fix(workflow): harden publishing and editor testing
Require high-risk actions to reference human confirmation outputs, filter deleted workflows from dashboard lists, and keep template metadata synchronized. Add safe browser-side debug execution for business nodes without triggering real side effects.
This commit is contained in:
mlogclub
+214
@@ -0,0 +1,214 @@
|
||||
import {
|
||||
WorkflowStatus,
|
||||
type IReport,
|
||||
type TaskRunInput,
|
||||
type WorkflowInputs,
|
||||
type WorkflowOutputs,
|
||||
} from '@flowgram.ai/runtime-interface';
|
||||
|
||||
type DebugNode = {
|
||||
id: string;
|
||||
type: string;
|
||||
data?: {
|
||||
config?: Record<string, unknown>;
|
||||
inputsValues?: Record<string, DebugValue>;
|
||||
};
|
||||
};
|
||||
|
||||
type DebugEdge = {
|
||||
sourceNodeID: string;
|
||||
targetNodeID: string;
|
||||
};
|
||||
|
||||
type DebugSchema = {
|
||||
nodes?: DebugNode[];
|
||||
edges?: DebugEdge[];
|
||||
};
|
||||
|
||||
type DebugValue = {
|
||||
type?: string;
|
||||
content?: unknown;
|
||||
};
|
||||
|
||||
const BUSINESS_NODE_TYPES = new Set([
|
||||
'conversation_understanding',
|
||||
'reply_policy',
|
||||
'knowledge_retrieve',
|
||||
'answerability_gate',
|
||||
'llm_reply',
|
||||
'analyze_conversation',
|
||||
'prepare_ticket_draft',
|
||||
'human_confirm',
|
||||
'create_ticket',
|
||||
'handoff_to_human',
|
||||
'send_reply',
|
||||
]);
|
||||
|
||||
export function isBusinessDebugSchema(input: TaskRunInput): boolean {
|
||||
const schema = parseSchema(input.schema);
|
||||
return (schema.nodes ?? []).some((node) => BUSINESS_NODE_TYPES.has(node.type));
|
||||
}
|
||||
|
||||
export function buildBusinessDebugReport(input: TaskRunInput, taskID: string): IReport {
|
||||
const schema = parseSchema(input.schema);
|
||||
const nodes = schema.nodes ?? [];
|
||||
const nodesByID = new Map(nodes.map((node) => [node.id, node]));
|
||||
const outgoing = new Map<string, DebugEdge[]>();
|
||||
for (const edge of schema.edges ?? []) {
|
||||
outgoing.set(edge.sourceNodeID, [...(outgoing.get(edge.sourceNodeID) ?? []), edge]);
|
||||
}
|
||||
|
||||
const values = new Map<string, WorkflowOutputs>();
|
||||
const reports: IReport['reports'] = {};
|
||||
const startedAt = Date.now();
|
||||
let current = nodes.find((node) => node.type === 'start');
|
||||
let workflowOutputs: WorkflowOutputs = { status: 'completed', debug: true };
|
||||
|
||||
for (let step = 0; current && step < 128; step += 1) {
|
||||
const nodeInputs = resolveNodeInputs(current, values);
|
||||
const nodeOutputs = simulateNode(current, nodeInputs, input.inputs);
|
||||
values.set(current.id, nodeOutputs);
|
||||
const now = Date.now();
|
||||
reports[current.id] = {
|
||||
id: current.id,
|
||||
status: WorkflowStatus.Succeeded,
|
||||
terminated: true,
|
||||
startTime: now,
|
||||
endTime: now,
|
||||
timeCost: 0,
|
||||
snapshots: [
|
||||
{
|
||||
id: `${taskID}-${current.id}`,
|
||||
nodeID: current.id,
|
||||
inputs: nodeInputs,
|
||||
outputs: nodeOutputs,
|
||||
data: { debug: true },
|
||||
},
|
||||
],
|
||||
};
|
||||
if (current.type === 'end') {
|
||||
workflowOutputs = { ...nodeOutputs, status: 'completed', debug: true };
|
||||
break;
|
||||
}
|
||||
const nextEdge = (outgoing.get(current.id) ?? [])[0];
|
||||
current = nextEdge ? nodesByID.get(nextEdge.targetNodeID) : undefined;
|
||||
}
|
||||
|
||||
const endedAt = Date.now();
