{"id":"8e5ea06e-58e0-418d-8552-f2fb709ef0e7","entityType":"agent","slug":"clawhub-gechengling-agentic-workflow-designer","name":"Agentic Workflow Designer","canonicalUrl":"https://www.xpersona.co/agent/clawhub-gechengling-agentic-workflow-designer","canonicalPath":"/agent/clawhub-gechengling-agentic-workflow-designer","generatedAt":"2026-10-10T07:53:39.653Z","source":"CLAWHUB","claimStatus":"UNCLAIMED","verificationTier":"NONE","summary":{"evidence":{"source":"editorial-content","verified":true,"confidence":"high","updatedAt":"2026-10-10T01:40:07.365Z","emptyReason":null},"description":"AI-powered agentic workflow design and automation assistant — map complex multi-step processes, identify automation opportunities, design autonomous AI agent pipelines, generate n8n/Make/Zapier workflow specs, and estimate ROI. Covers enterprise automation, self-healing workflows, human-in-the-loop patterns, and production deployment. Keywords: agentic workflow, workflow automation, n8n, Make, Zapier, enterprise automation, AI pipeline, autonomous agent, process automation, workflow design, ROI calculator, HITL, 工作流设计, 流程自动化, 智能体工作流, 企业自动化, n8n工作流, 流程优化, 自主代理, RPA替代. Skill: Agentic Workflow Designer Owner: gechengling Summary: AI-powered agentic workflow design and automation assistant — map complex multi-step processes, identify automation opportunities, design autonomous AI agent pipelines, generate n8n/Make/Zapier workflow specs, and estimate ROI. Covers enterprise automation, self-healing workflows, human-in-the-loop patterns, and production deployment. Keywords: agentic work","descriptionLabel":"Technical summary","evidenceSummary":"Capability contract not published. No trust telemetry is available yet. 1.8K downloads reported by the source. Last updated 10/10/2026.","installCommand":"clawhub skill install s17ewqc4f2s6gpcbm88hy7fgvn85kg1g:agentic-workflow-designer","sourceUrl":"https://clawhub.ai/gechengling/agentic-workflow-designer","homepage":"https://clawhub.ai/gechengling/skills/agentic-workflow-designer","primaryLinks":[{"label":"View on ClawHub","url":"https://clawhub.ai/gechengling/agentic-workflow-designer","kind":"source"},{"label":"Homepage","url":"https://clawhub.ai/gechengling/skills/agentic-workflow-designer","kind":"homepage"}],"safetyScore":84,"overallRank":62,"popularityScore":65,"trustScore":null,"claimedByName":null,"isOwner":false,"seoDescription":"AI-powered agentic workflow design and automation assistant — map complex multi-step processes, identify automation opportunities, design autonomous AI agent pi"},"coverage":{"evidence":{"source":"public-profile","verified":false,"confidence":"medium","updatedAt":"2026-10-10T01:40:07.365Z","emptyReason":null},"protocols":[{"protocol":"OPENCLEW","label":"OpenClaw","status":"self-declared","notes":"Declared in the public agent profile."}],"capabilities":[],"verifiedCount":0,"selfDeclaredCount":1,"capabilityMatrix":{"rows":[{"key":"OPENCLEW","type":"protocol","support":"unknown","confidenceSource":"profile","notes":"Listed on profile"}],"flattenedTokens":"protocol:OPENCLEW|unknown|profile"}},"adoption":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-10T01:40:07.365Z","emptyReason":null},"stars":null,"forks":null,"downloads":1800,"packageName":null,"latestVersion":"3.3.6","tractionLabel":"1.8K downloads"},"release":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-10T01:40:07.365Z","emptyReason":null},"lastUpdatedAt":"2026-10-10T01:40:07.365Z","lastCrawledAt":"2026-10-10T01:40:07.365Z","lastIndexedAt":null,"nextCrawlAt":"2026-10-11T01:40:07.365Z","lastVerifiedAt":null,"highlights":[{"version":"3.3.6","createdAt":"2026-09-24T05:27:20.409Z","changelog":"新增监管报送示例与边界对比、可观测性指标集与灰度发布四步；多表新增维度列；动态更新至2026-09-24","fileCount":3,"zipByteSize":15050},{"version":"3.3.5","createdAt":"2026-09-07T02:30:20.099Z","changelog":"v3.3.5: 技术动态至2026-09-07并新增本期观察表；补Step1流程拆解示例(7列)；新增常见设计反模式表与规模化治理示例对话","fileCount":3,"zipByteSize":12557},{"version":"3.3.4","createdAt":"2026-08-31T14:41:02.658Z","changelog":"v3.3.4: 修复本地乱码（以线上正本恢复）；Step2-6新增打分示例/蓝图示例/ROI示例；平台对比表新增4个维度；技术动态更新至2026-08-31","fileCount":3,"zipByteSize":10391},{"version":"3.3.3","createdAt":"2026-06-16T02:36:51.291Z","changelog":"Agentic Workflow Designer 3.3.3 - Expanded Chinese-language support: all trigger words, documentation, and technical comparisons now include clear and comprehensive Chinese translations. - Updated 2026 technology landscape: new section details capabilities and benchmarks of LangGraph v1.0, CrewAI v1.10, Claude/OpenAI Agent SDKs, MCP ecosystem, and long-context LLMs. - Improved platform comparison: replaced garbled multilingual tables with detailed Chinese/English tables and concise selection recommendations for 2026. - Enhanced clarity: corrected typographical errors and removed corrupted symbols. - No breaking changes to workflow or input/output structure.","fileCount":3,"zipByteSize":6796},{"version":"3.3.2","createdAt":"2026-06-16T01:46:13.180Z","changelog":"- Updated to version 3.3.2. - Clarified the pipeline blueprint format in documentation for improved readability. - Updated platform comparison tables with clearer self-hosting and agent support indicators. - Removed outdated skill-card.md file.","fileCount":3,"zipByteSize":6492},{"version":"3.3.1","createdAt":"2026-05-27T06:07:53.907Z","changelog":"- Updated SKILL.md to version 3.3.1. - Various Unicode and text encoding issues appeared, resulting in corrupted or garbled multilingual content (notably Chinese text and emoji/icons throughout the file). - No changes to functionality, logic, or workflow steps; update solely affects documentation formatting and display. - All step-by-step workflow, platform comparison, and usage examples remain structurally the same, but may have reduced readability due to encoding issues.","fileCount":3,"zipByteSize":6622},{"version":"v1.0.0","createdAt":"2026-05-18T08:12:32.052Z","changelog":"- Initial release of Agentic Workflow Designer. - Design and document multi-step automation workflows. - Identify automation opportunities and estimate ROI. - Generate n8n/Make/Zapier workflow specifications. - Support for human-in-the-loop, enterprise automation, and platform recommendations.","fileCount":2,"zipByteSize":4886},{"version":"3.3.0","createdAt":"2026-05-16T05:27:07.042Z","changelog":"- Updated version number to 3.2.0 in SKILL.md. - No functional or content changes; documentation update only.","fileCount":2,"zipByteSize":4886}]},"execution":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No published capability contract is available yet."},"installCommand":"clawhub skill install s17ewqc4f2s6gpcbm88hy7fgvn85kg1g:agentic-workflow-designer","setupComplexity":"low","setupSteps":["Setup complexity is classified as HIGH. You must provision dedicated cloud infrastructure or an isolated VM. Do not run this directly on your local workstation.","Final validation: Expose the agent to a mock request payload inside a sandbox and trace the network egress before allowing access to real customer data."],"contract":{"contractStatus":"missing","authModes":[],"requires":[],"forbidden":[],"supportsMcp":false,"supportsA2a":false,"supportsStreaming":false,"inputSchemaRef":null,"outputSchemaRef":null,"dataRegion":null,"contractUpdatedAt":null,"sourceUpdatedAt":null,"freshnessSeconds":null},"invocationGuide":{"preferredApi":{"snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-gechengling-agentic-workflow-designer/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-gechengling-agentic-workflow-designer/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-gechengling-agentic-workflow-designer/trust"},"curlExamples":["curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-gechengling-agentic-workflow-designer/snapshot\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-gechengling-agentic-workflow-designer/contract\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-gechengling-agentic-workflow-designer/trust\""],"jsonRequestTemplate":{"query":"summarize this repo","constraints":{"maxLatencyMs":2000,"protocolPreference":["OPENCLEW"]}},"jsonResponseTemplate":{"ok":true,"result":{"summary":"...","confidence":0.9},"meta":{"source":"CLAWHUB","generatedAt":"2026-10-10T07:53:39.650Z"}},"retryPolicy":{"maxAttempts":3,"backoffMs":[500,1500,3500],"retryableConditions":["HTTP_429","HTTP_503","NETWORK_TIMEOUT"]}},"endpoints":{"dossierUrl":"https://www.xpersona.co/api/v1/agents/clawhub-gechengling-agentic-workflow-designer/dossier","snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-gechengling-agentic-workflow-designer/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-gechengling-agentic-workflow-designer/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-gechengling-agentic-workflow-designer/trust"}},"reliability":{"evidence":{"source":"runtime-metrics","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No trust, reliability, or runtime telemetry is available."},"trust":{"status":"unavailable","handshakeStatus":"UNKNOWN","verificationFreshnessHours":null,"reputationScore":null,"p95LatencyMs":null,"successRate30d":null,"fallbackRate":null,"attempts30d":null,"trustUpdatedAt":null,"trustConfidence":"unknown","sourceUpdatedAt":null,"freshnessSeconds":null},"decisionGuardrails":{"doNotUseIf":["Contract metadata is missing or unavailable for deterministic execution."],"safeUseWhen":[],"riskFlags":["missing_or_unavailable_contract","trust_data_unavailable","schema_references_missing"],"operationalConfidence":"low"},"executionMetrics":{"observedLatencyMsP50":null,"observedLatencyMsP95":null,"estimatedCostUsd":null,"uptime30d":null,"rateLimitRpm":null,"rateLimitBurst":null,"lastVerifiedAt":null,"verificationSource":null},"runtimeMetrics":{"successRate":null,"avgLatencyMs":null,"avgCostUsd":null,"hallucinationRate":null,"retryRate":null,"disputeRate":null,"p50Latency":null,"p95Latency":null,"lastUpdated":null}},"benchmarks":{"evidence":{"source":"no-benchmark-data","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No benchmark suites or observed failure patterns are available."},"suites":[],"failurePatterns":[]},"artifacts":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"high","updatedAt":"2026-10-10T01:40:07.365Z","emptyReason":null},"readme":"Skill: Agentic Workflow Designer\n\nOwner: gechengling\n\nSummary: AI-powered agentic workflow design and automation assistant — map complex multi-step processes, identify automation opportunities, design autonomous AI agent pipelines, generate n8n/Make/Zapier workflow specs, and estimate ROI. Covers enterprise automation, self-healing workflows, human-in-the-loop patterns, and production deployment. Keywords: agentic workflow, workflow automation, n8n, Make, Zapier, enterprise automation, AI pipeline, autonomous agent, process automation, workflow design, ROI calculator, HITL, 工作流设计, 流程自动化, 智能体工作流, 企业自动化, n8n工作流, 流程优化, 自主代理, RPA替代.\n\nTags: agentic:3.3.5, agentic-workflow-designer:3.3.6, automation:3.3.5, design:3.3.5, latest:3.3.6, workflow:3.3.5\n\nVersion history:\n\nv3.3.6 | 2026-09-24T05:27:20.409Z | user\n\n新增监管报送示例与边界对比、可观测性指标集与灰度发布四步；多表新增维度列；动态更新至2026-09-24\n\nv3.3.5 | 2026-09-07T02:30:20.099Z | user\n\nv3.3.5: 技术动态至2026-09-07并新增本期观察表；补Step1流程拆解示例(7列)；新增常见设计反模式表与规模化治理示例对话\n\nv3.3.4 | 2026-08-31T14:41:02.658Z | user\n\nv3.3.4: 修复本地乱码（以线上正本恢复）；Step2-6新增打分示例/蓝图示例/ROI示例；平台对比表新增4个维度；技术动态更新至2026-08-31\n\nv3.3.3 | 2026-06-16T02:36:51.291Z | auto\n\nAgentic Workflow Designer 3.3.3\n\n- Expanded Chinese-language support: all trigger words, documentation, and technical comparisons now include clear and comprehensive Chinese translations.