签名关联类型:AgentModelView – AgentRoleRequirements
本页由当前 SDK 声明自动生成。
AgentModelView · AgentNativeCapability · AgentPromptMessage · AgentPromptMessageRole · AgentReasoningApplication · AgentReasoningCapability · AgentReasoningCapabilitySource · AgentReasoningCapabilityState · AgentReasoningEffort · AgentReasoningOption · AgentRoleChannelCapabilities · AgentRoleDefinition · AgentRoleDescriptor · AgentRoleErrorCode · AgentRoleFailure · AgentRoleImplementation · AgentRoleJsonSchema · AgentRoleNativeCapabilities · AgentRoleOutput · AgentRoleOutputKind · AgentRolePrepareInput · AgentRolePrompt · AgentRoleRef · AgentRoleRequirements
AgentModelView
账户模型目录项(使用账户实际返回的 slug/displayName)。
export type AgentModelView = { id: string; label: string; reasoningEfforts?: readonly AgentReasoningEffort[]; reasoning?: AgentReasoningCapability;};关联类型:AgentReasoningEffort · AgentReasoningCapability
成员说明:
reasoningEfforts:旧版兼容字段;新投影同时携带reasoning。
AgentNativeCapability
export type AgentNativeCapability = (typeof AGENT_NATIVE_CAPABILITIES)[number];AgentPromptMessage
export type AgentPromptMessage = { role: AgentPromptMessageRole; content: string;};AgentPromptMessageRole
export type AgentPromptMessageRole = (typeof AGENT_PROMPT_MESSAGE_ROLES)[number];AgentReasoningApplication
一个回合请求与传输的 reasoning 事实;不保存隐藏推理文本。
export type AgentReasoningApplication = { requested?: AgentReasoningEffort; applied: true | false | "provider-default"; code?: string;};关联类型:AgentReasoningEffort
成员说明:
applied:true=已按显式值应用;false=传输拒绝;provider-default=省略字段交由上游默认。
AgentReasoningCapability
可追溯的模型 reasoning 能力。
supported 但没有 efforts 表示上游确认支持推理、却没有给出可安全展示的精确档位;
此时界面只保留默认语义。能力修订号用于回合冻结与目录刷新后的失效判断。
export type AgentReasoningCapability = { state: AgentReasoningCapabilityState; efforts?: readonly AgentReasoningOption[]; source: AgentReasoningCapabilitySource; revision: string;};关联类型:AgentReasoningCapabilityState · AgentReasoningOption · AgentReasoningCapabilitySource
AgentReasoningCapabilitySource
export type AgentReasoningCapabilitySource = (typeof AGENT_REASONING_CAPABILITY_SOURCES)[number];AgentReasoningCapabilityState
export type AgentReasoningCapabilityState = (typeof AGENT_REASONING_CAPABILITY_STATES)[number];AgentReasoningEffort
export type AgentReasoningEffort = (typeof AGENT_REASONING_EFFORTS)[number];AgentReasoningOption
export type AgentReasoningOption = { id: AgentReasoningEffort; label?: string;};关联类型:AgentReasoningEffort
AgentRoleChannelCapabilities
执行通道对本轮请求的真实能力声明(由通道/模型目录给出,不从角色推导)。
export type AgentRoleChannelCapabilities = { readonly toolCalling: boolean; readonly imageInput: boolean; readonly structuredOutput: AgentRoleStructuredOutput; readonly nativeCapabilities: readonly AgentNativeCapability[];};关联类型:AgentRoleStructuredOutput · AgentNativeCapability
AgentRoleDefinition
插件声明的角色定义。id 只是插件内名字,完整身份由宿主加扩展命名空间。
export type AgentRoleDefinition = { readonly id: string; readonly revision: string; readonly title: string; readonly specialty: string; readonly instructions: string; readonly inputSchema: AgentRoleJsonSchema; readonly contextSchema: AgentRoleJsonSchema; readonly tools: AgentRoleToolset; readonly nativeCapabilities: AgentRoleNativeCapabilities; readonly requirements: AgentRoleRequirements; readonly output: AgentRoleOutput;};关联类型:AgentRoleJsonSchema · AgentRoleToolset · AgentRoleNativeCapabilities · AgentRoleRequirements · AgentRoleOutput
成员说明:
instructions:基础角色指令;宿主放在规范消息最前面。contextSchema:领域预先授权并准备好的上下文形状。