|
||||
return {
|
||||
id: taskID,
|
||||
inputs: input.inputs,
|
||||
outputs: workflowOutputs,
|
||||
workflowStatus: {
|
||||
status: WorkflowStatus.Succeeded,
|
||||
terminated: true,
|
||||
startTime: startedAt,
|
||||
endTime: endedAt,
|
||||
timeCost: endedAt - startedAt,
|
||||
},
|
||||
reports,
|
||||
messages: {
|
||||
log: [],
|
||||
info: [],
|
||||
debug: [],
|
||||
error: [],
|
||||
warning: [],
|
||||
},
|
||||
};
|
||||
}
|
||||
|
||||
function parseSchema(raw: string): DebugSchema {
|
||||
try {
|
||||
return JSON.parse(raw) as DebugSchema;
|
||||
} catch {
|
||||
return {};
|
||||
}
|
||||
}
|
||||
|
||||
function resolveNodeInputs(
|
||||
node: DebugNode,
|
||||
values: Map<string, WorkflowOutputs>
|
||||
): WorkflowInputs {
|
||||
return Object.fromEntries(
|
||||
Object.entries(node.data?.inputsValues ?? {}).map(([name, value]) => [
|
||||
name,
|
||||
resolveValue(value, values),
|
||||
])
|
||||
);
|
||||
}
|
||||
|
||||
function resolveValue(value: DebugValue, values: Map<string, WorkflowOutputs>): unknown {
|
||||
if (value?.type === 'ref' && Array.isArray(value.content)) {
|
||||
const [nodeID, field] = value.content;
|
||||
if (typeof nodeID === 'string' && typeof field === 'string') {
|
||||
return values.get(nodeID)?.[field];
|
||||
}
|
||||
}
|
||||
return value?.content;
|
||||
}
|
||||
|
||||
function simulateNode(
|
||||
node: DebugNode,
|
||||
inputs: WorkflowInputs,
|
||||
workflowInputs: WorkflowInputs
|
||||
): WorkflowOutputs {
|
||||
const userMessage = String(workflowInputs.userMessage ?? workflowInputs.query ?? '测试消息');
|
||||
switch (node.type) {
|
||||
case 'start':
|
||||
return { ...workflowInputs, userMessage, query: userMessage };
|
||||
case 'conversation_understanding':
|
||||
return {
|
||||
normalizedMessage: userMessage,
|
||||
messageIntent: 'ticket_request',
|
||||
answerScope: 'needs_ticket',
|
||||
confidence: 1,
|
||||
riskSignals: [],
|
||||
reason: '调试模拟结果',
|
||||
};
|
||||
case 'reply_policy':
|
||||
return { action: 'prepare_ticket', requiresFlow: true, targetFlow: 'prepare_ticket' };
|
||||
case 'knowledge_retrieve':
|
||||
return { documents: [], count: 0, query: String(inputs.query ?? userMessage) };
|
||||
case 'answerability_gate':
|
||||
return { answerability: 'answerable', reason: '调试模拟结果' };
|
||||
case 'analyze_conversation':
|
||||
return { intent: 'ticket_request', riskLevel: 'low', needTicket: true };
|
||||
case 'prepare_ticket_draft': {
|
||||
const issue = String(inputs.issue ?? userMessage).trim() || '测试问题';
|
||||
const title = issue.length > 30 ? `${issue.slice(0, 30)}…` : issue;
|
||||
const ticketDraft = {
|
||||
ready: true,
|
||||
title,
|
||||
description: issue,
|
||||
missingFields: [],
|
||||
followUpQuestions: [],
|
||||
conversationFacts: [issue],
|
||||
};
|
||||
return { ticketDraft, ...ticketDraft };
|
||||
}
|
||||
case 'human_confirm':
|
||||
return { confirmed: true, responseText: '调试运行自动确认' };
|
||||
case 'create_ticket':
|
||||
return {
|
||||
ticketId: 0,
|
||||
ticketNo: '',
|
||||
created: false,
|
||||
skipped: true,
|
||||
message: '调试运行不会创建工单。',
|
||||
};
|
||||
case 'handoff_to_human':
|
||||
return { handoffId: 0, skipped: true, message: '调试运行不会转人工。' };
|
||||
case 'llm_reply': {
|
||||
const staticReply = String(node.data?.config?.staticReply ?? '').trim();
|
||||
return { replyText: staticReply || '调试回复' };
|
||||
}
|
||||