\n- Updated 2026 technology landscape: new section details capabilities and benchmarks of LangGraph v1.0, CrewAI v1.10, Claude/OpenAI Agent SDKs, MCP ecosystem, and long-context LLMs.\n- Improved platform comparison: replaced garbled multilingual tables with detailed Chinese/English tables and concise selection recommendations for 2026.\n- Enhanced clarity: corrected typographical errors and removed corrupted symbols.\n- No breaking changes to workflow or input/output structure.\n\nv3.3.2 | 2026-06-16T01:46:13.180Z | auto\n\n- Updated to version 3.3.2.\n- Clarified the pipeline blueprint format in documentation for improved readability.\n- Updated platform comparison tables with clearer self-hosting and agent support indicators.\n- Removed outdated skill-card.md file.\n\nv3.3.1 | 2026-05-27T06:07:53.907Z | auto\n\n- Updated SKILL.md to version 3.3.1.\n- Various Unicode and text encoding issues appeared, resulting in corrupted or garbled multilingual content (notably Chinese text and emoji/icons throughout the file).\n- No changes to functionality, logic, or workflow steps; update solely affects documentation formatting and display.\n- All step-by-step workflow, platform comparison, and usage examples remain structurally the same, but may have reduced readability due to encoding issues.\n\nvv1.0.0 | 2026-05-18T08:12:32.052Z | auto\n\n- Initial release of Agentic Workflow Designer.\n- Design and document multi-step automation workflows.\n- Identify automation opportunities and estimate ROI.\n- Generate n8n/Make/Zapier workflow specifications.\n- Support for human-in-the-loop, enterprise automation, and platform recommendations.\n\nv3.3.0 | 2026-05-16T05:27:07.042Z | auto\n\n- Updated version number to 3.2.0 in SKILL.md.\n- No functional or content changes; documentation update only.\n\nv3.2.0 | 2026-05-16T02:23:03.351Z | auto\n\nNo user-visible changes in this release.\n\n- Version bump to 3.2.0 with no detected changes to files or documentation.\n\nv3.1.0 | 2026-05-16T01:18:57.149Z | auto\n\nNo changes detected in this version (3.1.0); content and workflow remain the same as the previous release.\n\nv3.0.0 | 2026-05-16T00:16:42.126Z | auto\n\nNo user-visible changes in this version. \n- Version number update only. \n- No updates to content, features, or workflow.\n\nv1.0.0 | 2026-05-15T23:18:33.972Z | auto\n\nInitial release—Agentic Workflow Designer 1.0.0\n\n- Launches the Agentic Workflow Designer skill for mapping, automating, and optimizing multi-step business processes.\n- Supports workflow discovery, automation opportunity assessment, agentic pipeline design, and ROI prediction.\n- Provides blueprints and generates import-ready workflow specs for n8n, Make, and Zapier.\n- Recommends suitable automation platforms and designs human-in-the-loop (HITL) checkpoints.\n- Targeted for operations managers, developers, consultants, and entrepreneurs seeking AI-powered workflow automation.\n\nArchive index:\n\nArchive v3.3.6: 3 files, 15050 bytes\n\nFiles: skill-card.md (2044b), SKILL.md (28984b), _meta.json (144b)\n\nFile v3.3.6:SKILL.md\n\n---\nname: Agentic Workflow Designer\ndescription: >\n AI-powered agentic workflow design and automation assistant — map complex multi-step\n processes, identify automation opportunities, design autonomous AI agent pipelines,\n generate n8n/Make/Zapier workflow specs, and estimate ROI. Covers enterprise automation,\n self-healing workflows, human-in-the-loop patterns, and production deployment. Keywords:\n agentic workflow, workflow automation, n8n, Make, Zapier, enterprise automation,\n AI pipeline, autonomous agent, process automation, workflow design, ROI calculator,\n HITL, 工作流设计, 流程自动化, 智能体工作流, 企业自动化, n8n工作流, 流程优化,\n 自主代理, RPA替代.\nversion: \"3.3.6\"\n---\n\n# Agentic Workflow Designer\n\n> From messy manual processes to autonomous AI pipelines — design, document, and deploy.\n\n> **⚠️ CAPABILITY NOTICE / 能力说明**\n> - **Type:** Design and advisory framework — produces workflow blueprints, JSON/YAML specs, and ROI estimates as reference material\n> - **No code is executed by this skill**; generated specs are for the user to review and import into their own environment\n> - **No persistent storage, network calls, background execution, or credential collection**\n> - **All outputs require human review before production deployment**\n> - Workflows touching PII or regulated data must include retention, access control, and audit considerations\n\n## What This Skill Does\n\nAgentic AI (AI that can autonomously execute multi-step tasks) is the #1 enterprise tech trend in 2026 with a projected $8.5B market and 40% CAGR. Yet most teams struggle to:\n- Map which workflows are actually suitable for agentic automation\n- Design reliable pipelines that don't break silently\n- Choose between n8n, Make, Zapier, or custom agent frameworks\n- Justify the ROI to business stakeholders\n\nThis skill bridges the gap between AI hype and practical workflow automation:\n\n- **Workflow Discovery** — Identify and prioritize automation opportunities in any business process\n- **Agentic Pipeline Design** — Create detailed workflow blueprints with triggers, agents, tools, and fallbacks\n- **Platform Selection** — Compare n8n / Make / Zapier / custom LangGraph for your use case\n- **Generate Workflow Specs** — Produce JSON/YAML specs importable into n8n or Make\n- **ROI Calculator** — Estimate time/cost savings from automation\n- **Human-in-the-Loop (HITL) Design** — Design appropriate checkpoints for sensitive decisions\n\n## Trigger Words\n\nAgentic workflow, automate my process, workflow automation, n8n, Make automation, Zapier flow, design a workflow, workflow design, process automation, automate with AI, AI pipeline, autonomous workflow, HITL pattern, 工作流设计, 自动化工作流, 流程自动化, 智能体工作流, 帮我设计流程, 自动化这个流程, n8n工作流, 企业自动化, RPA替代, agentic AI pipeline\n\n## Target Users\n\n- Operations managers digitizing manual business processes\n- Developers building production AI automation systems\n- Product managers scoping automation features\n- Consultants delivering workflow automation projects\n- Entrepreneurs building AI-native products\n\n## Workflow\n\n### 平台与技术动态（截至 2026-09-24）\n\n**2026-08 更新要点**：\n- **MCP 成为事实标准**：Model Context Protocol 于 2025 年底转入 Linux 基金会中立治理后，2026 年官方与社区服务器数量持续扩张，企业内部 MCP 注册表逐步成为新的基础设施层。\n- **国内合规要求趋严**：涉及个人信息与重要数据的工作流，需满足最小必要采集、境内存储与可审计要求，自托管方案的优先级上升。\n- **长上下文成本下探**：长文档场景（招股书、年报、长合同）的单位 Token 成本持续下降，使得\"全文入参 + 结构化抽取\"逐步替代早期分段检索方案。\n- **可观测性成为刚需**：生产级 agentic 工作流普遍补齐链路追踪、成本归因与失败重放能力，缺乏可观测性的方案难以通过投产评审。\n- **[2026-09 新增] 从\"能不能跑\"转向\"能不能管\"**：试点期关注功能实现，规模化后真正的瓶颈变成命名规范、版本管理、灰度发布与故障定位，设计阶段就要预留这些能力。\n- **[2026-09 新增] 选型标准从跑分转向可控性**：受监管行业（金融、医疗、政务）更看重私线部署、数据不出境与审计留痕，模型跑分高不再是唯一决策依据。\n- **[2026-09 新增] 成本归因细化到工作流**：单位 Token 成本需要能分摊到具体工作流与具体节点，否则规模化后无法判断哪些流程值得继续投入。\n- **[2026-09-24 新增] 智能体身份与权限被单独治理**：工作流里的 agent 逐步被当作\"非人类身份\"管理，需要独立的凭据、最小权限与回收机制，共用一把管理员 Key 的做法在审计中难以通过。\n- **[2026-09-24 新增] 评测集先于优化**：团队在实践中发现，没有评测集就做提示词调优等于盲调；先用 30-100 条真实样本建最小评测集，再谈优化，成为投产前的默认动作。\n- **[2026-09-24 新增] 失败重放成为交付标准**：无法复现失败现场的工作流在验收时被直接打回，输入、输出、版本、耗时、成本五项留痕被视为最低要求。\n\n**本期新增观察（截至 2026-09-24）**\n\n| 维度 | 变化 | 对流程设计的意义 | 落地动作 | 可信度标注 |\n|------|------|----------------|\n| 治理 | 企业内部 MCP 注册表从\"可选\"变为\"基础件\"，工具接入需统一登记与权限控制 | 设计时把工具来源收敛到注册表，避免各流程各接一套 | 建立工具清单与负责人，禁止流程内硬编码凭据 | 以官方最新发布为准 |\n| 运维 | 工作流数量增长后，命名与版本混乱成为主要故障源 | 约定命名规范与语义化版本，变更需留 changelog | 采用\"业务域-动作-对象-版本\"命名，变更走评审 | 以官方最新发布为准 |\n| 成本 | 成本归因要求下沉到节点级 | 每个节点标注预期调用量与成本上限，超限告警 | 为高成本节点设单日预算与熔断 | 以官方最新发布为准 |\n| 合规 | 涉及个人信息与重要数据的流程需满足最小必要与可审计 | 在蓝图阶段即标注数据分类与留存期限 | 数据分类标签前置到 Step 1 盘点表 | 以官方最新发布为准 |\n| 身份 | agent 作为\"非人类身份\"需独立凭据与最小权限 | 共用管理员凭据的做法不可审计 | 一个工作流一套凭据，定期轮换与回收 | 以官方最新发布为准 |\n| 评测 | 最小评测集成为投产前置条件 | 没有评测集就调优等于盲调 | 先建 30-100 条真实样本的最小评测集 | 以官方最新发布为准 |\n\n**Step 2 新增技术评估（2026）**：\n- LangGraph v1.0生产就绪：状态机工作流/长期记忆/错误恢复三大核心能力，企业级部署支持Kubernetes自动扩缩容，GitHub Stars突破85K\n- CrewAI v1.10多智能体协作：支持6种角色类型+并行任务编排，内置20+企业级连接器（Slack/Notion/Airtable/GitHub），2026年Q1新增中文文档\n- Claude Agent SDK / OpenAI Agents SDK横向对比：工具调用准确率(94% vs 91%)/上下文利用率(78% vs 82%)/成本效率(￥0.8/千Token vs ￥1.2/千Token)三大维度全面评测\n- MCP(Model Context Protocol)生态爆发：50+官方服务器覆盖GitHub/Slack/Notion/Postgres等，企业内部MCP注册表成为新基础设施\n- LLM长上下文之战：Gemini 2M Token / Claude 200K / GPT-4o 128K技术选型指南，针对金融长文档(招股书/年报)场景给出最优性价比方案\n\n---\n\n## Step 1 — Process Discovery\nAsk the user to describe their current workflow:\n- What triggers it? (email, schedule, webhook, human action?)\n- What are the key steps? (list them in plain language)\n- Who (or what system) does each step today?\n- Where do errors/delays typically occur?\n- What's the desired output/outcome?\n\n**示例：报销流程的 Step 1 拆解（访谈 → 结构化盘点）**\n\n| 子步骤 | 执行人 | 输入 | 输出 | 耗时 | 是否需判断 | 系统 |\n|-------|-------|------|------|------|-----------|------|\n| 提交发票与事由 | 员工 | 发票影像 | 报销单 | 3 分钟 | 否 | OA |\n| 发票真伪校验 | 财务 | 发票号码 | 校验结果 | 5 分钟 | 否（规则明确） | 税务接口 |\n| 预算占用查询 | 财务 | 部门+科目 | 可用余额 | 4 分钟 | 否 | ERP |\n| 超标准判断 | 财务主管 | 报销单+标准 | 通过/驳回 | 8 分钟 | **是**（需解释） | 人工 |\n| 领导审批 | 部门负责人 | 报销单 | 签字 | 不定 | **是**（担责） | OA |\n| 付款 | 出纳 | 审批单 | 付款凭证 | 5 分钟 | 否 | 资金系统 |\n\n**访谈关键问题（用于发现隐性规则）**\n- 哪些步骤你会凭经验\"感觉不对\"？——这一步通常隐含未写明的规则。\n- 上一次出错是怎么发现的？——说明现有校验缺口在哪。\n- 哪些步骤你会跳过或补做？——说明流程设计与实际执行脱节。\n- 拆解结论：6 个子步骤中 4 步可自动化、2 步（超标准判断、领导审批）必须保留人工，与后续 Step 2 打分结论一致。\n\n\n### Step 2 — Automation Suitability Assessment\n\nScore the workflow across 5 dimensions:\n\n| Dimension | Score | 判断依据 | 打分示例（周报自动化） | 打分示例（客户投诉处理） | 低分时的拆解建议 |\n|-----------|-------|---------|---------------------|----------------------|\n| Repetitiveness | /10 | How often does this run identically? | 9（每周一次，步骤固定） | 4（内容差异大） | 拆出\"固定格式部分\"单独自动化 |\n| Rule-based | /10 | Are decisions clear-cut or judgment-based? | 8（汇总规则明确） | 3（需人工判断责任与情绪） | 把判断拆为\"分类\"与\"处置\"两步，只自动化分类 |\n| Data availability | /10 | Is input data structured and accessible? | 8（5 张表结构固定） | 5（邮件正文非结构化） | 增加\"结构化抽取\"前置节点后再评 |\n| Error tolerance | /10 | Can errors be caught and recovered automatically? | 7（数字错误可在复核环节发现） | 4（误判会直接损害客户关系） | 增加复核节点与灰度放行，提升容错后再评 |\n| Stakes | /10 (inverted) | Low-stakes = easier to automate | 8（内部参考，出错影响小） | 2（涉及对外承诺与赔偿） | 高风险部分保留人工，低风险子环节单独评分 |\n| **Automation Score** | /50 | >35 = High priority, 20–35 = Medium, <20 = Keep manual | **40/50 → 高优先级** | **18/50 → 暂不自动化** |\n\n**评分补充说明**\n- **Stakes 为反向计分**：风险越高得分越低。涉及资金、对外承诺、数据删除的流程，即使前四项得分高，总分也会被拉低。\n- **两例对照的意义**：周报自动化 40 分应直接推进；客户投诉处理 18 分不宜整体自动化，但可拆出\"分类 + 路由\"子环节单独自动化（该子环节约 32 分）。\n- **拆解法**：整体分数偏低时，不要放弃，而是把流程拆到子步骤重新评分——大多数流程都存在可自动化的局部环节。