAssertion;
+/** Asserts no typed stream event matches. */
+export declare function notEvent(matcher: EveEvalEventMatch): RunAssertion;
+/** Asserts typed event groups occur in order, ignoring unrelated events. */
+export declare function eventOrder(matchers: readonly EveEvalEventMatch[]): RunAssertion;
 /**
  * Asserts `result.output` (the final assistant message) deep-equals `value`.
  */

package/dist/src/evals/assertions/run.js

@@ -1,2 +1,2 @@
-import{deepEquals,subagentCallMatches,testRegExp,toolCallMatches}from"#evals/match.js";import{LOAD_SKILL_TOOL_NAME}from"#runtime/skills/fragment-context.js";const PASS={score:1},fail=(e,t)=>({score:0,message:e,metadata:t});function completed(){return{name:`completed`,evaluate(e){return e.status===`failed`?fail(failureDetail(`run failed`,e.derived.failureCode)):e.derived.parked?fail(`run parked on ${e.derived.inputRequests.length} unanswered input request(s)`):PASS}}}function waiting(){return{name:`waiting`,evaluate(e){return e.derived.parked?PASS:fail(`expected the run to park on HITL input; it ended "${e.status}" with no pending requests`)}}}function didNotFail(){return{name:`didNotFail`,evaluate(e){if(e.status===`failed`)return fail(failureDetail(`run failed`,e.derived.failureCode));let t=e.events.find(e=>e.type===`turn.failed`||e.type===`step.failed`);return t===void 0?PASS:fail(`${t.type} (${t.data.code}): ${t.data.message}`)}}}function messageIncludes(e){return{name:`messageIncludes(${String(e)})`,evaluate(t){let r=joinCompletedMessages(t.events);return(typeof e==`string`?r.includes(e):testRegExp(e,r))?PASS:fail(`assistant messages did not include ${String(e)}; got: ${truncate(r)}`)}}}function calledTool(e,t={}){return{name:`calledTool(${e})`,evaluate(n){let i=n.derived.toolCalls.filter(t=>t.name===e),a=i.filter(e=>toolCallMatches(e,t));if(t.times===void 0?a.length>0:a.length===t.times)return{score:1,metadata:{matchingCalls:a.length}};let o=i.length>0?`observed ${e} calls: ${i.map(e=>truncate(JSON.stringify(e.input))).join(`, `)}`:`observed tools: [${n.derived.toolCalls.map(e=>e.name).join(`, `)}]`;return fail(`${t.times===void 0?`expected a matching call to "${e}"`:`expected exactly ${t.times} matching call(s), found ${a.length}`}; ${o}`)}}}function loadedSkill(e,t={}){return{...calledTool(LOAD_SKILL_TOOL_NAME,{...t,input:{skill:e}}),name:`loadedSkill(${e})`}}function notCalledTool(e){return{name:`notCalledTool(${e})`,evaluate(t){let n=t.derived.toolCalls.filter(t=>t.name===e).length;return n===0?PASS:fail(`"${e}" was called ${n} time(s)`)}}}function toolOrder(e){return{name:`toolOrder(${e.join(` → `)})`,evaluate(t){let n=t.derived.toolCalls.map(e=>e.name),r=0;for(let t of n)if(t===e[r]&&(r+=1),r===e.length)break;return r===e.length?PASS:fail(`missing "${e[r]}" after [${e.slice(0,r).join(`, `)}]; observed order: [${n.join(`, `)}]`)}}}function usedNoTools(){return{name:`usedNoTools`,evaluate(e){let t=e.derived.toolCallCount;return t===0?PASS:fail(`expected no tool calls, got ${t}`,{toolCallCount:t})}}}function maxToolCalls(e){return{name:`maxToolCalls(${e})`,evaluate(t){let n=t.derived.toolCallCount;return n<=e?PASS:fail(`expected at most ${e} tool calls, got ${n}`,{maxAllowed:e,toolCallCount:n})}}}function noFailedActions(){return{name:`noFailedActions`,evaluate(e){let t=e.events.filter(e=>e.type===`action.result`&&(e.data.status===`failed`||e.data.result.isError===!0));if(t.length===0)return PASS;let n=t.map(e=>e.data.result.kind===`tool-result`?e.data.result.toolName:e.data.result.kind);return fail(`${t.length} failed action(s): ${n.join(`, `)}`)}}}function calledSubagent(e,n={}){return{name:`calledSubagent(${e})`,evaluate(r){let i=r.derived.subagentCalls.filter(t=>t.name===e);return i.filter(e=>subagentCallMatches(e,n)).length>0?PASS:i.length===0?fail(`subagent "${e}" was never called; observed: [${r.derived.subagentCalls.map(e=>e.name).join(`, `)}]`,{observedSubagentCalls:r.derived.subagentCalls}):fail(`subagent "${e}" was called but no call matched the constraints`,{observedSubagentCalls:i})}}}function event(e,t){return{name:`event(${t})`,evaluate(n){return e(n.events)?PASS:fail(`event predicate "${t}" did not hold`)}}}function outputEquals(t){return{name:`outputEquals`,evaluate(n){return deepEquals(n.output,t)?PASS:fail(`output ${truncate(JSON.stringify(n.output))} does not equal expected ${truncate(JSON.stringify(t))}`)}}}function outputMatches(e){return{name:`outputMatches`,async evaluate(t){let n=await e[`~standard`].validate(t.output);return!(`issues`in n)||n.issues===void 0?PASS:fail(`output failed schema validation: ${n.issues.map(e=>e.message).join(`; `)}`)}}}function joinCompletedMessages(e){let t=[];for(let n of e)n.type===`message.completed`&&n.data.message!==null&&t.push(n.data.message);return t.join(`
-`)}function failureDetail(e,t){return t===void 0?e:`${e} (code: ${t})`}function truncate(e,t=200){return e===void 0?`undefined`:e.length<=t?e:`${e.slice(0,t)}…`}export{calledSubagent,calledTool,completed,didNotFail,event,loadedSkill,maxToolCalls,messageIncludes,noFailedActions,notCalledTool,outputEquals,outputMatches,toolOrder,usedNoTools,waiting};