case 'send_reply':
|
||||
return { sent: Boolean(inputs.replyText), replyMessageId: 0 };
|
||||
case 'condition':
|
||||
return { matched: true };
|
||||
case 'end':
|
||||
return { ...inputs, status: 'completed' };
|
||||
default:
|
||||
return { ...inputs, debug: true };
|
||||
}
|
||||
}
|
||||
@@ -4,14 +4,27 @@
|
||||
*/
|
||||
|
||||
/* eslint-disable no-console */
|
||||
import { FlowGramAPIName, IRuntimeClient } from '@flowgram.ai/runtime-interface';
|
||||
import {
|
||||
FlowGramAPIName,
|
||||
IRuntimeClient,
|
||||
type TaskReportOutput,
|
||||
} from '@flowgram.ai/runtime-interface';
|
||||
import { injectable } from '@flowgram.ai/free-layout-editor';
|
||||
|
||||
import { buildBusinessDebugReport, isBusinessDebugSchema } from './business-debug-runtime';
|
||||
|
||||
@injectable()
|
||||
export class WorkflowRuntimeBrowserClient implements IRuntimeClient {
|
||||
private businessDebugTasks = new Map<string, TaskReportOutput>();
|
||||
|
||||
constructor() {}
|
||||
|
||||
public [FlowGramAPIName.TaskRun]: IRuntimeClient[FlowGramAPIName.TaskRun] = async (input) => {
|
||||
if (isBusinessDebugSchema(input)) {
|
||||
const taskID = `business-debug-${Date.now()}`;
|
||||
this.businessDebugTasks.set(taskID, buildBusinessDebugReport(input, taskID));
|
||||
return { taskID };
|
||||
}
|
||||
const { TaskRunAPI } = await import('@flowgram.ai/runtime-js'); // Load on demand - 按需加载
|
||||
return TaskRunAPI(input);
|
||||
};
|
||||
@@ -19,6 +32,9 @@ export class WorkflowRuntimeBrowserClient implements IRuntimeClient {
|
||||
public [FlowGramAPIName.TaskReport]: IRuntimeClient[FlowGramAPIName.TaskReport] = async (
|
||||
input
|
||||
) => {
|
||||
if (this.businessDebugTasks.has(input.taskID)) {
|
||||
return this.businessDebugTasks.get(input.taskID);
|
||||
}
|
||||
const { TaskReportAPI } = await import('@flowgram.ai/runtime-js'); // Load on demand - 按需加载
|
||||
return TaskReportAPI(input);
|
||||
};
|
||||
@@ -26,6 +42,10 @@ export class WorkflowRuntimeBrowserClient implements IRuntimeClient {
|
||||
public [FlowGramAPIName.TaskResult]: IRuntimeClient[FlowGramAPIName.TaskResult] = async (
|
||||
input
|
||||
) => {
|
||||
const report = this.businessDebugTasks.get(input.taskID);
|
||||
if (report) {
|
||||
return report.outputs;
|
||||
}
|
||||
const { TaskResultAPI } = await import('@flowgram.ai/runtime-js'); // Load on demand - 按需加载
|
||||
return TaskResultAPI(input);
|
||||
};
|
||||
@@ -33,6 +53,9 @@ export class WorkflowRuntimeBrowserClient implements IRuntimeClient {
|
||||
public [FlowGramAPIName.TaskCancel]: IRuntimeClient[FlowGramAPIName.TaskCancel] = async (
|
||||
input
|
||||
) => {
|
||||
if (this.businessDebugTasks.delete(input.taskID)) {
|
||||
return { success: true };
|
||||
}
|
||||
const { TaskCancelAPI } = await import('@flowgram.ai/runtime-js'); // Load on demand - 按需加载
|
||||
return TaskCancelAPI(input);
|
||||
};
|
||||
@@ -40,6 +63,9 @@ export class WorkflowRuntimeBrowserClient implements IRuntimeClient {
|
||||
public [FlowGramAPIName.TaskValidate]: IRuntimeClient[FlowGramAPIName.TaskValidate] = async (
|
||||
input
|
||||
) => {
|
||||
if (isBusinessDebugSchema(input)) {
|
||||
return { valid: true, errors: [] };
|
||||
}
|
||||
const { TaskValidateAPI } = await import('@flowgram.ai/runtime-js'); // Load on demand - 按需加载
|
||||
return TaskValidateAPI(input);
|
||||
};
|
||||
|
||||
Reference in New Issue
Block a user