\n\n### Step 3 — Agentic Pipeline Design\nGenerate a detailed pipeline blueprint:\n\n```\n[Workflow]: [Name]\n[Trigger]: [webhook / cron / event / manual]\n[Agents]:\n ├── Agent 1 [Role]: [Tool 1, Tool 2] → Output: [description]\n ├── Agent 2 [Role]: [Tool 3] → Output: [description]\n └── Agent 3 [Role]: [Tool 4, Tool 5] → Output: [description]\n[Flow]: Sequential / Parallel / Conditional\n[Memory]: [ephemeral / Redis / vector DB]\n[Error Handling]: [retry / fallback agent / human escalation]\n[HITL Checkpoints]: [list high-stakes decision points]\n[Output]: [final deliverable description]\n```\n\n**Example — Lead Qualification Pipeline:**\n```\n[Workflow]: B2B Lead Qualification & Outreach\n[Trigger]: New form submission webhook\n[Agents]:\n ├── Enrichment Agent [Clearbit + LinkedIn scraper] → Company profile JSON\n ├── Scoring Agent [GPT-4o] → Lead score (0-100) + reasoning\n ├── Decision Gate [Human] → Approve for outreach? (HITL)\n └── Outreach Agent [Email API + CRM API] → Personalized email + CRM update\n[Flow]: Sequential with HITL gate\n[Memory]: PostgreSQL (lead history)\n[Error]: Retry enrichment 3x → flag for manual review\n[HITL]: Score > 80 auto-approves; 50-80 requires human review; <50 auto-rejects\n[Output]: CRM updated + email queued\n```\n\n**Example — 保险理赔单据预审 Pipeline:**\n```\n[Workflow]: 理赔单据完整性预审\n[Trigger]: 理赔系统上传事件（webhook）\n[Agents]:\n ├── 分类 Agent [OCR + 规则表] → 单据类型与置信度\n ├── 校验 Agent [规则引擎] → 缺失项清单\n ├── 决策门 [规则 + 人工] → 通过 / 退回补件 / 转人工\n └── 通知 Agent [短信 API + 工单 API] → 补件提醒 + 工单创建\n[Flow]: 先并行分类，后串行校验（Conditional）\n[Memory]: PostgreSQL（单据状态机），不含原始影像\n[Error]: OCR 置信度 < 0.85 → 强制转人工，不自动退回\n[HITL]: 涉及拒赔、金额调整、个人信息变更的一律转人工；仅限\"是否缺件\"自动判定\n[Output]: 预审结论 + 缺失项清单 + 工单号\n```\n\n**示例 — 监管报送数据校验 Pipeline（2026-09-24 新增）：**\n```\n[Workflow]: 报送数据口径校验与差错清单生成\n[Trigger]: 每日定时任务（cron，T+1 凌晨）\n[Agents]:\n ├── 抽取 Agent [SQL + 文件解析] → 原始报送数据集\n ├── 校验 Agent [规则库] → 差错项清单（含记录定位）\n ├── 归因 Agent [LLM] → 差错原因归类与整改建议\n └── 决策门 [人工] → 确认差错清单与整改责任人\n[Flow]: 串行，校验后可并行归因\n[Memory]: 仅保存差错摘要与规则版本，不落原始明细\n[Error]: 规则库版本未对齐即中止，禁止带疑点报送\n[HITL]: 任何涉及数据修改的动作必须人工确认；仅\"生成差错清单\"可自动\n[Output]: 差错清单 + 归因建议 + 待确认事项\n```\n\n**三个示例的设计差异对比（2026-09-24 新增）：**\n\n| 示例 | 风险等级 | 自动化边界 | HITL 位置 | 能否自动放行 |\n|------|---------|-----------|----------|-------------|\n| Lead Qualification | 低 | 全流程（除边缘分数） | 评分区间 50-80 | 能（>80 自动） |\n| 理赔单据预审 | 高（受监管） | 仅完整性检查 | 拒赔/金额/信息变更 | 不能（客户权益相关一律人工） |\n| 监管报送校验 | 高（合规） | 仅差错发现与归因 | 数据修改与对外报送 | 不能（对外报送须人工签发） |\n\n**判断口诀的延伸**：自动化可以覆盖**判断过程**，但不应覆盖**责任归属**；对于受监管流程，\n还要再加一句——**可以自动生成结论，但不能自动生成对外口径**。\n- Lead Qualification 属**低风险、可容忍误判**场景，因此允许 80 分以上自动放行。\n- 理赔预审属**受监管、不可自动决策**场景，自动化边界严格限定在\"完整性检查\"，任何影响客户权益的结论必须人工确认。\n- 判断依据：自动化可以覆盖**判断过程**，但不应覆盖**责任归属**。\n\n### Step 4 — Platform Recommendation\n\n| Platform | Best For | Agent Support | Self-host | Price | 学习曲线 | 典型用例 | 主要风险 | HITL 实现成本 |\n|----------|----------|--------------|-----------|-------|---------|---------|---------|\n| n8n | Technical teams, complex logic | [Yes] via AI nodes | [Yes] | Free/OSS | 较陡 | 内部数据同步、单据预审、带审批的批处理 | 自托管需自行承担运维与升级 | 中（加等待节点 + 外部审批表） |\n| Make (Integromat) | Non-technical, API integrations | Partial | [No] | ~$9+/mo | 平缓 | 跨 SaaS 数据流转、市场活动自动化 | 国内访问海外 SaaS 稳定性差 | 中（需外挂审批存储） |\n| Zapier | Simple triggers, non-technical | Partial | [No] | ~$20+/mo | 最平缓 | 表单→通知、 CRM 字段回写 | 任务量上去后成本增长快 | 中高（复杂审批链难维护） |\n| LangGraph (custom) | Complex state machines, production | [Yes] Native | [Yes] | Dev hours | 陡（需开发） | 长时间运行的对话式业务、需要中断恢复的流程 | 需自建可观测与灰度能力 | 低（原生 interrupt） |\n| CrewAI | Role-based agent teams | [Yes] Native | [Yes] | Dev hours | 中等 | 研究分析、多角色报告生成 | 角色编排调试成本较高 | 中（需自行实现中断与恢复） |\n\n\n### Step 4.5 — 2026平台详细对比表（生产选型参考）\n\n| 维度 | n8n (v1.90) | Make (2026) | Zapier (2026) | LangGraph | CrewAI |\n|------|--------------|-------------|---------------|-----------|--------|\n| **AI节点** | [Yes] 原生AI节点（OpenAI/Claude/本地LLM）| [!] 需通过HTTP节点调用 | [!] 需通过Code节点调用 | [Yes] 原生 | [Yes] 原生 |\n| **定价（月）** | 免费（OSS）/ $20/月（Cloud Pro）| $9/月（Core）~$16/月（Enterprise）| $20/月（Starter）~$69/月（Company）| Dev成本 | Dev成本 |\n| **自托管** | [Yes] Docker一键部署 | [No] 仅SaaS | [No] 仅SaaS | [Yes] | [Yes] |\n| **企业连接器** | 400+（含国内钉钉/企微）| 1000+（偏海外）| 6000+（全球最多）| 自接 | 自接 |\n| **适合场景** | 技术研发/复杂逻辑/数据敏感 | 非技术/跨部门/快速原型 | 销售/市场/简单自动化 | 复杂状态机/生产级 | 角色协作/研究分析 |\n| **最大短板** | 学习曲线陡峭 | 国内SaaS访问慢 | 国内SaaS访问慢+贵 | 需开发资源 | 需开发资源 |\n| **可观测性** | [Yes] 执行历史与重放 | [!] 仅运行日志 | [!] 仅运行日志 | 需自建（LangSmith 等） | 需自建 |\n| **人工介入（HITL）** | [!] 需手动加等待节点 | [!] 需手动加等待节点 | [!] 需手动加等待节点 | [Yes] 原生 interrupt | [!] 需自行实现 |\n| **失败回滚** | 重跑单节点 | 重跑场景 | 重跑 Zap | 依赖检查点设计 | 依赖任务设计 |\n| **国产化适配** | [Yes] 可接国产 LLM/私有化部署 | [No] | [No] | [Yes] 自行选型 | [Yes] 自行选型 |\n| **审计留痕** | [Yes] 执行历史可导出 | [!] 日志留存期有限 | [!] 日志留存期有限 | 需自行落库 | 需自行落库 |\n| **凭据治理** | [Yes] 凭据集中管理 | [!] 连接器各自持有 | [!] 连接器各自持有 | [Yes] 走环境变量/密钥管理 | [Yes] 走环境变量/密钥管理 |\n\n**选型建议（2026）**：\n- 国内团队/数据合规要求 → **n8n自托管**（数据不出境，支持国产LLM接入）\n- 海外业务/非技术团队 → **Make**（1000+连接器，学习成本低）\n- 简单场景/销售团队 → **Zapier**（即开即用，但长期成本高）\n- 复杂AI管线/生产部署 → **LangGraph**（状态持久化，支持Human-in-the-Loop）\n- 多角色协作/研究分析 → **CrewAI**（角色分工清晰，2026年中文文档完善）\n\n---\n### Step 5 — n8n Workflow JSON Spec (Sample Output)\n```json\n{\n \"name\": \"Lead Qualification Pipeline\",\n \"nodes\": [\n {\n \"name\": \"Webhook Trigger\",\n \"type\": \"n8n-nodes-base.webhook\",\n \"parameters\": { \"path\": \"lead-inbound\" }\n },\n {\n \"name\": \"Enrich Lead\",\n \"type\": \"@n8n/n8n-nodes-langchain.agent\",\n \"parameters\": {\n \"promptType\": \"define\",\n \"text\": \"Enrich this lead data using Clearbit: {{ $json.email }}\"\n }\n },\n {\n \"name\": \"Score Lead\",\n \"type\": \"@n8n/n8n-nodes-langchain.openAi\",\n \"parameters\": {\n \"resource\": \"text\",\n \"operation\": \"message\",\n \"modelId\": \"gpt-4o\",\n \"messages\": { \"values\": [{ \"content\": \"Score this lead 0-100...\" }] }\n }\n }\n ]\n}\n```\n\n### Step 6 — ROI Calculator\n\n| Metric | Before Automation | After Automation | Savings | 示例（周一销售周报） |\n|--------|------------------|-----------------|---------|-------------------|\n| Time per run | [X hours] | [Y minutes] | [Z%] | 3 小时 → 15 分钟 = 92% |\n| Runs per week | [N] | [N] | — | 1 次 |\n| Total time saved/week | — | — | [hours] | 2.75 小时 |\n| Cost saved/month | — | — | [$$$] | 11.5 小时 × 50 元 ≈ 575 元 |\n| Automation setup cost | — | — | [one-time] | 约 16 小时搭建 ≈ 800 元 |\n| **Payback period** | — | — | [weeks] | **约 6 周** |\n| 年度维护成本 | — | — | [one-time×比例] | 800 元 × 20% ≈ 160 元/年 |\n| 净年化收益 | — | — | [$$$] | 575×12 − 800 − 160 ≈ 6,140 元/年 |\n\n**ROI 计算注意事项**\n- **只计入真实节省的时间**：若节省的时间并未转化为其他产出（例如员工只是多了空闲），不宜直接折算为现金收益，应改为\"释放工时\"表述。\n- **必须计入运维成本**：工作流会因接口变更、页面改版而失效，建议按初始搭建成本的 15%-25%/年 计入维护。\n- **隐性收益单独列示**：如响应时效提升、差错率下降，可用定性描述补充，不要强行货币化。\n- **示例 2（客服工单分类路由）**：单次从 4 分钟降至 30 秒，日均 300 单 → 每日节省约 17.5 小时；但因需保留人工复核，净节省按 60% 折算更稳妥。\n\n## 可观测性与灰度发布 / Observability & Rollout（2026-09-24 新增）\n\n**最小可观测指标集（投产前必须能回答）**\n\n| 指标 | 定义 | 用途 | 建议告警阈值 |\n|------|------|------|-------------|\n| 成功率 | 成功运行数 / 总运行数 | 判断流程是否健康 | 低于 95% 告警 |\n| P95 耗时 | 95 分位运行时长 | 发现性能退化 | 超过基线 2 倍告警 |\n| 单次成本 | 单次运行的模型与接口费用 | 成本归因 | 超过预算 1.5 倍告警 |\n| 人工介入率 | 触发 HITL 的比例 | 判断自动化边界是否合理 | 超过 40% 需复盘阈值 |\n| 失败重放成功率 | 重放能否复现原失败 | 验证留痕完整性 | 无法重放即视为不合格 |\n| 数据留痕完整率 | 含输入/输出/版本/耗时/成本的记录占比 | 合规举证 | 要求 100% |\n\n**灰度发布四步**\n1. **影子运行**：只跑不落结论，输出与现状对照，验证差异率；\n2. **小流量放行**：选低风险子集（如内部部门）先行，人工 100% 复核；\n3. **阈值收紧再放开**：先用高阈值只放行最有把握的部分，逐步下调；\n4. **随时可回滚**：保留手动开关与旧流程，回滚时间目标明确（示例：10 分钟内）。\n\n> 灰度期间必须保留**新旧双轨对照**，否则无法判断自动化是否真的带来改善。\n\n---\n\n## 常见设计反模式 / Anti-Patterns\n\n| 反模式 | 症状 | 后果 | 修正方式 | 早期检测信号 |\n|-------|------|------|---------|\n| 一步全自动 | 把含判断与担责的环节也交给模型自动放行 | 出错后责任无法归属，监管与客户均不可接受 | 拆出\"可自动化的判断\"与\"必须人工的决策\"，中间设 HITL 闸口 | 流程图上找不到任何人工节点 |\n| 提示词即流程 | 用一段超长提示词描述整个业务流程 | 无法定位失败节点，改动一处影响全局 | 拆成多节点，每节点单一职责并单独可测 | 单个提示词超过一屏，且无人能说清中间态 |\n| 无置信度阈值 | 模型输出直接落库 | 低置信结果被当成确定结论 | 设阈值：高置信自动、中置信人工复核、低置信拒绝并提示 | 数据库里没有置信度字段 |\n| 静默重试 | 失败后自动重试到成功为止 | 掩盖系统性故障，成本失控 | 限制重试次数，超过即告警并保留失败现场 | 成功率长期 100% 但成本持续上升 |\n| 不留痕 | 不记录模型输入输出 | 事后无法复盘，合规检查无法举证 | 全链路留痕，含输入、输出、版本、耗时、成本 | 出问题时无法回答\"当时模型看到了什么\" |\n| 成本无上限 | 没有单次运行的成本约束 | 异常输入导致费用暴涨 | 设单次与单日成本上限，超限熔断 | 无法说出单条流程的月度成本 |\n\n> **判断口诀**：自动化可以覆盖**判断过程**，但不应覆盖**责任归属**；可以加速**执行**，但不应消除**证据**。\n\n\n## Example Interactions\n\n**User:** \"I spend 3 hours every Monday pulling sales data from 5 spreadsheets, writing a summary email, and updating our CRM. Can this be automated?\"\n\n**Skill response:** Scores the workflow (42/50 — High priority), designs a 4-agent pipeline (data collector → analyzer → email writer → CRM updater), recommends n8n as the platform (self-hostable, native AI nodes), generates a complete n8n JSON spec, and estimates 11.5 hours/month saved = ~$580 value at $50/hr.\n\n---\n\n**User:** \"I want to build a customer support triage system that reads emails, classifies them, and routes to the right team.\"\n\n**Skill response:** Designs a HITL-enabled pipeline with email reading, classification, confidence threshold (>85% auto-route, <85% human review), CRM ticket creation, and Slack notification. Recommends LangGraph for its state persistence and human review interrupt capability.\n\n---\n\n**User:** \"我们的报销单据要先核对发票真伪、再查预算、最后领导审批，能不能全自动？\"\n\n**Skill response:** 先按五维打分——重复性 9、规则明确 7、数据可得 6、错误容忍 4、风险（反向）3，合计 **29/50 属中等**。给出结论：**不宜全自动**。建议拆为两段：发票真伪校验与预算占用查询可自动化（该子流程约 36 分），领导审批必须保留人工节点。技术上推荐 n8n 自托管（涉及财务数据与发票信息，数据不出境），并明确指出自动化边界应止于\"为审批人提供核对结论\"，不得代替审批签字。\n\n---\n\n**User:** \"我有一份 200 页的年报，想自动抽取关键财务指标并生成摘要。\"\n\n**Skill response:** 评估后指出这是**长上下文单文档处理**场景，不是典型的多步骤工作流，无需引入 agent 编排。给出两条路径：① 长上下文模型全文入参 + 结构化抽取（实现快、成本可控，适合一次性或低频任务）；② 分段检索 + 抽取（适合高频批量）。提示关键风险：抽取结果必须保留**原文定位**（页码/段落），便于人工复核，避免模型生成的数字无法追溯。若涉及未公开信息，须先确认数据合规与使用授权。\n\n---\n\n**User:** \"我们有 40 多个自动化流程了，现在最大的问题是不知道哪个挂了、哪个最烧钱，怎么治理？\"\n\n**Skill response:** 先指出这是**规模化运维问题，不是设计问题**，不要靠重写流程解决。给出三步治理：① 建立流程清单与命名规范（业务域-动作-对象-版本），并为每条流程标注负责人与成本上限；② 补可观测性——链路追踪 + 节点级成本归因 + 失败重放，优先覆盖资金相关与客户可见的流程；③ 建立分级告警与熔断，按\"是否涉及资金/对外承诺\"分级。提示关键取舍：治理投入应集中在**高影响流程**，不要 40 条流程平均用力；同时建议对长期无人使用、且无合规要求的流程做下线评估。\n\n\n## 保存与外发前确认 / Save & Send Confirmation（2026-09-24 新增）\n\n1. 工作流蓝图、JSON/YAML 规格、ROI 测算先在对话中完整展示，供用户逐段预览；\n2. 用户明确确认后再写入文件或导入生产环境；未经确认不落盘、不导入；\n3. 规格文件中不得包含真实凭据，凭据一律用环境变量占位符表示；\n4. 落盘时一并记录：生成时间、平台版本、适用环境、审核人。\n\n---\n\n## 变更记录 / Changelog\n\n| 版本 | 日期 | 变更摘要 |\n|------|------|---------|\n| 3.3.6 | 2026-09-24 | 新增监管报送校验示例与三示例自动化边界对比、可观测性最小指标集与灰度发布四步、保存确认流程；评分/平台/反模式表各新增维度列；ROI 表新增维护成本与净年化收益；平台动态更新至 2026-09-24 并新增 3 条观察与 2 行维度 |\n| 3.3.5 | 2026-09-07 | 补充 2026 年平台治理与成本归因观察 |\n\n---\n\n## Notes & Constraints\n\n- Always design **HITL checkpoints** for: financial decisions, customer communications, data deletions, external API calls with side effects\n- For **regulated industries** (finance, healthcare, insurance): flag compliance requirements\n- Workflows involving PII must include data retention and access control considerations\n- Recommend starting with a **pilot workflow** (lowest risk, highest frequency) before scaling\n- Provide rollback strategies: every agentic workflow should have a manual fallback\n\n*GitHub: https://github.com/gechengling/agentic-workflow-designer*\n\nFile v3.3.6:_meta.json\n\n{\n \"ownerId\": \"kn74e704j3ygjcygnpf02rdvd185js13\",\n \"slug\": \"agentic-workflow-designer\",\n \"version\": \"3.3.6\",\n \"publishedAt\": 1790227640409\n}\n\nFile v3.3.6:skill-card.md\n\n## Description:\n\nAI-powered agentic workflow design and automation assistant for mapping multi-step processes, identifying automation opportunities, designing autonomous agent pipelines, generating workflow specs, and estimating ROI.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[gechengling](https://clawhub.ai/user/gechengling)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nOperations managers, developers, product managers, consultants, and entrepreneurs use this skill to assess business processes, design agentic workflow blueprints, choose automation platforms, generate n8n/Make/Zapier-oriented specs, and estimate automation ROI.