+import{deepEquals,eventMatches,subagentCallMatches,testRegExp,toolCallMatches}from"#evals/match.js";import{LOAD_SKILL_TOOL_NAME}from"#runtime/skills/fragment-context.js";const PASS={score:1},fail=(e,t)=>({score:0,message:e,metadata:t});function succeeded(){return{name:`succeeded`,evaluate(e){let t=runFailure(e);return t===void 0?e.derived.parked?fail(`run parked on ${e.derived.inputRequests.length} unanswered input request(s)`):PASS:t}}}function parked(){return{name:`parked`,evaluate(e){let t=runFailure(e);return t===void 0?e.derived.parked?PASS:fail(`expected the run to park on HITL input; it ended "${e.status}" with no pending requests`):t}}}function messageIncludes(e){return{name:`messageIncludes(${String(e)})`,evaluate(t){let n=joinCompletedMessages(t.events);return(typeof e==`string`?n.includes(e):testRegExp(e,n))?PASS:fail(`assistant messages did not include ${String(e)}; got: ${truncate(n)}`)}}}function calledTool(e,t={}){return validateCount(t.count),{name:`calledTool(${e})`,evaluate(n){let r=n.derived.toolCalls.filter(t=>t.name===e),a=r.filter(e=>toolCallMatches(e,t));if(t.count===void 0?a.length>0:a.length===t.count)return{score:1,metadata:{matchingCalls:a.length}};let o=r.length>0?`observed ${e} calls: ${r.map(e=>truncate(JSON.stringify(e.input))).join(`, `)}`:`observed tools: [${n.derived.toolCalls.map(e=>e.name).join(`, `)}]`;return fail(`${t.count===void 0?`expected a matching call to "${e}"`:`expected exactly ${t.count} matching call(s), found ${a.length}`}; ${o}`)}}}function loadedSkill(e,t={}){return{...calledTool(LOAD_SKILL_TOOL_NAME,{...t,input:{skill:e}}),name:`loadedSkill(${e})`}}function notCalledTool(e){return{name:`notCalledTool(${e})`,evaluate(t){let n=t.derived.toolCalls.filter(t=>t.name===e).length;return n===0?PASS:fail(`"${e}" was called ${n} time(s)`)}}}function toolOrder(e){return{name:`toolOrder(${e.join(` → `)})`,evaluate(t){let n=requestedTools(t.events).map(e=>e.name),r=missingOrderedName(e,n);return r===void 0?PASS:fail(`missing "${r.name}" after [${e.slice(0,r.cursor).join(`, `)}]; observed request order: [${n.join(`, `)}]`)}}}function usedNoTools(){return{name:`usedNoTools`,evaluate(e){let t=e.derived.toolCallCount;return t===0?PASS:fail(`expected no tool calls, got ${t}`,{toolCallCount:t})}}}function maxToolCalls(e){return validateCount(e),{name:`maxToolCalls(${e})`,evaluate(t){let n=t.derived.toolCallCount;return n<=e?PASS:fail(`expected at most ${e} tool calls, got ${n}`,{maxAllowed:e,toolCallCount:n})}}}function noFailedActions(){return{name:`noFailedActions`,evaluate(e){let t=e.events.filter(e=>e.type===`action.result`&&(e.data.status===`failed`||e.data.result.isError===!0));if(t.length===0)return PASS;let n=t.map(e=>e.data.result.kind===`tool-result`?e.data.result.toolName:e.data.result.kind);return fail(`${t.length} failed action(s): ${n.join(`, `)}`)}}}function calledSubagent(e,t={}){return validateCount(t.count),{name:`calledSubagent(${e})`,evaluate(r){let i=r.derived.subagentCalls.filter(t=>t.name===e),a=i.filter(e=>subagentCallMatches(e,t));return(t.count===void 0?a.length>0:a.length===t.count)?{score:1,metadata:{matchingCalls:a.length}}:i.length===0?fail(`subagent "${e}" was never called; observed: [${r.derived.subagentCalls.map(e=>e.name).join(`, `)}]`,{observedSubagentCalls:r.derived.subagentCalls}):fail(`subagent "${e}" was called but ${a.length} call(s) matched the constraints`,{observedSubagentCalls:i})}}}function eventsSatisfy(e,t){return{name:`eventsSatisfy(${e})`,evaluate(n){return t(n.events)?PASS:fail(`event predicate "${e}" did not hold`)}}}function typedEvent(e){return validateCount(e.count),{name:`event(${e.type})`,evaluate(n){let r=n.events.filter(n=>eventMatches(n,e));return(e.count===void 0?r.length>0:r.length===e.count)?{score:1,metadata:{matchingEvents:r.length}}:fail(`expected ${e.count===void 0?`at least one`:`exactly ${e.count}`} matching ${e.type} event(s), found ${r.length}; observed: [${n.events.map(e=>e.type).join(`, `)}]`)}}}function notEvent(e){return{name:`notEvent(${e.type})`,evaluate(n){let r=n.events.filter(n=>eventMatches(n,e));return r.length===0?PASS:fail(`expected no matching ${e.type} events, found ${r.length}`)}}}function eventOrder(e){for(let t of e)validateCount(t.count);return{name:`eventOrder(${e.map(e=>e.type).join(` → `)})`,evaluate(n){let r=-1;for(let i of e){let e=n.events.flatMap((e,n)=>eventMatches(e,i)?[n]:[]);if(!(i.count===void 0?e.length>0:e.length===i.count))return fail(`expected ${i.count===void 0?`at least one`:`exactly ${i.count}`} matching ${i.type} event(s), found ${e.length}`);let a=e[0];if(a===void 0||a<=r)return fail(`event group ${i.type} did not occur after the previous group`);r=e[e.length-1]??r}return PASS}}}function outputEquals(t){return{name:`outputEquals`,evaluate(n){return deepEquals(n.output,t)?PASS:fail(`output ${truncate(JSON.stringify(n.output))} does not equal expected ${truncate(JSON.stringify(t))}`)}}}function outputMatches(e){return{name:`outputMatches`,async evaluate(t){let n=await e[`~standard`].validate(t.output);return!(`issues`in n)||n.issues===void 0?PASS:fail(`output failed schema validation: ${n.issues.map(e=>e.message).join(`; `)}`)}}}function joinCompletedMessages(e){let t=[];for(let n of e)n.type===`message.completed`&&n.data.message!==null&&t.push(n.data.message);return t.join(`
+`)}function failureDetail(e,t){return t===void 0?e:`${e} (code: ${t})`}function runFailure(e){if(e.status===`failed`)return fail(failureDetail(`run failed`,e.derived.failureCode));let t=e.events.find(e=>e.type===`turn.failed`||e.type===`step.failed`);return t===void 0?void 0:fail(`${t.type} (${t.data.code}): ${t.data.message}`)}function truncate(e,t=200){return e===void 0?`undefined`:e.length<=t?e:`${e.slice(0,t)}…`}function validateCount(e){if(e!==void 0&&(!Number.isInteger(e)||e<0))throw TypeError(`Assertion count must be a non-negative integer; received ${e}.`)}function missingOrderedName(e,t){let n=0;for(let r of t)if(r===e[n]&&(n+=1),n===e.length)return;let r=e[n];return r===void 0?void 0:{cursor:n,name:r}}function requestedTools(e){let t=[],n=new Set,append=(e,r)=>{n.has(e)||(n.add(e),t.push({callId:e,name:r}))};for(let t of e)if(t.type===`actions.requested`)for(let e of t.data.actions)e.kind===`tool-call`&&append(e.callId,e.toolName);else if(t.type===`input.requested`)for(let e of t.data.requests){let{action:t}=e;t.kind===`tool-call`&&append(t.callId,t.toolName)}return t}export{calledSubagent,calledTool,eventOrder,eventsSatisfy,loadedSkill,maxToolCalls,messageIncludes,noFailedActions,notCalledTool,notEvent,outputEquals,outputMatches,parked,succeeded,toolOrder,typedEvent,usedNoTools};