AgentRoleDescriptor
只读目录项:不含运行回调、凭据或完整权威上下文。
export type AgentRoleDescriptor = { readonly ref: AgentRoleRef; readonly title: string; readonly specialty: string; readonly requirements: AgentRoleRequirements; readonly tools?: AgentRoleToolset; readonly nativeCapabilities?: AgentRoleNativeCapabilities; readonly outputKind: AgentRoleOutputKind; readonly available: boolean; readonly reason?: string; readonly missing?: readonly string[];};关联类型:AgentRoleRef · AgentRoleRequirements · AgentRoleToolset · AgentRoleNativeCapabilities · AgentRoleOutputKind
成员说明:
tools:宿主内部/受信扩展可见;公开 HTTP 目录会省略工具与内建能力清单。reason:不可用原因(缺失工具/通道能力/插件停用);可用时省略。missing:结构化缺失项(tool:<name>/native:<capability>/ 通道能力名)。
AgentRoleErrorCode
export type AgentRoleErrorCode = (typeof AGENT_ROLE_ERROR_CODES)[number];AgentRoleFailure
export type AgentRoleFailure = { readonly ok: false; readonly code: AgentRoleErrorCode; readonly message: string; readonly missing?: readonly string[];};关联类型:AgentRoleErrorCode
成员说明:
missing:缺失的必要能力/工具/内建能力(requirement-unsatisfied)。
AgentRoleImplementation
插件注册的角色实现;prepare 必须是确定性同步模板构造。
export type AgentRoleImplementation = { readonly definition: AgentRoleDefinition; prepare(input: AgentRolePrepareInput): AgentRolePrompt;};关联类型:AgentRoleDefinition · AgentRolePrepareInput · AgentRolePrompt
AgentRoleJsonSchema
可序列化 JSON schema 形状(复用项目既有 Record<string, unknown> 约定,不引入验证框架)。
export type AgentRoleJsonSchema = Readonly<Record<string, unknown>>;AgentRoleNativeCapabilities
角色允许的内建能力;缺省关闭。
export type AgentRoleNativeCapabilities = { readonly required: readonly AgentNativeCapability[]; readonly optional: readonly AgentNativeCapability[];};AgentRoleOutput
交付形态;与用户的 generate/modify/discuss 动作正交。
export type AgentRoleOutput = { readonly kind: "reply";} | { readonly kind: "candidate"; readonly schema: AgentRoleJsonSchema;} | { readonly kind: "tool-receipts";};关联类型:AgentRoleJsonSchema
AgentRoleOutputKind
export type AgentRoleOutputKind = AgentRoleOutput["kind"];关联类型:AgentRoleOutput
AgentRolePrepareInput
prepare 只接收已校验的输入、宿主授权的领域上下文与宿主冻结的目标身份。
export type AgentRolePrepareInput = { readonly input: unknown; readonly context: unknown; readonly target: AgentRoleTargetRef; readonly requestId: string; readonly expectedContentHash?: string; readonly nativeCapabilities: readonly AgentNativeCapability[];};关联类型:AgentRoleTargetRef · AgentNativeCapability
成员说明:
target:受理时冻结的目标身份(只含稳定 id/范围,不含 owner/租约/grant)。requestId:本回合稳定 requestId(只用于让角色把同一身份写进指令/工具参数)。expectedContentHash:领域冻结的内容前置条件(若有);角色只能转述,不能自行计算。nativeCapabilities:宿主/通道实际开启的内建能力(可选能力只在真实支持时出现)。
AgentRolePrompt
prepare 产出的规范消息与可选的工具收窄集合。
export type AgentRolePrompt = { readonly messages: readonly AgentPromptMessage[]; readonly tools?: readonly string[];};关联类型:AgentPromptMessage
成员说明:
tools:只能进一步收窄 definition.tools;越界或未声明即视为非法准备。
AgentRoleRef
宿主确认的完整角色身份。
descriptorDigest 是定义内容摘要,artifactId 是实际装载的实现制品身份;两者分开保存,
不能用函数 toString 摘要冒充实现身份。
export type AgentRoleRef = { readonly namespacedId: string; readonly roleId: string; readonly extensionId: string; readonly revision: string; readonly descriptorDigest: string; readonly artifactId: string; readonly configVersion?: number;};成员说明:
namespacedId:<extensionId>/<roleId>;业务入口必须显式选择完整身份。
AgentRoleRequirements
角色对执行通道的要求;缺失即在发送前明确不可用。
export type AgentRoleRequirements = { imageInput: boolean; toolCalling: boolean; structuredOutput: AgentRoleStructuredOutput;};