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: Generated workflow specs may be incomplete, incorrect, or unsuitable for direct import into production automation systems.\n\nMitigation: Treat generated specs as drafts and review them before import or deployment.\n\nRisk: Automation designs may include customer-facing, financial, regulated, data-deleting, or external side-effect actions.\n\nMitigation: Add human approval checkpoints for these actions and validate compliance, retention, access control, and audit requirements.\n\n## Reference(s):\n\n- [ClawHub skill page](https://clawhub.ai/gechengling/skills/agentic-workflow-designer)\n\n## Skill Output:\n\n**Output Type(s):** [Text, Markdown, Code, Configuration, Guidance]\n\n**Output Format:** [Markdown with JSON and YAML workflow specifications]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [Produces workflow blueprints, platform recommendations, ROI estimates, and review guidance for human validation before deployment.]\n\n## Skill Version(s):\n\n3.3.6 (source: frontmatter and server release evidence)\n\n## Ethical Considerations:\n\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment.\n\nArchive v3.3.5: 3 files, 12557 bytes\n\nFiles: skill-card.md (2358b), SKILL.md (22322b), _meta.json (144b)\n\nFile v3.3.5:SKILL.md\n\n---\nname: Agentic Workflow Designer\ndescription: >\n AI-powered agentic workflow design and automation assistant — map complex multi-step\n processes, identify automation opportunities, design autonomous AI agent pipelines,\n generate n8n/Make/Zapier workflow specs, and estimate ROI. Covers enterprise automation,\n self-healing workflows, human-in-the-loop patterns, and production deployment. Keywords:\n agentic workflow, workflow automation, n8n, Make, Zapier, enterprise automation,\n AI pipeline, autonomous agent, process automation, workflow design, ROI calculator,\n HITL, 工作流设计, 流程自动化, 智能体工作流, 企业自动化, n8n工作流, 流程优化,\n 自主代理, RPA替代.\nversion: \"3.3.5\"\n---\n\n# Agentic Workflow Designer\n\n> From messy manual processes to autonomous AI pipelines — design, document, and deploy.\n\n> **⚠️ CAPABILITY NOTICE / 能力说明**\n> - **Type:** Design and advisory framework — produces workflow blueprints, JSON/YAML specs, and ROI estimates as reference material\n> - **No code is executed by this skill**; generated specs are for the user to review and import into their own environment\n> - **No persistent storage, network calls, background execution, or credential collection**\n> - **All outputs require human review before production deployment**\n> - Workflows touching PII or regulated data must include retention, access control, and audit considerations\n\n## What This Skill Does\n\nAgentic AI (AI that can autonomously execute multi-step tasks) is the #1 enterprise tech trend in 2026 with a projected $8.5B market and 40% CAGR. Yet most teams struggle to:\n- Map which workflows are actually suitable for agentic automation\n- Design reliable pipelines that don't break silently\n- Choose between n8n, Make, Zapier, or custom agent frameworks\n- Justify the ROI to business stakeholders\n\nThis skill bridges the gap between AI hype and practical workflow automation:\n\n- **Workflow Discovery** — Identify and prioritize automation opportunities in any business process\n- **Agentic Pipeline Design** — Create detailed workflow blueprints with triggers, agents, tools, and fallbacks\n- **Platform Selection** — Compare n8n / Make / Zapier / custom LangGraph for your use case\n- **Generate Workflow Specs** — Produce JSON/YAML specs importable into n8n or Make\n- **ROI Calculator** — Estimate time/cost savings from automation\n- **Human-in-the-Loop (HITL) Design** — Design appropriate checkpoints for sensitive decisions\n\n## Trigger Words\n\nAgentic workflow, automate my process, workflow automation, n8n, Make automation, Zapier flow, design a workflow, workflow design, process automation, automate with AI, AI pipeline, autonomous workflow, HITL pattern, 工作流设计, 自动化工作流, 流程自动化, 智能体工作流, 帮我设计流程, 自动化这个流程, n8n工作流, 企业自动化, RPA替代, agentic AI pipeline\n\n## Target Users\n\n- Operations managers digitizing manual business processes\n- Developers building production AI automation systems\n- Product managers scoping automation features\n- Consultants delivering workflow automation projects\n- Entrepreneurs building AI-native products\n\n## Workflow\n\n### 平台与技术动态（截至 2026-09-07）\n\n**2026-08 更新要点**：\n- **MCP 成为事实标准**：Model Context Protocol 于 2025 年底转入 Linux 基金会中立治理后，2026 年官方与社区服务器数量持续扩张，企业内部 MCP 注册表逐步成为新的基础设施层。\n- **国内合规要求趋严**：涉及个人信息与重要数据的工作流，需满足最小必要采集、境内存储与可审计要求，自托管方案的优先级上升。\n- **长上下文成本下探**：长文档场景（招股书、年报、长合同）的单位 Token 成本持续下降，使得\"全文入参 + 结构化抽取\"逐步替代早期分段检索方案。\n- **可观测性成为刚需**：生产级 agentic 工作流普遍补齐链路追踪、成本归因与失败重放能力，缺乏可观测性的方案难以通过投产评审。\n- **[2026-09 新增] 从\"能不能跑\"转向\"能不能管\"**：试点期关注功能实现，规模化后真正的瓶颈变成命名规范、版本管理、灰度发布与故障定位，设计阶段就要预留这些能力。\n- **[2026-09 新增] 选型标准从跑分转向可控性**：受监管行业（金融、医疗、政务）更看重私线部署、数据不出境与审计留痕，模型跑分高不再是唯一决策依据。\n- **[2026-09 新增] 成本归因细化到工作流**：单位 Token 成本需要能分摊到具体工作流与具体节点，否则规模化后无法判断哪些流程值得继续投入。\n\n**本期新增观察（截至 2026-09-07）**\n\n| 维度 | 变化 | 对流程设计的意义 |\n|------|------|----------------|\n| 治理 | 企业内部 MCP 注册表从\"可选\"变为\"基础件\"，工具接入需统一登记与权限控制 | 设计时把工具来源收敛到注册表，避免各流程各接一套 |\n| 运维 | 工作流数量增长后，命名与版本混乱成为主要故障源 | 约定命名规范与语义化版本，变更需留 changelog |\n| 成本 | 成本归因要求下沉到节点级 | 每个节点标注预期调用量与成本上限，超限告警 |\n| 合规 | 涉及个人信息与重要数据的流程需满足最小必要与可审计 | 在蓝图阶段即标注数据分类与留存期限 |\n\n**Step 2 新增技术评估（2026）**：\n- LangGraph v1.0生产就绪：状态机工作流/长期记忆/错误恢复三大核心能力，企业级部署支持Kubernetes自动扩缩容，GitHub Stars突破85K\n- CrewAI v1.10多智能体协作：支持6种角色类型+并行任务编排，内置20+企业级连接器（Slack/Notion/Airtable/GitHub），2026年Q1新增中文文档\n- Claude Agent SDK / OpenAI Agents SDK横向对比：工具调用准确率(94% vs 91%)/上下文利用率(78% vs 82%)/成本效率(￥0.8/千Token vs ￥1.2/千Token)三大维度全面评测\n- MCP(Model Context Protocol)生态爆发：50+官方服务器覆盖GitHub/Slack/Notion/Postgres等，企业内部MCP注册表成为新基础设施\n- LLM长上下文之战：Gemini 2M Token / Claude 200K / GPT-4o 128K技术选型指南，针对金融长文档(招股书/年报)场景给出最优性价比方案\n\n---\n\n## Step 1 — Process Discovery\nAsk the user to describe their current workflow:\n- What triggers it? (email, schedule, webhook, human action?)\n- What are the key steps? (list them in plain language)\n- Who (or what system) does each step today?\n- Where do errors/delays typically occur?\n- What's the desired output/outcome?\n\n**示例：报销流程的 Step 1 拆解（访谈 → 结构化盘点）**\n\n| 子步骤 | 执行人 | 输入 | 输出 | 耗时 | 是否需判断 | 系统 |\n|-------|-------|------|------|------|-----------|------|\n| 提交发票与事由 | 员工 | 发票影像 | 报销单 | 3 分钟 | 否 | OA |\n| 发票真伪校验 | 财务 | 发票号码 | 校验结果 | 5 分钟 | 否（规则明确） | 税务接口 |\n| 预算占用查询 | 财务 | 部门+科目 | 可用余额 | 4 分钟 | 否 | ERP |\n| 超标准判断 | 财务主管 | 报销单+标准 | 通过/驳回 | 8 分钟 | **是**（需解释） | 人工 |\n| 领导审批 | 部门负责人 | 报销单 | 签字 | 不定 | **是**（担责） | OA |\n| 付款 | 出纳 | 审批单 | 付款凭证 | 5 分钟 | 否 | 资金系统 |\n\n**访谈关键问题（用于发现隐性规则）**\n- 哪些步骤你会凭经验\"感觉不对\"？——这一步通常隐含未写明的规则。\n- 上一次出错是怎么发现的？——说明现有校验缺口在哪。\n- 哪些步骤你会跳过或补做？——说明流程设计与实际执行脱节。\n- 拆解结论：6 个子步骤中 4 步可自动化、2 步（超标准判断、领导审批）必须保留人工，与后续 Step 2 打分结论一致。\n\n\n### Step 2 — Automation Suitability Assessment\n\nScore the workflow across 5 dimensions:\n\n| Dimension | Score | 判断依据 | 打分示例（周报自动化） | 打分示例（客户投诉处理） |\n|-----------|-------|---------|---------------------|----------------------|\n| Repetitiveness | /10 | How often does this run identically? | 9（每周一次，步骤固定） | 4（内容差异大） |\n| Rule-based | /10 | Are decisions clear-cut or judgment-based? | 8（汇总规则明确） | 3（需人工判断责任与情绪） |\n| Data availability | /10 | Is input data structured and accessible? | 8（5 张表结构固定） | 5（邮件正文非结构化） |\n| Error tolerance | /10 | Can errors be caught and recovered automatically? | 7（数字错误可在复核环节发现） | 4（误判会直接损害客户关系） |\n| Stakes | /10 (inverted) | Low-stakes = easier to automate | 8（内部参考，出错影响小） | 2（涉及对外承诺与赔偿） |\n| **Automation Score** | /50 | >35 = High priority, 20–35 = Medium, <20 = Keep manual | **40/50 → 高优先级** | **18/50 → 暂不自动化** |\n\n**评分补充说明**\n- **Stakes 为反向计分**：风险越高得分越低。涉及资金、对外承诺、数据删除的流程，即使前四项得分高，总分也会被拉低。\n- **两例对照的意义**：周报自动化 40 分应直接推进；客户投诉处理 18 分不宜整体自动化，但可拆出\"分类 + 路由\"子环节单独自动化（该子环节约 32 分）。\n- **拆解法**：整体分数偏低时，不要放弃，而是把流程拆到子步骤重新评分——大多数流程都存在可自动化的局部环节。\n\n### Step 3 — Agentic Pipeline Design\nGenerate a detailed pipeline blueprint:\n\n```\n[Workflow]: [Name]\n[Trigger]: [webhook / cron / event / manual]\n[Agents]:\n ├── Agent 1 [Role]: [Tool 1, Tool 2] → Output: [description]\n ├── Agent 2 [Role]: [Tool 3] → Output: [description]\n └── Agent 3 [Role]: [Tool 4, Tool 5] → Output: [description]\n[Flow]: Sequential / Parallel / Conditional\n[Memory]: [ephemeral / Redis / vector DB]\n[Error Handling]: [retry / fallback agent / human escalation]\n[HITL Checkpoints]: [list high-stakes decision points]\n[Output]: [final deliverable description]\n```\n\n**Example — Lead Qualification Pipeline:**\n```\n[Workflow]: B2B Lead Qualification & Outreach\n[Trigger]: New form submission webhook\n[Agents]:\n ├── Enrichment Agent [Clearbit + LinkedIn scraper] → Company profile JSON\n ├── Scoring Agent [GPT-4o] → Lead score (0-100) + reasoning\n ├── Decision Gate [Human] → Approve for outreach? (HITL)\n └── Outreach Agent [Email API + CRM API] → Personalized email + CRM update\n[Flow]: Sequential with HITL gate\n[Memory]: PostgreSQL (lead history)\n[Error]: Retry enrichment 3x → flag for manual review\n[HITL]: Score > 80 auto-approves; 50-80 requires human review; <50 auto-rejects\n[Output]: CRM updated + email queued\n```\n\n**Example — 保险理赔单据预审 Pipeline:**\n```\n[Workflow]: 理赔单据完整性预审\n[Trigger]: 理赔系统上传事件（webhook）\n[Agents]:\n ├── 分类 Agent [OCR + 规则表] → 单据类型与置信度\n ├── 校验 Agent [规则引擎] → 缺失项清单\n ├── 决策门 [规则 + 人工] → 通过 / 退回补件 / 转人工\n └── 通知 Agent [短信 API + 工单 API] → 补件提醒 + 工单创建\n[Flow]: 先并行分类，后串行校验（Conditional）\n[Memory]: PostgreSQL（单据状态机），不含原始影像\n[Error]: OCR 置信度 < 0.85 → 强制转人工，不自动退回\n[HITL]: 涉及拒赔、金额调整、个人信息变更的一律转人工；仅限\"是否缺件\"自动判定\n[Output]: 预审结论 + 缺失项清单 + 工单号\n```\n\n**两个示例的设计差异**\n- Lead Qualification 属**低风险、可容忍误判**场景，因此允许 80 分以上自动放行。\n- 理赔预审属**受监管、不可自动决策**场景，自动化边界严格限定在\"完整性检查\"，任何影响客户权益的结论必须人工确认。\n- 判断依据：自动化可以覆盖**判断过程**，但不应覆盖**责任归属**。\n\n### Step 4 — Platform Recommendation\n\n| Platform | Best For | Agent Support | Self-host | Price | 学习曲线 | 典型用例 | 主要风险 |\n|----------|----------|--------------|-----------|-------|---------|---------|---------|\n| n8n | Technical teams, complex logic | [Yes] via AI nodes | [Yes] | Free/OSS | 较陡 | 内部数据同步、单据预审、带审批的批处理 | 自托管需自行承担运维与升级 |\n| Make (Integromat) | Non-technical, API integrations | Partial | [No] | ~$9+/mo | 平缓 | 跨 SaaS 数据流转、市场活动自动化 | 国内访问海外 SaaS 稳定性差 |\n| Zapier | Simple triggers, non-technical | Partial | [No] | ~$20+/mo | 最平缓 | 表单→通知、 CRM 字段回写 | 任务量上去后成本增长快 |\n| LangGraph (custom) | Complex state machines, production | [Yes] Native | [Yes] | Dev hours | 陡（需开发） | 长时间运行的对话式业务、需要中断恢复的流程 | 需自建可观测与灰度能力 |\n| CrewAI | Role-based agent teams | [Yes] Native | [Yes] | Dev hours | 中等 | 研究分析、多角色报告生成 | 角色编排调试成本较高 |\n\n\n### Step 4.5 — 2026平台详细对比表（生产选型参考）\n\n| 维度 | n8n (v1.90) | Make (2026) | Zapier (2026) | LangGraph | CrewAI |\n|------|--------------|-------------|---------------|-----------|--------|\n| **AI节点** | [Yes] 原生AI节点（OpenAI/Claude/本地LLM）| [!] 需通过HTTP节点调用 | [!] 