package/dist/src/evals/assertions/scoped.d.ts

@@ -0,0 +1,17 @@
+import { AssertionCollector } from "#evals/assertions/collector.js";
+import type { EveEvalAssertionSubject } from "#evals/assertions/run.js";
+import type { EveEvalAssertions, EveEvalOutputAssertions, EveEvalTaskResult } from "#evals/types.js";
+type AssertionScope = {
+    readonly timing: "final";
+    select(result: EveEvalTaskResult): EveEvalAssertionSubject;
+} | {
+    readonly timing: "snapshot";
+    select(): EveEvalAssertionSubject;
+};
+/** Binds the shared assertion vocabulary to one aggregate, session, or turn scope. */
+export declare function createScopedAssertions(collector: AssertionCollector, scope: AssertionScope): EveEvalAssertions;
+/** Binds output-only assertions to a session or immutable turn snapshot. */
+export declare function createOutputAssertions(collector: AssertionCollector, scope: Extract<AssertionScope, {
+    timing: "snapshot";
+}>): EveEvalOutputAssertions;
+export {};

package/dist/src/evals/assertions/scoped.js

@@ -0,0 +1 @@
+import"#evals/assertions/collector.js";import*as RunAssertions from"#evals/assertions/run.js";function createScopedAssertions(t,n){let r=createRecorder(t,n);return{succeeded:()=>r(RunAssertions.succeeded()),parked:()=>r(RunAssertions.parked()),messageIncludes:t=>r(RunAssertions.messageIncludes(t)),calledTool:(t,n)=>r(RunAssertions.calledTool(t,n)),loadedSkill:(t,n)=>r(RunAssertions.loadedSkill(t,n)),notCalledTool:t=>r(RunAssertions.notCalledTool(t)),toolOrder:t=>r(RunAssertions.toolOrder(t)),usedNoTools:()=>r(RunAssertions.usedNoTools()),maxToolCalls:t=>r(RunAssertions.maxToolCalls(t)),calledSubagent:(t,n)=>r(RunAssertions.calledSubagent(t,n)),noFailedActions:()=>r(RunAssertions.noFailedActions()),event:(t,n)=>r(RunAssertions.typedEvent({...n,type:t})),notEvent:(t,n)=>r(RunAssertions.notEvent({...n,type:t})),eventOrder:t=>r(RunAssertions.eventOrder(t)),eventsSatisfy:(t,n)=>r(RunAssertions.eventsSatisfy(t,n))}}function createOutputAssertions(t,n){let r=createRecorder(t,n);return{outputEquals:t=>r(RunAssertions.outputEquals(t)),outputMatches:t=>r(RunAssertions.outputMatches(t))}}function createRecorder(e,t){return t.timing===`final`?n=>e.recordScoped(n,t.select):n=>{let r=t.select();return e.recordScoped(n,()=>r)}}export{createOutputAssertions,createScopedAssertions};

package/dist/src/evals/context.d.ts

@@ -9,6 +9,7 @@ import type { EveEvalContext, EveEvalJudgeConfig, EveEvalTargetHandle } from "#e
  */
 export declare function createEvalContext(deps: {
     readonly manager: EvalSessionManager;
+    readonly collector: AssertionCollector;
     readonly target: EveEvalTargetHandle;
     readonly signal: AbortSignal;
     readonly judge: EveEvalJudgeConfig | undefined;

package/dist/src/evals/context.js

@@ -1 +1 @@
-import"#evals/session.js";import{AssertionCollector}from"#evals/assertions/collector.js";import*as RunAssertions from"#evals/assertions/run.js";import{buildJudgeContext}from"#evals/judge.js";function createEvalContext(r){let i=new AssertionCollector,a=``,primary=()=>r.manager.primary,replyMessage=()=>r.manager.lastTurnSession()?.lastTurn?.message??null,o=buildJudgeContext({collector:i,getReply:replyMessage,getInput:()=>a,judge:r.judge});return{context:{get events(){return primary().events},get pendingInputRequests(){return primary().pendingInputRequests},get state(){return primary().state},get sessionId(){return primary().sessionId},expectInputRequests:e=>primary().expectInputRequests(e),respond:(...e)=>primary().respond(...e),respondAll:e=>primary().respondAll(e),send:e=>(a=promptText(e),primary().send(e)),sendFile:(e,t,n)=>(a=e,primary().sendFile(e,t,n)),signal:r.signal,target:r.target,get reply(){return replyMessage()},log:r.log,sleep:e=>sleep(e,r.signal),newSession:()=>r.manager.newSession(),completed:()=>i.recordRun(RunAssertions.completed()),didNotFail:()=>i.recordRun(RunAssertions.didNotFail()),waiting:()=>i.recordRun(RunAssertions.waiting()),messageIncludes:e=>i.recordRun(RunAssertions.messageIncludes(e)),calledTool:(e,n)=>i.recordRun(RunAssertions.calledTool(e,n)),loadedSkill:(e,n)=>i.recordRun(RunAssertions.loadedSkill(e,n)),notCalledTool:e=>i.recordRun(RunAssertions.notCalledTool(e)),toolOrder:e=>i.recordRun(RunAssertions.toolOrder(e)),usedNoTools:()=>i.recordRun(RunAssertions.usedNoTools()),maxToolCalls:e=>i.recordRun(RunAssertions.maxToolCalls(e)),calledSubagent:(e,n)=>i.recordRun(RunAssertions.calledSubagent(e,n)),noFailedActions:()=>i.recordRun(RunAssertions.noFailedActions()),event:(e,n)=>i.recordRun(RunAssertions.event(e,n)),outputEquals:e=>i.recordRun(RunAssertions.outputEquals(e)),outputMatches:e=>i.recordRun(RunAssertions.outputMatches(e)),check:(e,t)=>recordCheck(i,e,t),judge:o},collector:i}}function recordCheck(e,t,n){return e.recordValue({name:n.name,severity:n.severity,threshold:n.threshold,score:async()=>({score:await n.score(t)})})}function promptText(e){if(typeof e==`string`)return e;let t=e.message;return typeof t==`string`?t:``}function sleep(e=1e3,t){if(!Number.isFinite(e)||e<0)throw Error(`sleep() duration must be a non-negative finite number.`);return t?.aborted?Promise.reject(t.reason):new Promise((n,r)=>{let i=setTimeout(n,e);t?.addEventListener(`abort`,()=>{clearTimeout(i),r(t.reason)},{once:!0})})}export{createEvalContext};
+import"#evals/assertions/collector.js";import"#evals/session.js";import{createScopedAssertions}from"#evals/assertions/scoped.js";import{buildJudgeContext}from"#evals/judge.js";import{EvalRequirementFailed,EvalSkipped}from"#evals/control-flow.js";function createEvalContext(n){let i=n.collector,a=``,primary=()=>n.manager.primary,replyMessage=()=>n.manager.lastTurnSession()?.lastTurn?.message??null,o=buildJudgeContext({collector:i,getReply:replyMessage,getInput:()=>a,judge:n.judge});return{context:{get events(){return primary().events},get pendingInputRequests(){return primary().pendingInputRequests},get state(){return primary().state},get sessionId(){return primary().sessionId},requireInputRequest:e=>primary().requireInputRequest(e),respond:(...e)=>primary().respond(...e),respondAll:e=>primary().respondAll(e),send:e=>(a=promptText(e),primary().send(e)),sendFile:(e,t,n)=>(a=e,primary().sendFile(e,t,n)),signal:n.signal,target:n.target,get reply(){return replyMessage()},log:n.log,sleep:e=>sleep(e,n.signal),newSession:()=>n.manager.newSession(),...createScopedAssertions(i,{timing:`final`,select:e=>e}),check:(e,t)=>recordCheck(i,e,t),require:(e,t)=>requireCheck(i,e,t),skip:e=>{throw e.trim().length===0?Error(`skip() requires a non-empty reason.`):i.hasEntries||n.manager.hasActivity()?Error(`skip() must be called before sending messages or recording assertions.`):new EvalSkipped(e)},judge:o},collector:i}}async function requireCheck(e,t,r){let i=r.gate(r.threshold);if(!await e.recordRequirement({name:i.name,threshold:i.threshold,score:async()=>({score:await i.score(t)})}))throw new EvalRequirementFailed;return t}function recordCheck(e,t,n){return e.recordValue({name:n.name,severity:n.severity,threshold:n.threshold,score:async()=>({score:await n.score(t)})})}function promptText(e){if(typeof e==`string`)return e;let t=e.message;return typeof t==`string`?t:``}function sleep(e=1e3,t){if(!Number.isFinite(e)||e<0)throw Error(`sleep() duration must be a non-negative finite number.`);return t?.aborted?Promise.reject(t.reason):new Promise((n,r)=>{let i=setTimeout(n,e);t?.addEventListener(`abort`,()=>{clearTimeout(i),r(t.reason)},{once:!0})})}export{createEvalContext};

