工作流编排

工作流编排中心 - 协调全流程，管理任务依赖和状态

Security Scan

The skill's manifest, README, and code broadly match an orchestrator for ad-production workflows, but there are potential execution-safety concerns (template/condition evaluation and arbitrary skill invocation) that need review before trusting it in production.

Purpose & Capability

Name/description (workflow orchestration for ad production) aligns with the SKILL.md and the included TypeScript: it defines workflows, steps, dependencies, start/pause/resume, and templates for ad pipelines. The declared requirements (no env vars, no binaries) are consistent with a pure in-agent orchestrator.

Instruction Scope

SKILL.md and the code show the orchestrator invokes other skills/actions via api.executeAction with user-provided step.skill and step.command. The workflow supports condition strings ("condition") and template placeholders ("{{...}}") that are evaluated against runtime context. If those evaluations are implemented using eval/Function or unsafe template evaluation, they could execute arbitrary code or be used to escalate actions. The SKILL.md does not describe safety/permission boundaries or validation of invoked skill names/commands.

Install Mechanism

No install spec or external downloads; it's instruction-only plus a TypeScript source file. No packages or external binaries are pulled in, which reduces supply-chain risk.

Credentials

The skill requests no environment variables, credentials, or config paths. That is proportionate for an in-agent orchestrator. There are no surprising credential requests in the manifest.

Persistence & Privilege

always:false (not force-installed). disable-model-invocation:false (agent may call the skill autonomously) — this is the platform default. Because the orchestrator can invoke arbitrary skills/actions, autonomous invocation increases blast radius if workflows are maliciously crafted; consider limiting which workflows or which target skills it may call.

What to consider before installing

This skill appears to implement a legitimate workflow orchestrator, but you should not install it into a production agent without a few checks: 1) Inspect the complete index.ts (search for use of eval, new Function, vm, template engines, child_process, fetch/http requests, or any direct filesystem/network calls). If condition evaluation or template substitution uses eval/new Function, request that the author replace it with a safe evaluator or whitelist variables and operations. 2) Confirm the orchestrator enforces an allowlist of skills/actions it may call (don’t let workflows invoke arbitrary sensitive skills like credential managers or deployers). 3) Run it in a sandboxed agent or test environment first and exercise workflows with controlled inputs. 4) If you cannot review the full code, treat this as higher-risk and avoid granting it access to agents that can perform sensitive operations. If you want, provide the full (untruncated) index.ts and I can look specifically for eval/unsafe template patterns and other risky APIs.

Workflow Orchestrator - 工作流编排

负责协调广告创意生产的全流程，管理任务依赖和状态。

Setup

无需额外依赖，TypeScript编译后使用。

When to Use

• 创建和管理工作流
• 协调多个技能协作
• 管理任务依赖关系
• 监控工作流状态

Architecture

workflow-orchestrator/
├── index.ts          # 主入口，工作流编排逻辑
├── package.json      # 依赖配置
└── README.md         # 详细文档

Core Commands

创建工作流

await api.executeAction('workflow-orchestrator.create', {
  name: string,           // 工作流名称
  steps: Array<{          // 步骤定义
    skill: string,        // 使用的技能
    action: string,       // 动作
    input?: any,          // 输入数据
    dependsOn?: string[]  // 依赖的步骤
  }>
});

启动工作流

await api.executeAction('workflow-orchestrator.start', {
  workflowId: string      // 工作流ID
});

暂停工作流

await api.executeAction('workflow-orchestrator.pause', {
  workflowId: string      // 工作流ID
});

恢复工作流

await api.executeAction('workflow-orchestrator.resume', {
  workflowId: string      // 工作流ID
});

获取状态

await api.executeAction('workflow-orchestrator.get-status', {
  workflowId: string      // 工作流ID
});

终止工作流

await api.executeAction('workflow-orchestrator.terminate', {
  workflowId: string      // 工作流ID
});

响应事件

• workflow-orchestrator.started - 工作流启动
• workflow-orchestrator.paused - 工作流暂停
• workflow-orchestrator.resumed - 工作流恢复
• workflow-orchestrator.completed - 工作流完成
• workflow-orchestrator.failed - 工作流失败
• workflow-orchestrator.step-completed - 步骤完成

Configuration

工作流配置示例：

{
  "timeout": 3600,        // 超时时间（秒）
  "maxRetries": 3,        // 最大重试次数
  "parallelLimit": 5      // 最大并行数
}