需通过Code节点调用 | [Yes] 原生 | [Yes] 原生 |\n| **定价（月）** | 免费（OSS）/ $20/月（Cloud Pro）| $9/月（Core）~$16/月（Enterprise）| $20/月（Starter）~$69/月（Company）| Dev成本 | Dev成本 |\n| **自托管** | [Yes] Docker一键部署 | [No] 仅SaaS | [No] 仅SaaS | [Yes] | [Yes] |\n| **企业连接器** | 400+（含国内钉钉/企微）| 1000+（偏海外）| 6000+（全球最多）| 自接 | 自接 |\n| **适合场景** | 技术研发/复杂逻辑/数据敏感 | 非技术/跨部门/快速原型 | 销售/市场/简单自动化 | 复杂状态机/生产级 | 角色协作/研究分析 |\n| **最大短板** | 学习曲线陡峭 | 国内SaaS访问慢 | 国内SaaS访问慢+贵 | 需开发资源 | 需开发资源 |\n| **可观测性** | [Yes] 执行历史与重放 | [!] 仅运行日志 | [!] 仅运行日志 | 需自建（LangSmith 等） | 需自建 |\n| **人工介入（HITL）** | [!] 需手动加等待节点 | [!] 需手动加等待节点 | [!] 需手动加等待节点 | [Yes] 原生 interrupt | [!] 需自行实现 |\n| **失败回滚** | 重跑单节点 | 重跑场景 | 重跑 Zap | 依赖检查点设计 | 依赖任务设计 |\n| **国产化适配** | [Yes] 可接国产 LLM/私有化部署 | [No] | [No] | [Yes] 自行选型 | [Yes] 自行选型 |\n\n**选型建议（2026）**：\n- 国内团队/数据合规要求 → **n8n自托管**（数据不出境，支持国产LLM接入）\n- 海外业务/非技术团队 → **Make**（1000+连接器，学习成本低）\n- 简单场景/销售团队 → **Zapier**（即开即用，但长期成本高）\n- 复杂AI管线/生产部署 → **LangGraph**（状态持久化，支持Human-in-the-Loop）\n- 多角色协作/研究分析 → **CrewAI**（角色分工清晰，2026年中文文档完善）\n\n---\n### Step 5 — n8n Workflow JSON Spec (Sample Output)\n```json\n{\n \"name\": \"Lead Qualification Pipeline\",\n \"nodes\": [\n {\n \"name\": \"Webhook Trigger\",\n \"type\": \"n8n-nodes-base.webhook\",\n \"parameters\": { \"path\": \"lead-inbound\" }\n },\n {\n \"name\": \"Enrich Lead\",\n \"type\": \"@n8n/n8n-nodes-langchain.agent\",\n \"parameters\": {\n \"promptType\": \"define\",\n \"text\": \"Enrich this lead data using Clearbit: {{ $json.email }}\"\n }\n },\n {\n \"name\": \"Score Lead\",\n \"type\": \"@n8n/n8n-nodes-langchain.openAi\",\n \"parameters\": {\n \"resource\": \"text\",\n \"operation\": \"message\",\n \"modelId\": \"gpt-4o\",\n \"messages\": { \"values\": [{ \"content\": \"Score this lead 0-100...\" }] }\n }\n }\n ]\n}\n```\n\n### Step 6 — ROI Calculator\n\n| Metric | Before Automation | After Automation | Savings | 示例（周一销售周报） |\n|--------|------------------|-----------------|---------|-------------------|\n| Time per run | [X hours] | [Y minutes] | [Z%] | 3 小时 → 15 分钟 = 92% |\n| Runs per week | [N] | [N] | — | 1 次 |\n| Total time saved/week | — | — | [hours] | 2.75 小时 |\n| Cost saved/month | — | — | [$$$] | 11.5 小时 × 50 元 ≈ 575 元 |\n| Automation setup cost | — | — | [one-time] | 约 16 小时搭建 ≈ 800 元 |\n| **Payback period** | — | — | [weeks] | **约 6 周** |\n\n**ROI 计算注意事项**\n- **只计入真实节省的时间**：若节省的时间并未转化为其他产出（例如员工只是多了空闲），不宜直接折算为现金收益，应改为\"释放工时\"表述。\n- **必须计入运维成本**：工作流会因接口变更、页面改版而失效，建议按初始搭建成本的 15%-25%/年 计入维护。\n- **隐性收益单独列示**：如响应时效提升、差错率下降，可用定性描述补充，不要强行货币化。\n- **示例 2（客服工单分类路由）**：单次从 4 分钟降至 30 秒，日均 300 单 → 每日节省约 17.5 小时；但因需保留人工复核，净节省按 60% 折算更稳妥。\n\n## 常见设计反模式 / Anti-Patterns\n\n| 反模式 | 症状 | 后果 | 修正方式 |\n|-------|------|------|---------|\n| 一步全自动 | 把含判断与担责的环节也交给模型自动放行 | 出错后责任无法归属，监管与客户均不可接受 | 拆出\"可自动化的判断\"与\"必须人工的决策\"，中间设 HITL 闸口 |\n| 提示词即流程 | 用一段超长提示词描述整个业务流程 | 无法定位失败节点，改动一处影响全局 | 拆成多节点，每节点单一职责并单独可测 |\n| 无置信度阈值 | 模型输出直接落库 | 低置信结果被当成确定结论 | 设阈值：高置信自动、中置信人工复核、低置信拒绝并提示 |\n| 静默重试 | 失败后自动重试到成功为止 | 掩盖系统性故障，成本失控 | 限制重试次数，超过即告警并保留失败现场 |\n| 不留痕 | 不记录模型输入输出 | 事后无法复盘，合规检查无法举证 | 全链路留痕，含输入、输出、版本、耗时、成本 |\n| 成本无上限 | 没有单次运行的成本约束 | 异常输入导致费用暴涨 | 设单次与单日成本上限，超限熔断 |\n\n> **判断口诀**：自动化可以覆盖**判断过程**，但不应覆盖**责任归属**；可以加速**执行**，但不应消除**证据**。\n\n\n## Example Interactions\n\n**User:** \"I spend 3 hours every Monday pulling sales data from 5 spreadsheets, writing a summary email, and updating our CRM. Can this be automated?\"\n\n**Skill response:** Scores the workflow (42/50 — High priority), designs a 4-agent pipeline (data collector → analyzer → email writer → CRM updater), recommends n8n as the platform (self-hostable, native AI nodes), generates a complete n8n JSON spec, and estimates 11.5 hours/month saved = ~$580 value at $50/hr.\n\n---\n\n**User:** \"I want to build a customer support triage system that reads emails, classifies them, and routes to the right team.\"\n\n**Skill response:** Designs a HITL-enabled pipeline with email reading, classification, confidence threshold (>85% auto-route, <85% human review), CRM ticket creation, and Slack notification. Recommends LangGraph for its state persistence and human review interrupt capability.\n\n---\n\n**User:** \"我们的报销单据要先核对发票真伪、再查预算、最后领导审批，能不能全自动？\"\n\n**Skill response:** 先按五维打分——重复性 9、规则明确 7、数据可得 6、错误容忍 4、风险（反向）3，合计 **29/50 属中等**。给出结论：**不宜全自动**。建议拆为两段：发票真伪校验与预算占用查询可自动化（该子流程约 36 分），领导审批必须保留人工节点。技术上推荐 n8n 自托管（涉及财务数据与发票信息，数据不出境），并明确指出自动化边界应止于\"为审批人提供核对结论\"，不得代替审批签字。\n\n---\n\n**User:** \"我有一份 200 页的年报，想自动抽取关键财务指标并生成摘要。\"\n\n**Skill response:** 评估后指出这是**长上下文单文档处理**场景，不是典型的多步骤工作流，无需引入 agent 编排。给出两条路径：① 长上下文模型全文入参 + 结构化抽取（实现快、成本可控，适合一次性或低频任务）；② 分段检索 + 抽取（适合高频批量）。提示关键风险：抽取结果必须保留**原文定位**（页码/段落），便于人工复核，避免模型生成的数字无法追溯。若涉及未公开信息，须先确认数据合规与使用授权。\n\n---\n\n**User:** \"我们有 40 多个自动化流程了，现在最大的问题是不知道哪个挂了、哪个最烧钱，怎么治理？\"\n\n**Skill response:** 先指出这是**规模化运维问题，不是设计问题**，不要靠重写流程解决。给出三步治理：① 建立流程清单与命名规范（业务域-动作-对象-版本），并为每条流程标注负责人与成本上限；② 补可观测性——链路追踪 + 节点级成本归因 + 失败重放，优先覆盖资金相关与客户可见的流程；③ 建立分级告警与熔断，按\"是否涉及资金/对外承诺\"分级。提示关键取舍：治理投入应集中在**高影响流程**，不要 40 条流程平均用力；同时建议对长期无人使用、且无合规要求的流程做下线评估。\n\n\n## Notes & Constraints\n\n- Always design **HITL checkpoints** for: financial decisions, customer communications, data deletions, external API calls with side effects\n- For **regulated industries** (finance, healthcare, insurance): flag compliance requirements\n- Workflows involving PII must include data retention and access control considerations\n- Recommend starting with a **pilot workflow** (lowest risk, highest frequency) before scaling\n- Provide rollback strategies: every agentic workflow should have a manual fallback\n\n*GitHub: https://github.com/gechengling/agentic-workflow-designer*\n\nFile v3.3.5:_meta.json\n\n{\n \"ownerId\": \"kn74e704j3ygjcygnpf02rdvd185js13\",\n \"slug\": \"agentic-workflow-designer\",\n \"version\": \"3.3.5\",\n \"publishedAt\": 1788748220099\n}\n\nFile v3.3.5:skill-card.md\n\n## Description:\n\nAI-powered agentic workflow design and automation assistant for mapping multi-step processes, identifying automation opportunities, designing autonomous AI agent pipelines, generating n8n, Make, and Zapier workflow specs, and estimating ROI.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[gechengling](https://clawhub.ai/user/gechengling)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nOperations managers, developers, product managers, consultants, and entrepreneurs use this skill to assess automation fit, design agentic workflow blueprints, compare automation platforms, plan human review checkpoints, and estimate ROI before implementation.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: Generated workflow specifications may trigger side effects if imported without review.\n\nMitigation: Review n8n, Make, Zapier, and agent specs before import, and require explicit human approval for financial decisions, customer communications, data deletion, regulated data, and other side-effecting actions.\n\nRisk: Workflow designs may involve PII or regulated data.\n\nMitigation: Add retention, access control, audit, and data minimization requirements before production deployment.\n\nRisk: Automation recommendations and ROI estimates may be incomplete for a specific operating environment.\n\nMitigation: Validate recommendations with a low-risk pilot workflow and update platform, cost, and fallback assumptions before scaling.\n\n## Reference(s):\n\n- [ClawHub skill page](https://clawhub.ai/gechengling/skills/agentic-workflow-designer)\n\n## Skill Output:\n\n**Output Type(s):** [Text, Markdown, Code, Configuration, Guidance]\n\n**Output Format:** [Markdown guidance with workflow blueprints, JSON/YAML workflow specifications, comparison tables, and ROI estimates]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [Generated workflow specs are reference material and require human review before import or production deployment.]\n\n## Skill Version(s):\n\n3.3.5 (source: frontmatter and release evidence)\n\n## Ethical Considerations:\n\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment.\n\nArchive v3.3.4: 3 files, 10391 bytes\n\nFiles: skill-card.md (2200b), SKILL.md (17619b), _meta.json (144b)\n\nFile v3.3.4:SKILL.md\n\n---\nname: Agentic Workflow Designer\ndescription: >\n AI-powered agentic workflow design and automation assistant — map complex multi-step\n processes, identify automation opportunities, design autonomous AI agent pipelines,\n generate n8n/Make/Zapier workflow specs, and estimate ROI. Covers enterprise automation,\n self-healing workflows, human-in-the-loop patterns, and production deployment. Keywords:\n agentic workflow, workflow automation, n8n, Make, Zapier, enterprise automation,\n AI pipeline, autonomous agent, process automation, workflow design, ROI calculator,\n HITL, 工作流设计, 流程自动化, 智能体工作流, 企业自动化, n8n工作流, 流程优化,\n 自主代理, RPA替代.\nversion: \"3.3.4\"\n---\n\n# Agentic Workflow Designer\n\n> From messy manual processes to autonomous AI pipelines — design, document, and deploy.\n\n> **⚠️ CAPABILITY NOTICE / 能力说明**\n> - **Type:** Design and advisory framework — produces workflow blueprints, JSON/YAML specs, and ROI estimates as reference material\n> - **No code is executed by this skill**; generated specs are for the user to review and import into their own environment\n> - **No persistent storage, network calls, background execution, or credential collection**\n> - **All outputs require human review before production deployment**\n> - Workflows touching PII or regulated data must include retention, access control, and audit considerations\n\n## What This Skill Does\n\nAgentic AI (AI that can autonomously execute multi-step tasks) is the #1 enterprise tech trend in 2026 with a projected $8.5B market and 40% CAGR. Yet most teams struggle to:\n- Map which workflows are actually suitable for agentic automation\n- Design reliable pipelines that don't break silently\n- Choose between n8n, Make, Zapier, or custom agent frameworks\n- Justify the ROI to business stakeholders\n\nThis skill bridges the gap between AI hype and practical workflow automation:\n\n- **Workflow Discovery** — Identify and prioritize automation opportunities in any business process\n- **Agentic Pipeline Design** — Create detailed workflow blueprints with triggers, agents, tools, and fallbacks\n- **Platform Selection** — Compare n8n / Make / Zapier / custom LangGraph for your use case\n- **Generate Workflow Specs** — Produce JSON/YAML specs importable into n8n or Make\n- **ROI Calculator** — Estimate time/cost savings from automation\n- **Human-in-the-Loop (HITL) Design** — Design appropriate checkpoints for sensitive decisions\n\n## Trigger Words\n\nAgentic workflow, automate my process, workflow automation, n8n, Make automation, Zapier flow, design a workflow, workflow design, process automation, automate with AI, AI pipeline, autonomous workflow, HITL pattern, 工作流设计, 自动化工作流, 流程自动化, 智能体工作流, 帮我设计流程, 自动化这个流程, n8n工作流, 企业自动化, RPA替代, agentic AI pipeline\n\n## Target Users\n\n- Operations managers digitizing manual business processes\n- Developers building production AI automation systems\n- Product managers scoping automation features\n- Consultants delivering workflow automation projects\n- Entrepreneurs building AI-native products\n\n## Workflow\n\n### 平台与技术动态（截至 2026-08-31）\n\n**2026-08 更新要点**：\n- **MCP 成为事实标准**：Model Context Protocol 于 2025 年底转入 Linux 基金会中立治理后，2026 年官方与社区服务器数量持续扩张，企业内部 MCP 注册表逐步成为新的基础设施层。\n- **国内合规要求趋严**：涉及个人信息与重要数据的工作流，需满足最小必要采集、境内存储与可审计要求，自托管方案的优先级上升。\n- **长上下文成本下探**：长文档场景（招股书、年报、长合同）的单位 Token 成本持续下降，使得\"全文入参 + 结构化抽取\"逐步替代早期分段检索方案。