CreateAgentTask

1. Overview

Use case: When calling the createAgentTask operation, serialize the fields of this class into JSON and pass them in. The server routes to different processing pipelines (text/audio) based on inputType, and routes to different instruction processors (key attribute extraction/service inspection/tag classification, etc.) based on instructionType.

2. Top-level field description

Field	Type	Required	Description
vid	String	No	The session ID. If not specified, a value is automatically generated using the current timestamp.
modelType	String	Yes	The large model type.
agentId	Long	Yes	The Agent configuration ID.
inputType	String	Yes	The analysis object type. Valid values: audio (speech) and text (text).
instructionType	String	Yes	The instruction type. Valid values: serviceInspection (service inspection), field (key attribute extraction), tagCategory (tag classification), multiLevelTag (multi-level tag recognition), and customPrompt (custom instruction).
instructionTypeParam	InstructionTypeParam	Yes	The parameters corresponding to the instruction type. Different instructionType values require different sub-parameters (see §4.1).
taskParam	TaskParam	Yes	The task data parameters, including text conversation or audio file information (see §3.1).

3. Detailed description of nested classes.

3.1 TaskParam — Task data parameters

AgentTaskParam.TaskParam carries the input data for the task. Different sub-parameters are used depending on the value of inputType.

Field	Type	Description
textParam	TextParam	The text parameters. Used when inputType = "text".
audioParam	AudioParam	The audio parameters. Used when inputType = "audio".
callbackUrl	String	The callback notification URL after the task is completed (optional).

Usage rules:

• When inputType = "text", you must provide textParam.

• When inputType = "audio", you must provide audioParam.

• textParam and audioParam are mutually exclusive. Provide only one of them for a single task.

3.2 TextParam — Text parameters

AgentTaskParam.TextParam encapsulates text-type input data.

Field	Type	Description
dialogue	List<Dialogue>	The dialogue list. Each entry is a Dialogue object.
dialogueContent	String	The raw dialogue text string. Provide either this field or dialogue. At least one must be specified.

Dialogue object structure:

Field	Type	Required	Description
role	String	Yes	The speaker role, such as "customer" or "agent".
words	String	Yes	The dialogue text content.
identity	String	No	The speaker identity identifier (such as the agent employee ID).
begin	Long	No	The relative offset of the sentence start time (in milliseconds).
end	Long	No	The relative offset of the sentence end time (in milliseconds).
beginTime	Date	No	The absolute start time of the sentence (yyyy-MM-dd HH:mm:ss).
hourMinSec	String	No	The time in hour:minute:second format (HH:mm:ss). Required for phone call scenarios.
emotionValue	Integer	No	The emotion energy value (1-10).
speechRate	Integer	No	The speech rate (words per minute).
silenceDuration	Integer	No	The silence duration (in seconds).

3.3 AudioParam — Audio parameters

AgentTaskParam.AudioParam encapsulates audio-type input data.

Field	Type	Required	Description
fileUrl	String	Yes	The access URL of the audio file.
fileName	String	No	The name of the audio file.
asrVersion	String	No	The ASR (speech-to-text) engine version.

---

4. InstructionTypeParam — Detailed description of instruction parameters

AgentManageParam.InstructionTypeParam carries different instruction configuration parameters based on the value of instructionType. Only fill in the sub-parameters corresponding to the current instructionType. Leave the other fields empty.

4.1 FieldsParam — Key attribute extraction (instructionType = "field")

Field	Type	Required	Description
nameDescPairList	List<NameDescPair>	Yes	The list of attribute dimensions to extract.

NameDescPair structure:

Field	Type	Description
name	String	The attribute name (required).
desc	String	The attribute description (optional).

4.2 ServiceInspectionParam — Service inspection (instructionType = "serviceInspection")

Field	Type	Required	Description
sceneName	String	Yes	The inspection scenario name.
sceneDescription	String	No	The inspection scenario description.
dimensions	List<Dimension>	Yes	The list of inspection dimensions.

Dimension structure:

Field	Type	Description
dimension	String	The dimension name (required).
desc	String	The dimension description (optional).

4.3 TagCategoryParam — Tag classification (instructionType = "tagCategory")

Field	Type	Required	Description
nameDescPairList	List<NameDescValueListPair>	Yes	The list of tag classification dimensions.

NameDescValueListPair structure:

Field	Type	Description
name	String	The dimension name.
desc	String	The dimension description.
valueList	List<String>	The list of candidate tag values.

4.4 CustomPrompt — Custom instruction (instructionType = "customPrompt")

Field	Type	Required	Description
customPrompt	String	Yes	The user-defined prompt template content.

4.5 TagTreeLevelParam — Multi-level tag recognition (instructionType = "multiLevelTag")

Field	Type	Required	Description
tagIds	List<Long>	No	The list of runtime starting parent tag IDs. The first ID is used as the matched parent tag, and recognition starts from its child nodes level by level (skipping preceding levels). This field also takes effect when agentId is specified.
tagTreeLevels	List<TagTreeLevelInfoParam>	Yes	The recognition configuration for each level, arranged in hierarchical order.

TagTreeLevelInfoParam structure:

Field	Type	Description
tagTreeIds	List<Long>	The list of candidate tag IDs for this level. A value of -1 indicates that all tags at this level are selected (required).
prompt	String	The analysis prompt for this level. This prompt is concatenated with tag data and sent to the large model (required).

---