package/dist/src/evals/control-flow.d.ts

@@ -0,0 +1,9 @@
+/** Internal signal used to stop an eval after a recorded required assertion fails. */
+export declare class EvalRequirementFailed extends Error {
+    constructor();
+}
+/** Internal signal used by `t.skip()` to produce a skipped verdict. */
+export declare class EvalSkipped extends Error {
+    readonly reason: string;
+    constructor(reason: string);
+}

package/dist/src/evals/control-flow.js

@@ -0,0 +1 @@
+var EvalRequirementFailed=class extends Error{constructor(){super(`A required eval assertion failed.`),this.name=`EvalRequirementFailed`}},EvalSkipped=class extends Error{reason;constructor(e){super(e),this.name=`EvalSkipped`,this.reason=e}};export{EvalRequirementFailed,EvalSkipped};

package/dist/src/evals/define-eval.d.ts

@@ -2,7 +2,7 @@ import type { EveEvalDefinition, EveEvalInput } from "#evals/types.js";
 /**
  * Defines one eve eval. Each eval file is exactly one case: an imperative
  * `test(t)` function that drives the agent (`t.send`, `t.respond`, …) and
- * asserts on what it produced (`t.completed()`, `t.check(...)`,
+ * asserts on what it produced (`t.succeeded()`, `t.check(...)`,
  * `t.judge.autoevals.*`). Organize related evals with directory nesting under
  * `evals/`, or default-export an array of evals to fan one file out over a
  * dataset.

package/dist/src/evals/define-eval.js

@@ -1 +1 @@
-function defineEval(e){return validateEvalInput(e),{...e,_tag:`EveEval`}}function validateEvalInput(e){if(`id`in e)throw Error("Eval must not specify `id`. Eval identity is derived from the file path under evals/.");if(`name`in e)throw Error("Eval must not specify `name`. Eval identity is derived from the file path under evals/.");if(rejectLegacyKey(e,`input`,"Send the prompt inside `test`: `async test(t) { await t.send(...) }`."),rejectLegacyKey(e,`run`,"Rename `run` to `test`; it receives the same context `t`."),rejectLegacyKey(e,`checks`,"Assert inline inside `test` (e.g. `t.completed()`, `t.calledTool(...)`)."),rejectLegacyKey(e,`scores`,"Use soft assertions inside `test`: `t.check(...).atLeast(n)` or `t.judge.autoevals.*`."),rejectLegacyKey(e,`expected`,"Pass the reference value to the assertion (e.g. `t.check(t.reply, includes(value))`)."),rejectLegacyKey(e,`thresholds`,"Put the threshold on the assertion: `.atLeast(n)`."),rejectLegacyKey(e,`parseOutput`,"Read the value you want inside `test` and assert on it directly."),rejectLegacyKey(e,`model`,"Rename `model` to `judge: { model }`."),rejectLegacyKey(e,`modelOptions`,"Move it under `judge: { model, modelOptions }`."),rejectLegacyKey(e,`cases`,"Each eval file is one case; default-export an array of `defineEval(...)` for datasets."),rejectLegacyKey(e,`requires`,`Point real-model evals at credentialed targets directly; dev-only routes are enforced from the live target.`),typeof e.test!=`function`)throw Error("Eval requires a `test(t)` function.");if(e.timeoutMs!==void 0&&(e.timeoutMs<0||!Number.isFinite(e.timeoutMs)))throw Error("Eval `timeoutMs` must be a non-negative finite number.")}function rejectLegacyKey(e,t,n){if(t in e)throw Error(`Eval \`${t}\` is no longer supported. ${n}`)}export{defineEval};
+function defineEval(e){return validateEvalInput(e),{...e,_tag:`EveEval`}}function validateEvalInput(e){if(`id`in e)throw Error("Eval must not specify `id`. Eval identity is derived from the file path under evals/.");if(`name`in e)throw Error("Eval must not specify `name`. Eval identity is derived from the file path under evals/.");if(rejectLegacyKey(e,`input`,"Send the prompt inside `test`: `async test(t) { await t.send(...) }`."),rejectLegacyKey(e,`run`,"Rename `run` to `test`; it receives the same context `t`."),rejectLegacyKey(e,`checks`,"Assert inline inside `test` (e.g. `t.succeeded()`, `t.calledTool(...)`)."),rejectLegacyKey(e,`scores`,"Use soft assertions inside `test`: `t.check(...).atLeast(n)` or `t.judge.autoevals.*`."),rejectLegacyKey(e,`expected`,"Pass the reference value to the assertion (e.g. `t.check(t.reply, includes(value))`)."),rejectLegacyKey(e,`thresholds`,"Put the threshold on the assertion: `.atLeast(n)`."),rejectLegacyKey(e,`parseOutput`,"Read the value you want inside `test` and assert on it directly."),rejectLegacyKey(e,`model`,"Rename `model` to `judge: { model }`."),rejectLegacyKey(e,`modelOptions`,"Move it under `judge: { model, modelOptions }`."),rejectLegacyKey(e,`cases`,"Each eval file is one case; default-export an array of `defineEval(...)` for datasets."),rejectLegacyKey(e,`requires`,`Point real-model evals at credentialed targets directly; dev-only routes are enforced from the live target.`),typeof e.test!=`function`)throw Error("Eval requires a `test(t)` function.");if(e.timeoutMs!==void 0&&(e.timeoutMs<0||!Number.isFinite(e.timeoutMs)))throw Error("Eval `timeoutMs` must be a non-negative finite number.")}function rejectLegacyKey(e,t,n){if(t in e)throw Error(`Eval \`${t}\` is no longer supported. ${n}`)}export{defineEval};