\n- **可观测性成为刚需**：生产级 agentic 工作流普遍补齐链路追踪、成本归因与失败重放能力，缺乏可观测性的方案难以通过投产评审。\n\n**Step 2 新增技术评估（2026）**：\n- LangGraph v1.0生产就绪：状态机工作流/长期记忆/错误恢复三大核心能力，企业级部署支持Kubernetes自动扩缩容，GitHub Stars突破85K\n- CrewAI v1.10多智能体协作：支持6种角色类型+并行任务编排，内置20+企业级连接器（Slack/Notion/Airtable/GitHub），2026年Q1新增中文文档\n- Claude Agent SDK / OpenAI Agents SDK横向对比：工具调用准确率(94% vs 91%)/上下文利用率(78% vs 82%)/成本效率(￥0.8/千Token vs ￥1.2/千Token)三大维度全面评测\n- MCP(Model Context Protocol)生态爆发：50+官方服务器覆盖GitHub/Slack/Notion/Postgres等，企业内部MCP注册表成为新基础设施\n- LLM长上下文之战：Gemini 2M Token / Claude 200K / GPT-4o 128K技术选型指南，针对金融长文档(招股书/年报)场景给出最优性价比方案\n\n---\n\n## Step 1 — Process Discovery\nAsk the user to describe their current workflow:\n- What triggers it? (email, schedule, webhook, human action?)\n- What are the key steps? (list them in plain language)\n- Who (or what system) does each step today?\n- Where do errors/delays typically occur?\n- What's the desired output/outcome?\n\n### Step 2 — Automation Suitability Assessment\n\nScore the workflow across 5 dimensions:\n\n| Dimension | Score | 判断依据 | 打分示例（周报自动化） | 打分示例（客户投诉处理） |\n|-----------|-------|---------|---------------------|----------------------|\n| Repetitiveness | /10 | How often does this run identically? | 9（每周一次，步骤固定） | 4（内容差异大） |\n| Rule-based | /10 | Are decisions clear-cut or judgment-based? | 8（汇总规则明确） | 3（需人工判断责任与情绪） |\n| Data availability | /10 | Is input data structured and accessible? | 8（5 张表结构固定） | 5（邮件正文非结构化） |\n| Error tolerance | /10 | Can errors be caught and recovered automatically? | 7（数字错误可在复核环节发现） | 4（误判会直接损害客户关系） |\n| Stakes | /10 (inverted) | Low-stakes = easier to automate | 8（内部参考，出错影响小） | 2（涉及对外承诺与赔偿） |\n| **Automation Score** | /50 | >35 = High priority, 20–35 = Medium, <20 = Keep manual | **40/50 → 高优先级** | **18/50 → 暂不自动化** |\n\n**评分补充说明**\n- **Stakes 为反向计分**：风险越高得分越低。涉及资金、对外承诺、数据删除的流程，即使前四项得分高，总分也会被拉低。\n- **两例对照的意义**：周报自动化 40 分应直接推进；客户投诉处理 18 分不宜整体自动化，但可拆出\"分类 + 路由\"子环节单独自动化（该子环节约 32 分）。\n- **拆解法**：整体分数偏低时，不要放弃，而是把流程拆到子步骤重新评分——大多数流程都存在可自动化的局部环节。\n\n### Step 3 — Agentic Pipeline Design\nGenerate a detailed pipeline blueprint:\n\n```\n[Workflow]: [Name]\n[Trigger]: [webhook / cron / event / manual]\n[Agents]:\n ├── Agent 1 [Role]: [Tool 1, Tool 2] → Output: [description]\n ├── Agent 2 [Role]: [Tool 3] → Output: [description]\n └── Agent 3 [Role]: [Tool 4, Tool 5] → Output: [description]\n[Flow]: Sequential / Parallel / Conditional\n[Memory]: [ephemeral / Redis / vector DB]\n[Error Handling]: [retry / fallback agent / human escalation]\n[HITL Checkpoints]: [list high-stakes decision points]\n[Output]: [final deliverable description]\n```\n\n**Example — Lead Qualification Pipeline:**\n```\n[Workflow]: B2B Lead Qualification & Outreach\n[Trigger]: New form submission webhook\n[Agents]:\n ├── Enrichment Agent [Clearbit + LinkedIn scraper] → Company profile JSON\n ├── Scoring Agent [GPT-4o] → Lead score (0-100) + reasoning\n ├── Decision Gate [Human] → Approve for outreach? (HITL)\n └── Outreach Agent [Email API + CRM API] → Personalized email + CRM update\n[Flow]: Sequential with HITL gate\n[Memory]: PostgreSQL (lead history)\n[Error]: Retry enrichment 3x → flag for manual review\n[HITL]: Score > 80 auto-approves; 50-80 requires human review; <50 auto-rejects\n[Output]: CRM updated + email queued\n```\n\n**Example — 保险理赔单据预审 Pipeline:**\n```\n[Workflow]: 理赔单据完整性预审\n[Trigger]: 理赔系统上传事件（webhook）\n[Agents]:\n ├── 分类 Agent [OCR + 规则表] → 单据类型与置信度\n ├── 校验 Agent [规则引擎] → 缺失项清单\n ├── 决策门 [规则 + 人工] → 通过 / 退回补件 / 转人工\n └── 通知 Agent [短信 API + 工单 API] → 补件提醒 + 工单创建\n[Flow]: 先并行分类，后串行校验（Conditional）\n[Memory]: PostgreSQL（单据状态机），不含原始影像\n[Error]: OCR 置信度 < 0.85 → 强制转人工，不自动退回\n[HITL]: 涉及拒赔、金额调整、个人信息变更的一律转人工；仅限\"是否缺件\"自动判定\n[Output]: 预审结论 + 缺失项清单 + 工单号\n```\n\n**两个示例的设计差异**\n- Lead Qualification 属**低风险、可容忍误判**场景，因此允许 80 分以上自动放行。\n- 理赔预审属**受监管、不可自动决策**场景，自动化边界严格限定在\"完整性检查\"，任何影响客户权益的结论必须人工确认。\n- 判断依据：自动化可以覆盖**判断过程**，但不应覆盖**责任归属**。\n\n### Step 4 — Platform Recommendation\n\n| Platform | Best For | Agent Support | Self-host | Price | 学习曲线 | 典型用例 | 主要风险 |\n|----------|----------|--------------|-----------|-------|---------|---------|---------|\n| n8n | Technical teams, complex logic | [Yes] via AI nodes | [Yes] | Free/OSS | 较陡 | 内部数据同步、单据预审、带审批的批处理 | 自托管需自行承担运维与升级 |\n| Make (Integromat) | Non-technical, API integrations | Partial | [No] | ~$9+/mo | 平缓 | 跨 SaaS 数据流转、市场活动自动化 | 国内访问海外 SaaS 稳定性差 |\n| Zapier | Simple triggers, non-technical | Partial | [No] | ~$20+/mo | 最平缓 | 表单→通知、 CRM 字段回写 | 任务量上去后成本增长快 |\n| LangGraph (custom) | Complex state machines, production | [Yes] Native | [Yes] | Dev hours | 陡（需开发） | 长时间运行的对话式业务、需要中断恢复的流程 | 需自建可观测与灰度能力 |\n| CrewAI | Role-based agent teams | [Yes] Native | [Yes] | Dev hours | 中等 | 研究分析、多角色报告生成 | 角色编排调试成本较高 |\n\n\n### Step 4.5 — 2026平台详细对比表（生产选型参考）\n\n| 维度 | n8n (v1.90) | Make (2026) | Zapier (2026) | LangGraph | CrewAI |\n|------|--------------|-------------|---------------|-----------|--------|\n| **AI节点** | [Yes] 原生AI节点（OpenAI/Claude/本地LLM）| [!] 需通过HTTP节点调用 | [!] 需通过Code节点调用 | [Yes] 原生 | [Yes] 原生 |\n| **定价（月）** | 免费（OSS）/ $20/月（Cloud Pro）| $9/月（Core）~$16/月（Enterprise）| $20/月（Starter）~$69/月（Company）| Dev成本 | Dev成本 |\n| **自托管** | [Yes] Docker一键部署 | [No] 仅SaaS | [No] 仅SaaS | [Yes] | [Yes] |\n| **企业连接器** | 400+（含国内钉钉/企微）| 1000+（偏海外）| 6000+（全球最多）| 自接 | 自接 |\n| **适合场景** | 技术研发/复杂逻辑/数据敏感 | 非技术/跨部门/快速原型 | 销售/市场/简单自动化 | 复杂状态机/生产级 | 角色协作/研究分析 |\n| **最大短板** | 学习曲线陡峭 | 国内SaaS访问慢 | 国内SaaS访问慢+贵 | 需开发资源 | 需开发资源 |\n| **可观测性** | [Yes] 执行历史与重放 | [!] 仅运行日志 | [!] 仅运行日志 | 需自建（LangSmith 等） | 需自建 |\n| **人工介入（HITL）** | [!] 需手动加等待节点 | [!] 需手动加等待节点 | [!] 需手动加等待节点 | [Yes] 原生 interrupt | [!] 需自行实现 |\n| **失败回滚** | 重跑单节点 | 重跑场景 | 重跑 Zap | 依赖检查点设计 | 依赖任务设计 |\n| **国产化适配** | [Yes] 可接国产 LLM/私有化部署 | [No] | [No] | [Yes] 自行选型 | [Yes] 自行选型 |\n\n**选型建议（2026）**：\n- 国内团队/数据合规要求 → **n8n自托管**（数据不出境，支持国产LLM接入）\n- 海外业务/非技术团队 → **Make**（1000+连接器，学习成本低）\n- 简单场景/销售团队 → **Zapier**（即开即用，但长期成本高）\n- 复杂AI管线/生产部署 → **LangGraph**（状态持久化，支持Human-in-the-Loop）\n- 多角色协作/研究分析 → **CrewAI**（角色分工清晰，2026年中文文档完善）\n\n---\n### Step 5 — n8n Workflow JSON Spec (Sample Output)\n```json\n{\n \"name\": \"Lead Qualification Pipeline\",\n \"nodes\": [\n {\n \"name\": \"Webhook Trigger\",\n \"type\": \"n8n-nodes-base.webhook\",\n \"parameters\": { \"path\": \"lead-inbound\" }\n },\n {\n \"name\": \"Enrich Lead\",\n \"type\": \"@n8n/n8n-nodes-langchain.agent\",\n \"parameters\": {\n \"promptType\": \"define\",\n \"text\": \"Enrich this lead data using Clearbit: {{ $json.email }}\"\n }\n },\n {\n \"name\": \"Score Lead\",\n \"type\": \"@n8n/n8n-nodes-langchain.openAi\",\n \"parameters\": {\n \"resource\": \"text\",\n \"operation\": \"message\",\n \"modelId\": \"gpt-4o\",\n \"messages\": { \"values\": [{ \"content\": \"Score this lead 0-100...\" }] }\n }\n }\n ]\n}\n```\n\n### Step 6 — ROI Calculator\n\n| Metric | Before Automation | After Automation | Savings | 示例（周一销售周报） |\n|--------|------------------|-----------------|---------|-------------------|\n| Time per run | [X hours] | [Y minutes] | [Z%] | 3 小时 → 15 分钟 = 92% |\n| Runs per week | [N] | [N] | — | 1 次 |\n| Total time saved/week | — | — | [hours] | 2.75 小时 |\n| Cost saved/month | — | — | [$$$] | 11.5 小时 × 50 元 ≈ 575 元 |\n| Automation setup cost | — | — | [one-time] | 约 16 小时搭建 ≈ 800 元 |\n| **Payback period** | — | — | [weeks] | **约 6 周** |\n\n**ROI 计算注意事项**\n- **只计入真实节省的时间**：若节省的时间并未转化为其他产出（例如员工只是多了空闲），不宜直接折算为现金收益，应改为\"释放工时\"表述。\n- **必须计入运维成本**：工作流会因接口变更、页面改版而失效，建议按初始搭建成本的 15%-25%/年 计入维护。\n- **隐性收益单独列示**：如响应时效提升、差错率下降，可用定性描述补充，不要强行货币化。\n- **示例 2（客服工单分类路由）**：单次从 4 分钟降至 30 秒，日均 300 单 → 每日节省约 17.5 小时；但因需保留人工复核，净节省按 60% 折算更稳妥。\n\n## Example Interactions\n\n**User:** \"I spend 3 hours every Monday pulling sales data from 5 spreadsheets, writing a summary email, and updating our CRM. Can this be automated?\"\n\n**Skill response:** Scores the workflow (42/50 — High priority), designs a 4-agent pipeline (data collector → analyzer → email writer → CRM updater), recommends n8n as the platform (self-hostable, native AI nodes), generates a complete n8n JSON spec, and estimates 11.5 hours/month saved = ~$580 value at $50/hr.\n\n---\n\n**User:** \"I want to build a customer support triage system that reads emails, classifies them, and routes to the right team.\"\n\n**Skill response:** Designs a HITL-enabled pipeline with email reading, classification, confidence threshold (>85% auto-route, <85% human review), CRM ticket creation, and Slack notification. Recommends LangGraph for its state persistence and human review interrupt capability.\n\n---\n\n**User:** \"我们的报销单据要先核对发票真伪、再查预算、最后领导审批，能不能全自动？\"\n\n**Skill response:** 先按五维打分——重复性 9、规则明确 7、数据可得 6、错误容忍 4、风险（反向）3，合计 **29/50 属中等**。给出结论：**不宜全自动**。建议拆为两段：发票真伪校验与预算占用查询可自动化（该子流程约 36 分），领导审批必须保留人工节点。技术上推荐 n8n 自托管（涉及财务数据与发票信息，数据不出境），并明确指出自动化边界应止于\"为审批人提供核对结论\"，不得代替审批签字。\n\n---\n\n**User:** \"我有一份 200 页的年报，想自动抽取关键财务指标并生成摘要。\"\n\n**Skill response:** 评估后指出这是**长上下文单文档处理**场景，不是典型的多步骤工作流，无需引入 agent 编排。给出两条路径：① 长上下文模型全文入参 + 结构化抽取（实现快、成本可控，适合一次性或低频任务）；② 分段检索 + 抽取（适合高频批量）。提示关键风险：抽取结果必须保留**原文定位**（页码/段落），便于人工复核，避免模型生成的数字无法追溯。若涉及未公开信息，须先确认数据合规与使用授权。\n\n## Notes & Constraints\n\n- Always design **HITL checkpoints** for: financial decisions, customer communications, data deletions, external API calls with side effects\n- For **regulated industries** (finance, healthcare, insurance): flag compliance requirements\n- Workflows involving PII must include data retention and access control considerations\n- Recommend starting with a **pilot workflow** (lowest risk, highest frequency) before scaling\n- Provide rollback strategies: every agentic workflow should have a manual fallback\n\n*GitHub: https://github.com/gechengling/agentic-workflow-designer*\n\nFile v3.3.4:_meta.json\n\n{\n \"ownerId\": \"kn74e704j3ygjcygnpf02rdvd185js13\",\n \"slug\": \"agentic-workflow-designer\",\n \"version\": \"3.3.4\",\n \"publishedAt\": 1788187262658\n}\n\nFile v3.3.4:skill-card.md\n\n## Description:\n\nAgentic Workflow Designer helps users map business processes, assess automation suitability, design agentic workflow blueprints, generate n8n/Make/Zapier specs, and estimate ROI.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[gechengling](https://clawhub.ai/user/gechengling)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nOperations managers, developers, product managers, consultants, and entrepreneurs use this skill to assess business processes for agentic automation, design human-reviewed workflow pipelines, choose automation platforms, generate workflow specs, and estimate ROI.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: Generated workflow specifications may be imported into systems that handle PII, regulated data, customer communications, financial decisions, external APIs, or update/delete actions.\n\nMitigation: Review every generated specification before import, add human-in-the-loop checkpoints for high-stakes decisions, and verify retention, access control, audit, and manual fallback requirements.\n\nRisk: Workflow recommendations and ROI estimates are advisory and may not match the user's actual operating constraints.\n\nMitigation: Validate assumptions with real workflow data and start with a low-risk pilot before broader production rollout.