Answering catalog: tasks, source reads, and check references

One task definition ties together the input Schema, reader, output, and check, with real references verified at assembly.

defineAnswerReaderCatalog holds real read ports; defineAnswerCheckCatalog compiles a synchronous check list. defineAnswerTaskCatalog uses those catalogs to verify readerId, outputId, checkIds, goalIds, and the input Schema.

task.read(request,context) only allows synchronous readers; readAsync awaits asynchronous readers at the same task entry and checks cancellation. The check catalog runs in checkIds order and keeps the corrected value plus all issues; any failed wins first, then unavailable, and only an all-pass run is passed. The task catalog's describe projects inputJsonSchema without performing reads or model calls.

Catalog validation is not read authorization. Readers obtain trusted sources from server-side App dependencies; a model cannot gain permission through taskId, sourceStatus, or outputId.

Progressive examples for the same business case

Register one read task for explanatory material

Beginner · Node.js

TypeScript

import { Answering } from '@jangts/agent-core/kits/answering';
import { Presentation } from '@jangts/agent-core/kits/presentation';
const readers = Answering.defineAnswerReaderCatalog([{ id: 'manual', read: (input: string) => ({ question: input, source: 'Agent Core 允许普通对象节点。' }) }]);
const checks = Answering.defineAnswerCheckCatalog<string, {}>([]);
const outputs = Presentation.defineOutputCatalog([{ id: 'reply', kind: 'text', fields: [{ id: 'text', owner: 'generated' }] }]);
const tasks = Answering.defineAnswerTaskCatalog([
  { id: 'explain', description: '根据说明资料解释接口', readerId: 'manual', outputId: 'reply', goalIds: ['explain'], checkIds: [], metadata: {},
    inputSchema: { jsonSchema: { type: 'string' }, parse(value: unknown) { if (typeof value !== 'string') throw new Error('需要问题'); return value; } } },
], { readers, checks, outputs });
console.log(tasks.read({ taskId: 'explain', input: '如何定义节点？' }, { app: {}, signal: new AbortController().signal }));

JavaScript

import { Answering } from '@jangts/agent-core/kits/answering';
import { Presentation } from '@jangts/agent-core/kits/presentation';
const readers = Answering.defineAnswerReaderCatalog([{ id: 'manual', read: (input) => ({ question: input, source: 'Agent Core 允许普通对象节点。' }) }]);
const checks = Answering.defineAnswerCheckCatalog([]);
const outputs = Presentation.defineOutputCatalog([{ id: 'reply', kind: 'text', fields: [{ id: 'text', owner: 'generated' }] }]);
const tasks = Answering.defineAnswerTaskCatalog([
    { id: 'explain', description: '根据说明资料解释接口', readerId: 'manual', outputId: 'reply', goalIds: ['explain'], checkIds: [], metadata: {},
        inputSchema: { jsonSchema: { type: 'string' }, parse(value) { if (typeof value !== 'string')
                throw new Error('需要问题'); return value; } } },
], { readers, checks, outputs });
console.log(tasks.read({ taskId: 'explain', input: '如何定义节点？' }, { app: {}, signal: new AbortController().signal }));

Add check references to the same task and check the generated result

Configuration · Node.js

TypeScript

import { Answering } from '@jangts/agent-core/kits/answering';
import { Presentation } from '@jangts/agent-core/kits/presentation';
const readers = Answering.defineAnswerReaderCatalog([{ id: 'manual', read: (input: string) => ({ question: input, source: 'Agent Core 允许普通对象节点。' }) }]);
const checks = Answering.defineAnswerCheckCatalog<string, {}>([
  { id: 'nonempty', check: value => ({ status: value.trim() ? 'passed' : 'failed', value: value.trim(), issues: value.trim() ? [] : [{ code: 'EMPTY', message: '答案为空' }] }) },
]);
const outputs = Presentation.defineOutputCatalog([{ id: 'reply', kind: 'text', fields: [{ id: 'text', owner: 'generated' }] }]);
const tasks = Answering.defineAnswerTaskCatalog([
  { id: 'explain', description: '根据说明资料解释接口', readerId: 'manual', outputId: 'reply', goalIds: ['explain'], checkIds: ['nonempty'], metadata: {},
    inputSchema: { jsonSchema: { type: 'string' }, parse(value: unknown) { if (typeof value !== 'string') throw new Error('需要问题'); return value; } } },
], { readers, checks, outputs });
const context = { app: {}, signal: new AbortController().signal };
const evidence = await tasks.readAsync({ taskId: 'explain', input: '如何定义节点？' }, context);
console.log(evidence, checks.check({ checkIds: tasks.require('explain').checkIds, value: ' 普通对象即可。 ' }, context));

JavaScript

import { Answering } from '@jangts/agent-core/kits/answering';
import { Presentation } from '@jangts/agent-core/kits/presentation';
const readers = Answering.defineAnswerReaderCatalog([{ id: 'manual', read: (input) => ({ question: input, source: 'Agent Core 允许普通对象节点。' }) }]);
const checks = Answering.defineAnswerCheckCatalog([
    { id: 'nonempty', check: value => ({ status: value.trim() ? 'passed' : 'failed', value: value.trim(), issues: value.trim() ? [] : [{ code: 'EMPTY', message: '答案为空' }] }) },
]);
const outputs = Presentation.defineOutputCatalog([{ id: 'reply', kind: 'text', fields: [{ id: 'text', owner: 'generated' }] }]);
const tasks = Answering.defineAnswerTaskCatalog([
    { id: 'explain', description: '根据说明资料解释接口', readerId: 'manual', outputId: 'reply', goalIds: ['explain'], checkIds: ['nonempty'], metadata: {},
        inputSchema: { jsonSchema: { type: 'string' }, parse(value) { if (typeof value !== 'string')
                throw new Error('需要问题'); return value; } } },
], { readers, checks, outputs });
const context = { app: {}, signal: new AbortController().signal };
const evidence = await tasks.readAsync({ taskId: 'explain', input: '如何定义节点？' }, context);
console.log(evidence, checks.check({ checkIds: tasks.require('explain').checkIds, value: ' 普通对象即可。 ' }, context));

Policies & implementations

• Task references and synchronous/asynchronous reads
• Check result aggregation