package/dist/src/evals/expect/index.d.ts

@@ -2,10 +2,12 @@ import type { StandardSchemaV1 } from "#compiled/@standard-schema/spec/index.js"
 import type { Assertion } from "#evals/types.js";
 export type { Assertion, AssertionHandle, AssertionSeverity } from "#evals/types.js";
 /**
- * Passes when the value (coerced to a string) contains `substring`. A hard
- * gate by default. Apply with `t.check(value, includes("..."))`.
+ * Passes when the value (coerced to a string) contains a substring or matches
+ * a RegExp. A hard gate by default.
  */
-export declare function includes(substring: string): Assertion;
+export declare function includes(substring: string | RegExp): Assertion;
+/** Passes when `predicate` returns true for the supplied value. */
+export declare function satisfies<T = unknown>(predicate: (value: T) => boolean, label: string): Assertion;
 /**
  * Passes when the value deep-equals `expected` (exact structural equality).
  * A hard gate by default.

package/dist/src/evals/expect/index.js

@@ -1 +1 @@
-import{Levenshtein}from"../../node_modules/.pnpm/autoevals@0.0.132_ws@8.21.0/node_modules/autoevals/jsdist/index.js";import{deepEquals}from"#evals/match.js";function makeAssertion(e){return{name:e.name,severity:e.severity,threshold:e.threshold,score:e.score,gate(t){return makeAssertion({...e,severity:`gate`,threshold:t})},soft(t){return makeAssertion({...e,severity:`soft`,threshold:t})},atLeast(t){return makeAssertion({...e,severity:`soft`,threshold:t})}}}function includes(e){return makeAssertion({name:`includes(${e})`,severity:`gate`,score:t=>+!!String(t??``).includes(e)})}function equals(e){return makeAssertion({name:`equals`,severity:`gate`,score:n=>+!!deepEquals(n,e)})}function matches(e){return makeAssertion({name:`matches`,severity:`gate`,score:async t=>{let n=await e[`~standard`].validate(t);return+(!(`issues`in n)||n.issues===void 0)}})}function similarity(t){return makeAssertion({name:`similarity`,severity:`soft`,score:async n=>(await Levenshtein({output:String(n??``),expected:t})).score??0})}export{equals,includes,matches,similarity};
+import{Levenshtein}from"../../node_modules/.pnpm/autoevals@0.0.132_ws@8.21.0/node_modules/autoevals/jsdist/index.js";import{deepEquals,testRegExp}from"#evals/match.js";function makeAssertion(e){return{name:e.name,severity:e.severity,threshold:e.threshold,score:e.score,gate(t){return makeAssertion({...e,severity:`gate`,threshold:t})},soft(t){return makeAssertion({...e,severity:`soft`,threshold:t})},atLeast(t){return makeAssertion({...e,severity:`soft`,threshold:t})}}}function includes(e){return makeAssertion({name:`includes(${e})`,severity:`gate`,score:t=>{let r=String(t??``);return(typeof e==`string`?r.includes(e):testRegExp(e,r))?1:0}})}function satisfies(e,t){if(t.trim().length===0)throw Error(`satisfies() requires a non-empty label.`);return makeAssertion({name:`satisfies(${t})`,severity:`gate`,score:t=>+!!e(t)})}function equals(e){return makeAssertion({name:`equals`,severity:`gate`,score:n=>+!!deepEquals(n,e)})}function matches(e){return makeAssertion({name:`matches`,severity:`gate`,score:async t=>{let n=await e[`~standard`].validate(t);return+(!(`issues`in n)||n.issues===void 0)}})}function similarity(t){return makeAssertion({name:`similarity`,severity:`soft`,score:async n=>(await Levenshtein({output:String(n??``),expected:t})).score??0})}export{equals,includes,matches,satisfies,similarity};

package/dist/src/evals/index.d.ts

@@ -3,5 +3,5 @@ export { defineEvalConfig } from "#evals/define-eval-config.js";
 export { EveEvalTurnFailedError } from "#evals/session.js";
 export type { RuntimeIdentity } from "#protocol/message.js";
 export type { InputRequest } from "#runtime/input/types.js";
-export type { EveEvalValueMatcher, EveEvalToolCallMatchOptions, EveEvalSkillLoadMatchOptions, EveEvalSubagentCallMatchOptions, } from "#evals/match.js";
-export type { Assertion, AssertionHandle, AssertionResult, AssertionSeverity, AutoevalsJudges, EveEvalContext, EveEvalDerivedFacts, EveEvalJudgeConfig, EveEvalRunSummary, EveEvalSession, EveEvalSessionResult, EveEvalScheduleDispatchResult, EveEvalSubagentCall, EveEval, EveEvalConfig, EveEvalConfigInput, EveEvalDefinition, EveEvalInput, EveEvalResult, EveEvalTarget, EveEvalTargetCapabilities, EveEvalTargetHandle, EveEvalTaskResult, EveEvalToolCall, EveEvalTurn, EveEvalVerdict, JudgeContext, JudgeOpts, } from "#evals/types.js";
+export type { EveEvalEventMatch, EveEvalInputRequestMatchOptions, EveEvalValueMatcher, EveEvalToolCallMatchOptions, EveEvalSkillLoadMatchOptions, EveEvalSubagentCallMatchOptions, } from "#evals/match.js";
+export type { Assertion, AssertionHandle, AssertionResult, AssertionSeverity, AutoevalsJudges, EveEvalActionStatus, EveEvalAssertions, EveEvalContext, EveEvalDerivedFacts, EveEvalJudgeConfig, EveEvalRunSummary, EveEvalSession, EveEvalSessionResult, EveEvalScheduleDispatchResult, EveEvalSubagentCall, EveEval, EveEvalConfig, EveEvalConfigInput, EveEvalDefinition, EveEvalInput, EveEvalResult, EveEvalTarget, EveEvalTargetCapabilities, EveEvalTargetHandle, EveEvalTaskResult, EveEvalToolCall, EveEvalTurn, EveEvalVerdict, JudgeContext, JudgeOpts, } from "#evals/types.js";