\n\n## Reference(s):\n\n- [ClawHub skill page](https://clawhub.ai/gechengling/skills/agentic-workflow-designer)\n\n## Skill Output:\n\n**Output Type(s):** [text, markdown, code, configuration, guidance]\n\n**Output Format:** [Markdown with workflow blueprints, scoring tables, ROI estimates, and JSON/YAML workflow specifications]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [Generated workflow specifications are advisory and require human review before import or production deployment.]\n\n## Skill Version(s):\n\n3.3.4 (source: frontmatter and server release evidence)\n\n## Ethical Considerations:\n\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment.\n\nArchive v3.3.3: 3 files, 6796 bytes\n\nFiles: skill-card.md (2304b), SKILL.md (10601b), _meta.json (144b)\n\nFile v3.3.3:SKILL.md\n\n---\nname: Agentic Workflow Designer\ndescription: >\n AI-powered agentic workflow design and automation assistant — map complex multi-step\n processes, identify automation opportunities, design autonomous AI agent pipelines,\n generate n8n/Make/Zapier workflow specs, and estimate ROI. Covers enterprise automation,\n self-healing workflows, human-in-the-loop patterns, and production deployment. Keywords:\n agentic workflow, workflow automation, n8n, Make, Zapier, enterprise automation,\n AI pipeline, autonomous agent, process automation, workflow design, ROI calculator,\n HITL, 工作流设计, 流程自动化, 智能体工作流, 企业自动化, n8n工作流, 流程优化,\n 自主代理, RPA替代.\nversion: \"3.3.3\"\n---\n\n# Agentic Workflow Designer\n\n> From messy manual processes to autonomous AI pipelines — design, document, and deploy.\n\n## What This Skill Does\n\nAgentic AI (AI that can autonomously execute multi-step tasks) is the #1 enterprise tech trend in 2026 with a projected $8.5B market and 40% CAGR. Yet most teams struggle to:\n- Map which workflows are actually suitable for agentic automation\n- Design reliable pipelines that don't break silently\n- Choose between n8n, Make, Zapier, or custom agent frameworks\n- Justify the ROI to business stakeholders\n\nThis skill bridges the gap between AI hype and practical workflow automation:\n\n- **Workflow Discovery** — Identify and prioritize automation opportunities in any business process\n- **Agentic Pipeline Design** — Create detailed workflow blueprints with triggers, agents, tools, and fallbacks\n- **Platform Selection** — Compare n8n / Make / Zapier / custom LangGraph for your use case\n- **Generate Workflow Specs** — Produce JSON/YAML specs importable into n8n or Make\n- **ROI Calculator** — Estimate time/cost savings from automation\n- **Human-in-the-Loop (HITL) Design** — Design appropriate checkpoints for sensitive decisions\n\n## Trigger Words\n\nAgentic workflow, automate my process, workflow automation, n8n, Make automation, Zapier flow, design a workflow, workflow design, process automation, automate with AI, AI pipeline, autonomous workflow, HITL pattern, 工作流设计, 自动化工作流, 流程自动化, 智能体工作流, 帮我设计流程, 自动化这个流程, n8n工作流, 企业自动化, RPA替代, agentic AI pipeline\n\n## Target Users\n\n- Operations managers digitizing manual business processes\n- Developers building production AI automation systems\n- Product managers scoping automation features\n- Consultants delivering workflow automation projects\n- Entrepreneurs building AI-native products\n\n## Workflow\n\n### 新增内容（2026版）\n**Step 2 新增技术评估（2026）**：\n- LangGraph v1.0生产就绪：状态机工作流/长期记忆/错误恢复三大核心能力，企业级部署支持Kubernetes自动扩缩容，GitHub Stars突破85K\n- CrewAI v1.10多智能体协作：支持6种角色类型+并行任务编排，内置20+企业级连接器（Slack/Notion/Airtable/GitHub），2026年Q1新增中文文档\n- Claude Agent SDK / OpenAI Agents SDK横向对比：工具调用准确率(94% vs 91%)/上下文利用率(78% vs 82%)/成本效率(￥0.8/千Token vs ￥1.2/千Token)三大维度全面评测\n- MCP(Model Context Protocol)生态爆发：50+官方服务器覆盖GitHub/Slack/Notion/Postgres等，企业内部MCP注册表成为新基础设施\n- LLM长上下文之战：Gemini 2M Token / Claude 200K / GPT-4o 128K技术选型指南，针对金融长文档(招股书/年报)场景给出最优性价比方案\n\n---\n\n## Step 1 — Process Discovery\nAsk the user to describe their current workflow:\n- What triggers it? (email, schedule, webhook, human action?)\n- What are the key steps? (list them in plain language)\n- Who (or what system) does each step today?\n- Where do errors/delays typically occur?\n- What's the desired output/outcome?\n\n### Step 2 — Automation Suitability Assessment\n\nScore the workflow across 5 dimensions:\n\n| Dimension | Score | Notes |\n|-----------|-------|-------|\n| Repetitiveness | /10 | How often does this run identically? |\n| Rule-based | /10 | Are decisions clear-cut or judgment-based? |\n| Data availability | /10 | Is input data structured and accessible? |\n| Error tolerance | /10 | Can errors be caught and recovered automatically? |\n| Stakes | /10 (inverted) | Low-stakes = easier to automate |\n| **Automation Score** | /50 | >35 = High priority, 20–35 = Medium, <20 = Keep manual |\n\n### Step 3 — Agentic Pipeline Design\nGenerate a detailed pipeline blueprint:\n\n```\n[Workflow]: [Name]\n[Trigger]: [webhook / cron / event / manual]\n[Agents]:\n ├── Agent 1 [Role]: [Tool 1, Tool 2] → Output: [description]\n ├── Agent 2 [Role]: [Tool 3] → Output: [description]\n └── Agent 3 [Role]: [Tool 4, Tool 5] → Output: [description]\n[Flow]: Sequential / Parallel / Conditional\n[Memory]: [ephemeral / Redis / vector DB]\n[Error Handling]: [retry / fallback agent / human escalation]\n[HITL Checkpoints]: [list high-stakes decision points]\n[Output]: [final deliverable description]\n```\n\n**Example — Lead Qualification Pipeline:**\n```\n[Workflow]: B2B Lead Qualification & Outreach\n[Trigger]: New form submission webhook\n[Agents]:\n ├── Enrichment Agent [Clearbit + LinkedIn scraper] → Company profile JSON\n ├── Scoring Agent [GPT-4o] → Lead score (0-100) + reasoning\n ├── Decision Gate [Human] → Approve for outreach? (HITL)\n └── Outreach Agent [Email API + CRM API] → Personalized email + CRM update\n[Flow]: Sequential with HITL gate\n[Memory]: PostgreSQL (lead history)\n[Error]: Retry enrichment 3x → flag for manual review\n[HITL]: Score > 80 auto-approves; 50-80 requires human review; <50 auto-rejects\n[Output]: CRM updated + email queued\n```\n\n### Step 4 — Platform Recommendation\n\n| Platform | Best For | Agent Support | Self-host | Price |\n|----------|----------|--------------|-----------|-------|\n| n8n | Technical teams, complex logic | [Yes] via AI nodes | [Yes] | Free/OSS |\n| Make (Integromat) | Non-technical, API integrations | Partial | [No] | ~$9+/mo |\n| Zapier | Simple triggers, non-technical | Partial | [No] | ~$20+/mo |\n| LangGraph (custom) | Complex state machines, production | [Yes] Native | [Yes] | Dev hours |\n| CrewAI | Role-based agent teams | [Yes] Native | [Yes] | Dev hours |\n\n\n### Step 4.5 — 2026平台详细对比表（生产选型参考）\n\n| 维度 | n8n (v1.90) | Make (2026) | Zapier (2026) | LangGraph | CrewAI |\n|------|--------------|-------------|---------------|-----------|--------|\n| **AI节点** | [Yes] 原生AI节点（OpenAI/Claude/本地LLM）| [!] 需通过HTTP节点调用 | [!] 需通过Code节点调用 | [Yes] 原生 | [Yes] 原生 |\n| **定价（月）** | 免费（OSS）/ $20/月（Cloud Pro）| $9/月（Core）~$16/月（Enterprise）| $20/月（Starter）~$69/月（Company）| Dev成本 | Dev成本 |\n| **自托管** | [Yes] Docker一键部署 | [No] 仅SaaS | [No] 仅SaaS | [Yes] | [Yes] |\n| **企业连接器** | 400+（含国内钉钉/企微）| 1000+（偏海外）| 6000+（全球最多）| 自接 | 自接 |\n| **适合场景** | 技术研发/复杂逻辑/数据敏感 | 非技术/跨部门/快速原型 | 销售/市场/简单自动化 | 复杂状态机/生产级 | 角色协作/研究分析 |\n| **最大短板** | 学习曲线陡峭 | 国内SaaS访问慢 | 国内SaaS访问慢+贵 | 需开发资源 | 需开发资源 |\n\n**选型建议（2026）**：\n- 国内团队/数据合规要求 → **n8n自托管**（数据不出境，支持国产LLM接入）\n- 海外业务/非技术团队 → **Make**（1000+连接器，学习成本低）\n- 简单场景/销售团队 → **Zapier**（即开即用，但长期成本高）\n- 复杂AI管线/生产部署 → **LangGraph**（状态持久化，支持Human-in-the-Loop）\n- 多角色协作/研究分析 → **CrewAI**（角色分工清晰，2026年中文文档完善）\n\n---\n### Step 5 — n8n Workflow JSON Spec (Sample Output)\n```json\n{\n \"name\": \"Lead Qualification Pipeline\",\n \"nodes\": [\n {\n \"name\": \"Webhook Trigger\",\n \"type\": \"n8n-nodes-base.webhook\",\n \"parameters\": { \"path\": \"lead-inbound\" }\n },\n {\n \"name\": \"Enrich Lead\",\n \"type\": \"@n8n/n8n-nodes-langchain.agent\",\n \"parameters\": {\n \"promptType\": \"define\",\n \"text\": \"Enrich this lead data using Clearbit: {{ $json.email }}\"\n }\n },\n {\n \"name\": \"Score Lead\",\n \"type\": \"@n8n/n8n-nodes-langchain.openAi\",\n \"parameters\": {\n \"resource\": \"text\",\n \"operation\": \"message\",\n \"modelId\": \"gpt-4o\",\n \"messages\": { \"values\": [{ \"content\": \"Score this lead 0-100...\" }] }\n }\n }\n ]\n}\n```\n\n### Step 6 — ROI Calculator\n\n| Metric | Before Automation | After Automation | Savings |\n|--------|------------------|-----------------|---------|\n| Time per run | [X hours] | [Y minutes] | [Z%] |\n| Runs per week | [N] | [N] | — |\n| Total time saved/week | — | — | [hours] |\n| Cost saved/month | — | — | [$$$] |\n| Automation setup cost | — | — | [one-time] |\n| **Payback period** | — | — | [weeks] |\n\n## Example Interactions\n\n**User:** \"I spend 3 hours every Monday pulling sales data from 5 spreadsheets, writing a summary email, and updating our CRM. Can this be automated?\"\n\n**Skill response:** Scores the workflow (42/50 — High priority), designs a 4-agent pipeline (data collector → analyzer → email writer → CRM updater), recommends n8n as the platform (self-hostable, native AI nodes), generates a complete n8n JSON spec, and estimates 11.5 hours/month saved = ~$580 value at $50/hr.\n\n---\n\n**User:** \"I want to build a customer support triage system that reads emails, classifies them, and routes to the right team.\"\n\n**Skill response:** Designs a HITL-enabled pipeline with email reading, classification, confidence threshold (>85% auto-route, <85% human review), CRM ticket creation, and Slack notification. Recommends LangGraph for its state persistence and human review interrupt capability.\n\n## Notes & Constraints\n\n- Always design **HITL checkpoints** for: financial decisions, customer communications, data deletions, external API calls with side effects\n- For **regulated industries** (finance, healthcare, insurance): flag compliance requirements\n- Workflows involving PII must include data retention and access control considerations\n- Recommend starting with a **pilot workflow** (lowest risk, highest frequency) before scaling\n- Provide rollback strategies: every agentic workflow should have a manual fallback\n\n*GitHub: https://github.com/gechengling/agentic-workflow-designer*\n\nFile v3.3.3:_meta.json\n\n{\n \"ownerId\": \"kn74e704j3ygjcygnpf02rdvd185js13\",\n \"slug\": \"agentic-workflow-designer\",\n \"version\": \"3.3.3\",\n \"publishedAt\": 1781577411291\n}\n\nFile v3.3.3:skill-card.md\n\n## Description:
\nAI-powered workflow design and automation guidance for mapping business processes, assessing automation fit, designing agentic pipelines, generating workflow specifications, and estimating ROI.
\n\nThis skill is ready for commercial/non-commercial use.
\n\n## Publisher:
\n[gechengling](https://clawhub.ai/user/gechengling)
\n\n### License/Terms of Use:
\nMIT-0
\n\n\n## Use Case:
\nOperations managers, developers, product managers, consultants, and founders use this skill to assess business processes for automation and design practical agentic workflows. It helps produce pipeline blueprints, platform recommendations, HITL checkpoints, workflow specs, and ROI estimates.
\n\n### Deployment Geography for Use:
\nGlobal
\n\n## Known Risks and Mitigations:
\nRisk: Broad workflow-automation triggers may route general automation planning requests to this skill.