package/dist/src/evals/match.d.ts

@@ -1,3 +1,6 @@
+import type { HandleMessageStreamEvent } from "#protocol/message.js";
+import type { InputRequest } from "#runtime/input/types.js";
+import type { JsonObject, JsonValue } from "#shared/json.js";
 import type { EveEvalSubagentCall, EveEvalToolCall } from "#evals/types.js";
 /**
  * One matcher accepted by the assertion options (`t.calledTool`,
@@ -9,26 +12,25 @@ import type { EveEvalSubagentCall, EveEvalToolCall } from "#evals/types.js";
  *   compare with `Object.is`
  * - a **RegExp** tests string values directly and the JSON serialization of
  *   anything else
- * - a **function** receives the observed value and either returns a boolean
- *   verdict, or returns a resolved expected value that is then compared like a
- *   literal — e.g. `(o) => o === process.env.EVE_WEATHER_AGENT_HOST`. To assert
- *   a literal boolean field, use the literal directly; boolean returns are
- *   always treated as verdicts.
+ * - a **function** receives the observed value and returns a boolean verdict
  */
-export type EveEvalValueMatcher<T = unknown> = T | RegExp | ((value: T) => unknown);
+export type EveEvalValueMatcher<T = JsonValue | undefined> = EveEvalDeepMatcher<T>;
+type EveEvalDeepMatcher<T> = RegExp | ((value: T) => boolean) | (T extends readonly (infer TEntry)[] ? readonly EveEvalDeepMatcher<TEntry>[] : T extends object ? {
+    readonly [K in keyof T]?: EveEvalDeepMatcher<T[K]>;
+} : T);
 /**
  * Constraints applied to tool calls by `t.calledTool`. All provided
  * constraints must hold for a call to match.
  */
 export interface EveEvalToolCallMatchOptions {
     /** Partial-deep matcher over the call input (see {@link EveEvalValueMatcher}). */
-    readonly input?: EveEvalValueMatcher;
+    readonly input?: EveEvalValueMatcher<JsonObject>;
     /** Matcher over the call output. */
     readonly output?: EveEvalValueMatcher;
-    /** Required error state of matching calls. */
-    readonly isError?: boolean;
+    /** Required lifecycle outcome. Defaults to `"completed"`. */
+    readonly status?: EveEvalToolCall["status"];
     /** Exact number of matching calls required. Defaults to "at least one". */
-    readonly times?: number;
+    readonly count?: number;
 }
 /**
  * Constraints applied to a `load_skill` call by `t.loadedSkill`. Identical to
@@ -41,10 +43,40 @@ export type EveEvalSkillLoadMatchOptions = Omit<EveEvalToolCallMatchOptions, "in
  */
 export interface EveEvalSubagentCallMatchOptions {
     /** Matcher over the `subagent.called` remote URL. */
-    readonly remoteUrl?: EveEvalValueMatcher;
+    readonly remoteUrl?: EveEvalValueMatcher<string | undefined>;
     /** Matcher over the `subagent.completed` output. */
     readonly output?: EveEvalValueMatcher;
+    /** Required lifecycle outcome. Defaults to `"completed"`. */
+    readonly status?: EveEvalSubagentCall["status"];
+    /** Exact number of matching delegations required. Defaults to "at least one". */
+    readonly count?: number;
 }
+/** Constraints accepted by `requireInputRequest`. */
+export interface EveEvalInputRequestMatchOptions {
+    /** Matcher over the request's display hint. */
+    readonly display?: EveEvalValueMatcher<InputRequest["display"]>;
+    /** Partial-deep matcher over a tool-call action's input. */
+    readonly input?: EveEvalValueMatcher<JsonObject>;
+    /** Matcher over the complete option-id list in request order. */
+    readonly optionIds?: EveEvalValueMatcher<readonly string[]>;
+    /** Matcher over the request prompt. */
+    readonly prompt?: EveEvalValueMatcher<string>;
+    /** Required tool name for a tool-call action. */
+    readonly toolName?: string;
+}
+/** One typed stream-event matcher used by scoped event assertions. */
+export type EveEvalEventMatch<TType extends HandleMessageStreamEvent["type"] = HandleMessageStreamEvent["type"]> = TType extends HandleMessageStreamEvent["type"] ? {
+    /** Stream-event type to match. */
+    readonly type: TType;
+    /** Partial-deep matcher over the event data. */
+    readonly data?: EveEvalDeepMatcher<Extract<HandleMessageStreamEvent, {
+        type: TType;
+    }> extends {
+        data: infer TData;
+    } ? TData : never>;
+    /** Exact number of matching events required. Defaults to "at least one". */
+    readonly count?: number;
+} : never;
 /**
  * Returns true when the observed value satisfies a matcher (literal, RegExp,
  * or function — see {@link EveEvalValueMatcher}).
@@ -52,7 +84,7 @@ export interface EveEvalSubagentCallMatchOptions {
 export declare function matchesValue(matcher: unknown, value: unknown): boolean;
 /**
  * Returns true when one derived tool call satisfies the `input`/`output`/
- * `isError` constraints (the `times` count is the caller's concern).
+ * lifecycle constraints (the `count` option is the caller's concern).
  */
 export declare function toolCallMatches(call: EveEvalToolCall, options: EveEvalToolCallMatchOptions): boolean;
 /**
@@ -60,8 +92,12 @@ export declare function toolCallMatches(call: EveEvalToolCall, options: EveEvalT
  * `output` constraints.
  */
 export declare function subagentCallMatches(call: EveEvalSubagentCall, options: EveEvalSubagentCallMatchOptions): boolean;
+/** Returns true when one HITL request satisfies every supplied constraint. */
+export declare function inputRequestMatches(request: InputRequest, options: EveEvalInputRequestMatchOptions): boolean;
+/** Returns true when one stream event satisfies a typed event matcher. */
+export declare function eventMatches(event: HandleMessageStreamEvent, matcher: EveEvalEventMatch): boolean;
 /**
- * Strict structural equality used by `t.outputEquals`: unlike matcher
+ * Strict structural equality used by scoped `outputEquals`: unlike matcher
  * comparison, objects must carry exactly the same keys on both sides.
  */
 export declare function deepEquals(a: unknown, b: unknown): boolean;
@@ -71,3 +107,4 @@ export declare function deepEquals(a: unknown, b: unknown): boolean;
  * a `g`/`y`-flagged pattern would otherwise return order-dependent results.
  */
 export declare function testRegExp(pattern: RegExp, text: string): boolean;
+export {};

package/dist/src/evals/match.js

@@ -1 +1 @@
-function matchesValue(e,t){if(e instanceof RegExp)return testRegExpAgainst(e,t);if(typeof e==`function`){let n=e(t);return typeof n==`boolean`?n:typeof n==`function`?Object.is(n,t):matchesValue(n,t)}return Array.isArray(e)?!Array.isArray(t)||t.length!==e.length?!1:e.every((e,n)=>matchesValue(e,t[n])):isPlainObject(e)?isPlainObject(t)?Object.entries(e).every(([e,n])=>matchesValue(n,t[e])):!1:Object.is(e,t)}function toolCallMatches(e,t){return!(t.input!==void 0&&!matchesValue(t.input,e.input)||t.output!==void 0&&!matchesValue(t.output,e.output)||t.isError!==void 0&&e.isError!==t.isError)}function subagentCallMatches(e,t){return!(t.remoteUrl!==void 0&&!matchesValue(t.remoteUrl,e.remoteUrl)||t.output!==void 0&&!matchesValue(t.output,e.output))}function deepEquals(e,t){if(Object.is(e,t))return!0;if(Array.isArray(e)||Array.isArray(t))return!Array.isArray(e)||!Array.isArray(t)||e.length!==t.length?!1:e.every((e,n)=>deepEquals(e,t[n]));if(isPlainObject(e)&&isPlainObject(t)){let n=Object.keys(e),r=Object.keys(t);return n.length===r.length?n.every(n=>deepEquals(e[n],t[n])):!1}return!1}function testRegExpAgainst(e,t){if(typeof t==`string`)return testRegExp(e,t);if(t===void 0)return!1;let n=JSON.stringify(t);return n!==void 0&&testRegExp(e,n)}function testRegExp(e,t){return(e.global||e.sticky)&&(e.lastIndex=0),e.test(t)}function isPlainObject(e){return typeof e==`object`&&!!e&&!Array.isArray(e)}export{deepEquals,matchesValue,subagentCallMatches,testRegExp,toolCallMatches};
+function matchesValue(e,t){return e instanceof RegExp?testRegExpAgainst(e,t):typeof e==`function`?e(t):Array.isArray(e)?!Array.isArray(t)||t.length!==e.length?!1:e.every((e,n)=>matchesValue(e,t[n])):isPlainObject(e)?isPlainObject(t)?Object.entries(e).every(([e,n])=>matchesValue(n,t[e])):!1:Object.is(e,t)}function toolCallMatches(e,t){return!(t.input!==void 0&&!matchesValue(t.input,e.input)||t.output!==void 0&&!matchesValue(t.output,e.output)||e.status!==(t.status??`completed`))}function subagentCallMatches(e,t){return!(t.remoteUrl!==void 0&&!matchesValue(t.remoteUrl,e.remoteUrl)||t.output!==void 0&&!matchesValue(t.output,e.output)||e.status!==(t.status??`completed`))}function inputRequestMatches(e,t){if(t.display!==void 0&&!matchesValue(t.display,e.display)||t.prompt!==void 0&&!matchesValue(t.prompt,e.prompt))return!1;if(t.optionIds!==void 0){let n=(e.options??[]).map(e=>e.id);if(!matchesValue(t.optionIds,n))return!1}return!(t.toolName!==void 0&&(e.action.kind!==`tool-call`||e.action.toolName!==t.toolName)||t.input!==void 0&&(e.action.kind!==`tool-call`||!matchesValue(t.input,e.action.input)))}function eventMatches(e,t){return e.type===t.type&&(t.data===void 0||matchesValue(t.data,`data`in e?e.data:void 0))}function deepEquals(e,t){if(Object.is(e,t))return!0;if(Array.isArray(e)||Array.isArray(t))return!Array.isArray(e)||!Array.isArray(t)||e.length!==t.length?!1:e.every((e,n)=>deepEquals(e,t[n]));if(isPlainObject(e)&&isPlainObject(t)){let n=Object.keys(e),r=Object.keys(t);return n.length===r.length?n.every(n=>deepEquals(e[n],t[n])):!1}return!1}function testRegExpAgainst(e,t){if(typeof t==`string`)return testRegExp(e,t);if(t===void 0)return!1;let n=JSON.stringify(t);return n!==void 0&&testRegExp(e,n)}function testRegExp(e,t){return(e.global||e.sticky)&&(e.lastIndex=0),e.test(t)}function isPlainObject(e){return typeof e==`object`&&!!e&&!Array.isArray(e)}export{deepEquals,eventMatches,inputRequestMatches,matchesValue,subagentCallMatches,testRegExp,toolCallMatches};

package/dist/src/evals/runner/artifacts.js

@@ -1,3 +1,3 @@
 import{dirname,join}from"node:path";import{mkdir,writeFile}from"node:fs/promises";function resolveArtifactDirectory(e){return join(e,`.eve`,`evals`,new Date().toISOString().replace(/[:.]/g,`-`).slice(0,19))}async function writeArtifacts(t,n){let r=join(t,`evals`);await mkdir(r,{recursive:!0}),await writeFile(join(t,`summary.json`),JSON.stringify(buildSummaryArtifact(n),null,2));let i=n.results.map(e=>JSON.stringify(buildResultLine(e))).join(`
 `);await writeFile(join(t,`results.jsonl`),`${i}\n`),await Promise.all(n.results.map(async t=>{let n=join(r,`${sanitizeArtifactPath(t.id)}.json`);await mkdir(dirname(n),{recursive:!0}),await writeFile(n,JSON.stringify(buildEvalArtifact(t),null,2));let i=t.result.events.map(e=>JSON.stringify(e)).join(`
-`);await writeFile(join(r,`${sanitizeArtifactPath(t.id)}.events.ndjson`),`${i}\n`)}))}function buildSummaryArtifact(e){return{target:e.target,startedAt:e.startedAt,completedAt:e.completedAt,passed:e.passed,failed:e.failed,scored:e.scored,errored:e.errored,totalEvals:e.results.length,evals:e.results.map(e=>({id:e.id,verdict:e.verdict,status:e.result.status,assertions:e.assertions.map(e=>({name:e.name,score:e.score,severity:e.severity,passed:e.passed})),error:e.error}))}}function buildResultLine(e){return{id:e.id,verdict:e.verdict,status:e.result.status,output:e.result.output,assertions:e.assertions,error:e.error}}function buildEvalArtifact(e){return{id:e.id,result:{output:e.result.output,finalMessage:e.result.finalMessage,sessionId:e.result.sessionId,status:e.result.status,logs:e.result.logs,derived:e.result.derived,sessions:e.result.sessions},verdict:e.verdict,assertions:e.assertions,error:e.error}}function sanitizeArtifactPath(e){return e.split(`/`).map(e=>e.replace(/[^a-zA-Z0-9_-]/g,`_`)).join(`/`)}export{resolveArtifactDirectory,writeArtifacts};
+`);await writeFile(join(r,`${sanitizeArtifactPath(t.id)}.events.ndjson`),`${i}\n`)}))}function buildSummaryArtifact(e){return{target:e.target,startedAt:e.startedAt,completedAt:e.completedAt,passed:e.passed,failed:e.failed,scored:e.scored,skipped:e.skipped,errored:e.errored,totalEvals:e.results.length,evals:e.results.map(e=>({id:e.id,verdict:e.verdict,status:e.result.status,assertions:e.assertions.map(e=>({name:e.name,score:e.score,severity:e.severity,passed:e.passed})),error:e.error,skipReason:e.skipReason}))}}function buildResultLine(e){return{id:e.id,verdict:e.verdict,status:e.result.status,output:e.result.output,assertions:e.assertions,error:e.error,skipReason:e.skipReason}}function buildEvalArtifact(e){return{id:e.id,result:{output:e.result.output,finalMessage:e.result.finalMessage,sessionId:e.result.sessionId,status:e.result.status,logs:e.result.logs,derived:e.result.derived,sessions:e.result.sessions},verdict:e.verdict,assertions:e.assertions,error:e.error,skipReason:e.skipReason}}function sanitizeArtifactPath(e){return e.split(`/`).map(e=>e.replace(/[^a-zA-Z0-9_-]/g,`_`)).join(`/`)}export{resolveArtifactDirectory,writeArtifacts};

package/dist/src/evals/runner/derive-run-facts.js

@@ -1 +1 @@
-const TURN_EPILOGUE_EVENT_TYPES=new Set([`turn.completed`,`session.waiting`,`session.completed`]);function deriveRunFacts(e,t){let n=t?.sessionId,r=[],i=new Map,a=[],o=new Map,s=[],c=-1,l=0,u=0,d,ensureSubagentCall=(e,t)=>{let r=o.get(e);if(r!==void 0)return r;let i={name:t,isError:!1,turnIndex:Math.max(c,0),sessionId:n};return a.push(i),o.set(e,i),i};for(let t of e)switch(t.type){case`turn.started`:c+=1;break;case`actions.requested`:for(let e of t.data.actions){if(e.kind!==`tool-call`)continue;let t={name:e.toolName,input:e.input,output:void 0,isError:!1,turnIndex:Math.max(c,0),sessionId:n};r.push(t),i.set(e.callId,t)}break;case`action.result`:{let{result:e,status:n}=t.data,r=n===`failed`||e.isError===!0;if(e.kind===`tool-result`){let t=i.get(e.callId);t!==void 0&&(t.output=e.output,t.isError=r)}else if(e.kind===`subagent-result`){let t=o.get(e.callId);t!==void 0&&(t.output=t.output??e.output,t.isError=r)}break}case`subagent.called`:{let e=ensureSubagentCall(t.data.callId,t.data.name);t.data.remote!==void 0&&(e.remoteUrl=t.data.remote.url);break}case`subagent.started`:ensureSubagentCall(t.data.callId,t.data.subagentName);break;case`subagent.completed`:{let e=ensureSubagentCall(t.data.callId,t.data.subagentName);e.output=t.data.output;break}case`input.requested`:s.push(...t.data.requests);break;case`message.completed`:t.data.finishReason!==`tool-calls`&&(l+=1);break;case`reasoning.completed`:u+=1;break;case`session.failed`:d=t.data.code;break}return{toolCalls:r,toolCallCount:r.length,subagentCalls:a,subagentCallCount:a.length,inputRequests:s,parked:endedParkedOnInput(e),messageCount:l,reasoningBlockCount:u,failureCode:d}}function createEmptyDerivedFacts(){return{toolCalls:[],toolCallCount:0,subagentCalls:[],subagentCallCount:0,inputRequests:[],parked:!1,messageCount:0,reasoningBlockCount:0}}function endedParkedOnInput(t){for(let n=t.length-1;n>=0;n--){let r=t[n];if(!(r===void 0||TURN_EPILOGUE_EVENT_TYPES.has(r.type)))return r.type===`input.requested`}return!1}export{createEmptyDerivedFacts,deriveRunFacts};
+const TURN_EPILOGUE_EVENT_TYPES=new Set([`turn.completed`,`session.waiting`,`session.completed`]);function deriveRunFacts(e,t){let n=t?.sessionId,r=[],i=new Map,a=[],o=new Map,s=[],c=-1,l=0,u=0,d,ensureToolCall=(e,t,a)=>{let o=i.get(e);if(o!==void 0)return o;let s={name:t,input:a,output:void 0,status:`pending`,turnIndex:Math.max(c,0),sessionId:n};return r.push(s),i.set(e,s),s},ensureSubagentCall=(e,t)=>{let r=o.get(e);if(r!==void 0)return r;let i={name:t,status:`pending`,turnIndex:Math.max(c,0),sessionId:n};return a.push(i),o.set(e,i),i};for(let t of e)switch(t.type){case`turn.started`:c+=1;break;case`actions.requested`:for(let e of t.data.actions)e.kind===`tool-call`&&ensureToolCall(e.callId,e.toolName,e.input);break;case`action.result`:{let{result:e,status:n}=t.data;if(e.kind===`tool-result`){let t=ensureToolCall(e.callId,e.toolName,{});t.output=e.output,t.status=n}else if(e.kind===`subagent-result`){let t=ensureSubagentCall(e.callId,e.subagentName);t.output=t.output??e.output,t.status=n}break}case`subagent.called`:{let e=ensureSubagentCall(t.data.callId,t.data.name);t.data.remote!==void 0&&(e.remoteUrl=t.data.remote.url);break}case`subagent.started`:ensureSubagentCall(t.data.callId,t.data.subagentName);break;case`subagent.completed`:{let e=ensureSubagentCall(t.data.callId,t.data.subagentName);e.output=t.data.output,e.status===`pending`&&(e.status=`completed`);break}case`input.requested`:s.push(...t.data.requests);for(let e of t.data.requests)ensureToolCall(e.action.callId,e.action.toolName,e.action.input);break;case`message.completed`:t.data.finishReason!==`tool-calls`&&(l+=1);break;case`reasoning.completed`:u+=1;break;case`session.failed`:d=t.data.code;break}return{toolCalls:r,toolCallCount:r.length,subagentCalls:a,subagentCallCount:a.length,inputRequests:s,parked:endedParkedOnInput(e),messageCount:l,reasoningBlockCount:u,failureCode:d}}function createEmptyDerivedFacts(){return{toolCalls:[],toolCallCount:0,subagentCalls:[],subagentCallCount:0,inputRequests:[],parked:!1,messageCount:0,reasoningBlockCount:0}}function endedParkedOnInput(t){for(let n=t.length-1;n>=0;n--){let r=t[n];if(!(r===void 0||TURN_EPILOGUE_EVENT_TYPES.has(r.type)))return r.type===`input.requested`}return!1}export{createEmptyDerivedFacts,deriveRunFacts};

package/dist/src/evals/runner/execute-eval.js

@@ -1 +1 @@
-import{toErrorMessage}from"#shared/errors.js";import{createEmptyDerivedFacts}from"#evals/runner/derive-run-facts.js";import{executeTask}from"#evals/runner/execute-task.js";import{computeEvalVerdict}from"#evals/runner/verdict.js";async function executeEval(n){let{evaluation:r,target:i,client:a}=n,o=new Date().toISOString(),s,c=[],l;try{let e=await executeTask({client:a,evaluation:r,onLog:n.onLog,target:i,timeoutMs:n.timeoutMs??r.timeoutMs});s=e.result,c=e.assertions,l=e.error}catch(t){l=toErrorMessage(t),s={output:null,finalMessage:null,status:`failed`,events:[],derived:createEmptyDerivedFacts()}}let u=computeEvalVerdict({error:l,assertions:c});return{id:r.id,result:s,assertions:c,verdict:u,error:l,startedAt:o,completedAt:new Date().toISOString()}}export{executeEval};
+import{toErrorMessage}from"#shared/errors.js";import{createEmptyDerivedFacts}from"#evals/runner/derive-run-facts.js";import{executeTask}from"#evals/runner/execute-task.js";import{computeEvalVerdict}from"#evals/runner/verdict.js";async function executeEval(n){let{evaluation:r,target:i,client:a}=n,o=new Date().toISOString(),s,c=[],l,u;try{let e=await executeTask({client:a,evaluation:r,onLog:n.onLog,target:i,timeoutMs:n.timeoutMs??r.timeoutMs});s=e.result,c=e.assertions,l=e.error,u=e.skipReason}catch(t){l=toErrorMessage(t),s={output:null,finalMessage:null,status:`failed`,events:[],derived:createEmptyDerivedFacts()}}let d=computeEvalVerdict({error:l,assertions:c,skipReason:u});return{id:r.id,result:s,assertions:c,verdict:d,error:l,skipReason:u,startedAt:o,completedAt:new Date().toISOString()}}export{executeEval};