\nMitigation: Enable it only when workflow-automation planning support is desired and review its recommendations before use.
\nRisk: Generated workflow designs may affect sensitive business data, customer records, payments, public posting, or external systems.
\nMitigation: Require human review, HITL checkpoints, access controls, and rollback plans before implementing or connecting generated workflows to third-party tools.
\n\n\n## Reference(s):
\n- [ClawHub Skill Page](https://clawhub.ai/gechengling/agentic-workflow-designer)
\n- [Skill Definition](artifact/SKILL.md)
\n\n\n## Skill Output:
\n**Output Type(s):** [text, markdown, code, configuration, guidance]
\n**Output Format:** [Markdown guidance with JSON/YAML workflow specifications, scoring tables, and configuration-oriented recommendations]
\n**Output Parameters:** [1D]
\n**Other Properties Related to Output:** [May include automation suitability scores, HITL checkpoints, platform comparisons, rollback guidance, and ROI estimates.]
\n\n## Skill Version(s):
\n3.3.3 (source: frontmatter and server evidence)
\n\n## Ethical Considerations:
\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment.
\n\nArchive v3.3.2: 3 files, 6492 bytes\n\nFiles: skill-card.md (2004b), SKILL.md (9952b), _meta.json (144b)\n\nFile v3.3.2:SKILL.md\n\n---\nname: Agentic Workflow Designer\ndescription: >\n AI-powered agentic workflow design and automation assistant �� map complex multi-step\n processes, identify automation opportunities, design autonomous AI agent pipelines,\n generate n8n/Make/Zapier workflow specs, and estimate ROI. Covers enterprise automation,\n self-healing workflows, human-in-the-loop patterns, and production deployment. Keywords:\n agentic workflow, workflow automation, n8n, Make, Zapier, enterprise automation,\n AI pipeline, autonomous agent, process automation, workflow design, ROI calculator,\n HITL, ���������, �����Զ���, �����幤����, ��ҵ�Զ���, n8n������, �����Ż�,\n ��������, RPA���.\nversion: \"3.3.2\"\n---\n\n# Agentic Workflow Designer\n\n> From messy manual processes to autonomous AI pipelines �� design, document, and deploy.\n\n## What This Skill Does\n\nAgentic AI (AI that can autonomously execute multi-step tasks) is the #1 enterprise tech trend in 2026 with a projected $8.5B market and 40% CAGR. Yet most teams struggle to:\n- Map which workflows are actually suitable for agentic automation\n- Design reliable pipelines that don't break silently\n- Choose between n8n, Make, Zapier, or custom agent frameworks\n- Justify the ROI to business stakeholders\n\nThis skill bridges the gap between AI hype and practical workflow automation:\n\n- **Workflow Discovery** �� Identify and prioritize automation opportunities in any business process\n- **Agentic Pipeline Design** �� Create detailed workflow blueprints with triggers, agents, tools, and fallbacks\n- **Platform Selection** �� Compare n8n / Make / Zapier / custom LangGraph for your use case\n- **Generate Workflow Specs** �� Produce JSON/YAML specs importable into n8n or Make\n- **ROI Calculator** �� Estimate time/cost savings from automation\n- **Human-in-the-Loop (HITL) Design** �� Design appropriate checkpoints for sensitive decisions\n\n## Trigger Words\n\nAgentic workflow, automate my process, workflow automation, n8n, Make automation, Zapier flow, design a workflow, workflow design, process automation, automate with AI, AI pipeline, autonomous workflow, HITL pattern, ���������, �Զ���������, �����Զ���, �����幤����, �����������, �Զ����������, n8n������, ��ҵ�Զ���, RPA���, agentic AI pipeline\n\n## Target Users\n\n- Operations managers digitizing manual business processes\n- Developers building production AI automation systems\n- Product managers scoping automation features\n- Consultants delivering workflow automation projects\n- Entrepreneurs building AI-native products\n\n## Workflow\n\n### �������ݣ�2026�棩\n**Step 2 ��������������2026��**��\n- LangGraph v1.0����������״̬��������/���ڼ���/����ָ����������������ҵ������֧��Kubernetes�Զ������ݣ�GitHub Starsͻ��85K\n- CrewAI v1.10��������Э����֧��6�ֽ�ɫ����+����������ţ�����20+��ҵ����������Slack/Notion/Airtable/GitHub����2026��Q1���������ĵ�\n- Claude Agent SDK / OpenAI Agents SDK����Աȣ����ߵ���ȷ��(94% vs 91%)/������������(78% vs 82%)/�ɱ�Ч��(��0.8/ǧToken vs ��1.2/ǧToken)����ά��ȫ������\n- MCP(Model Context Protocol)��̬������50+�ٷ�����������GitHub/Slack/Notion/Postgres�ȣ���ҵ�ڲ�MCPע�����Ϊ�»�����ʩ\n- LLM��������֮ս��Gemini 2M Token / Claude 200K / GPT-4o 128K����ѡ��ָ�ϣ���Խ��ڳ��ĵ�(�й���/�걨)�������������Լ۱ȷ���\n\n---\n\n## Step 1 �� Process Discovery\nAsk the user to describe their current workflow:\n- What triggers it? (email, schedule, webhook, human action?)\n- What are the key steps? (list them in plain language)\n- Who (or what system) does each step today?\n- Where do errors/delays typically occur?\n- What's the desired output/outcome?\n\n### Step 2 �� Automation Suitability Assessment\n\nScore the workflow across 5 dimensions:\n\n| Dimension | Score | Notes |\n|-----------|-------|-------|\n| Repetitiveness | /10 | How often does this run identically? |\n| Rule-based | /10 | Are decisions clear-cut or judgment-based? |\n| Data availability | /10 | Is input data structured and accessible? |\n| Error tolerance | /10 | Can errors be caught and recovered automatically? |\n| Stakes | /10 (inverted) | Low-stakes = easier to automate |\n| **Automation Score** | /50 | >35 = High priority, 20�C35 = Medium, <20 = Keep manual |\n\n### Step 3 �� Agentic Pipeline Design\nGenerate a detailed pipeline blueprint:\n\n```\n[Workflow]: [Name]\n[Trigger]: [webhook / cron / event / manual]\n[Agents]:\n ������ Agent 1 [Role]: [Tool 1, Tool 2] �� Output: [description]\n ������ Agent 2 [Role]: [Tool 3] �� Output: [description]\n ������ Agent 3 [Role]: [Tool 4, Tool 5] �� Output: [description]\n[Flow]: Sequential / Parallel / Conditional\n[Memory]: [ephemeral / Redis / vector DB]\n[Error Handling]: [retry / fallback agent / human escalation]\n[HITL Checkpoints]: [list high-stakes decision points]\n[Output]: [final deliverable description]\n```\n\n**Example �� Lead Qualification Pipeline:**\n```\n[Workflow]: B2B Lead Qualification & Outreach\n[Trigger]: New form submission webhook\n[Agents]:\n ������ Enrichment Agent [Clearbit + LinkedIn scraper] �� Company profile JSON\n ������ Scoring Agent [GPT-4o] �� Lead score (0-100) + reasoning\n ������ Decision Gate [Human] �� Approve for outreach? (HITL)\n ������ Outreach Agent [Email API + CRM API] �� Personalized email + CRM update\n[Flow]: Sequential with HITL gate\n[Memory]: PostgreSQL (lead history)\n[Error]: Retry enrichment 3x �� flag for manual review\n[HITL]: Score > 80 auto-approves; 50-80 requires human review; <50 auto-rejects\n[Output]: CRM updated + email queued\n```\n\n### Step 4 �� Platform Recommendation\n\n| Platform | Best For | Agent Support | Self-host | Price |\n|----------|----------|--------------|-----------|-------|\n| n8n | Technical teams, complex logic | [Yes] via AI nodes | [Yes] | Free/OSS |\n| Make (Integromat) | Non-technical, API integrations | Partial | [No] | ~$9+/mo |\n| Zapier | Simple triggers, non-technical | Partial | [No] | ~$20+/mo |\n| LangGraph (custom) | Complex state machines, production | [Yes] Native | [Yes] | Dev hours |\n| CrewAI | Role-based agent teams | [Yes] Native | [Yes] | Dev hours |\n\n\n### Step 4.5 �� 2026ƽ̨��ϸ�Աȱ�������ѡ�Ͳο���\n\n| ά�� | n8n (v1.90) | Make (2026) | Zapier (2026) | LangGraph | CrewAI |\n|------|--------------|-------------|---------------|-----------|--------|\n| **AI�ڵ�** | [Yes] ԭ��AI�ڵ㣨OpenAI/Claude/����LLM��| [!] ��ͨ��HTTP�ڵ���� | [!] ��ͨ��Code�ڵ���� | [Yes] ԭ�� | [Yes] ԭ�� |\n| **���ۣ��£�** | ��ѣ�OSS��/ $20/�£�Cloud Pro��| $9/�£�Core��~$16/�£�Enterprise��| $20/�£�Starter��~$69/�£�Company��| Dev�ɱ� | Dev�ɱ� |\n| **���й�** | [Yes] Dockerһ������ | [No] ��SaaS | [No] ��SaaS | [Yes] | [Yes] |\n| **��ҵ������** | 400+�������ڶ���/����| 1000+��ƫ���⣩| 6000+��ȫ����ࣩ| �Խ� | �Խ� |\n| **�ʺϳ���** | �����з�/������/�������� | �Ǽ���/�粿��/����ԭ�� | ����/�г�/���Զ��� | ����״̬��/������ | ��ɫЭ��/�о����� |\n| **���̰�** | ѧϰ���߶��� | ����SaaS������ | ����SaaS������+�� | �迪����Դ | �迪����Դ |\n\n**ѡ�ͽ��飨2026��**��\n- �����Ŷ�/���ݺϹ�Ҫ�� �� **n8n���й�**�����ݲ�������֧�ֹ���LLM���룩\n- ����ҵ��/�Ǽ����Ŷ� �� **Make**��1000+��������ѧϰ�ɱ��ͣ�\n- ����/�����Ŷ� �� **Zapier**���������ã������ڳɱ��ߣ�\n- ����AI����/�������� �� **LangGraph**��״̬�־û���֧��Human-in-the-Loop��\n- ���ɫЭ��/�о����� �� **CrewAI**����ɫ�ֹ�������2026�������ĵ����ƣ�\n\n---\n### Step 5 �� n8n Workflow JSON Spec (Sample Output)\n```json\n{\n \"name\": \"Lead Qualification Pipeline\",\n \"nodes\": [\n {\n \"name\": \"Webhook Trigger\",\n \"type\": \"n8n-nodes-base.webhook\",\n \"parameters\": { \"path\": \"lead-inbound\" }\n },\n {\n \"name\": \"Enrich Lead\",\n \"type\": \"@n8n/n8n-nodes-langchain.agent\",\n \"parameters\": {\n \"promptType\": \"define\",\n \"text\": \"Enrich this lead data using Clearbit: {{ $json.email }}\"\n }\n },\n {\n \"name\": \"Score Lead\",\n \"type\": \"@n8n/n8n-nodes-langchain.openAi\",\n \"parameters\": {\n \"resource\": \"text\",\n \"operation\": \"message\",\n \"modelId\": \"gpt-4o\",\n \"messages\": { \"values\": [{ \"content\": \"Score this lead 0-100...\" }] }\n }\n }\n ]\n}\n```\n\n### Step 6 �� ROI Calculator\n\n| Metric | Before Automation | After Automation | Savings |\n|--------|------------------|-----------------|---------|\n| Time per run | [X hours] | [Y minutes] | [Z%] |\n| Runs per week | [N] | [N] | �� |\n| Total time saved/week | �� | �� | [hours] |\n| Cost saved/month | �� | �� | [$$$] |\n| Automation setup cost | �� | �� | [one-time] |\n| **Payback period** | �� | �� | [weeks] |\n\n## Example Interactions\n\n**User:** \"I spend 3 hours every Monday pulling sales data from 5 spreadsheets, writing a summary email, and updating our CRM. Can this be automated?\"\n\n**Skill response:** Scores the workflow (42/50 �� High priority), designs a 4-agent pipeline (data collector �� analyzer �� email writer �� CRM updater), recommends n8n as the platform (self-hostable, native AI nodes), generates a complete n8n JSON spec, and estimates 11.5 hours/month saved = ~$580 value at $50/hr.\n\n---\n\n**User:** \"I want to build a customer support triage system that reads emails, classifies them, and routes to the right team.\"\n\n**Skill response:** Designs a HITL-enabled pipeline with email reading, classification, confidence threshold (>85% auto-route, <85% human review), CRM ticket creation, and Slack notification. Recommends LangGraph for its state persistence and human review interrupt capability.\n\n## Notes & Constraints\n\n- Always design **HITL checkpoints** for: financial decisions, customer communications, data deletions, external API calls with side effects\n- For **regulated industries** (finance, healthcare, insurance): flag compliance requirements\n- Workflows involving PII must include data retention and access control considerations\n- Recommend starting with a **pilot workflow** (lowest risk, highest frequency) before scaling\n- Provide rollback strategies: every agentic workflow should have a manual fallback\n\n*GitHub: https://github.com/gechengling/agentic-workflow-designer*\n\nFile v3.3.2:_meta.json\n\n{\n \"ownerId\": \"kn74e704j3ygjcygnpf02rdvd185js13\",\n \"slug\": \"agentic-workflow-designer\",\n \"version\": \"3.3.2\",\n \"publishedAt\": 1781574373180\n}\n\nFile v3.3.2:skill-card.md\n\n## Description:
\nDesigns agentic workflow automations by mapping business processes, scoring automation fit, recommending platforms, drafting agent pipelines, producing workflow specs, and estimating ROI.
\n\nThis skill is ready for commercial/non-commercial use.
\n\n## Publisher:
\n[gechengling](https://clawhub.ai/user/gechengling)
\n\n### License/Terms of Use:
\nMIT-0
\n\n\n## Use Case:
\nOperations managers, developers, product managers, consultants, and entrepreneurs use this skill to turn manual business processes into automation blueprints with platform recommendations, human-in-the-loop checkpoints, workflow specs, and ROI estimates.
\n\n### Deployment Geography for Use:
\nGlobal
\n\n## Known Risks and Mitigations:
\nRisk: Generated automation specs may touch PII, customer messages, financial decisions, deletions, credentials, or production APIs.
\nMitigation: Treat generated specs as drafts and review them before importing, connecting, or running workflows against real systems.
\n\n\n## Reference(s):
\n- [Agentic Workflow Designer on ClawHub](https://clawhub.ai/gechengling/agentic-workflow-designer)
\n\n\n## Skill Output:
\n**Output Type(s):** [text, markdown, code, configuration, guidance]
\n**Output Format:** [Markdown with JSON or YAML workflow specifications when requested]
\n**Output Parameters:** [1D]
\n**Other Properties Related to Output:** [Outputs may include process assessments, automation scores, platform recommendations, agent pipeline blueprints, n8n/Make/Zapier-style specs, HITL checkpoints, rollback guidance, and ROI estimates.]
\n\n## Skill Version(s):
\n3.3.2 (source: server release evidence and skill frontmatter)
\n\n## Ethical Considerations:
\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment.
\n\nArchive v3.3.1: 3 files, 6622 bytes\n\nFiles: skill-card.md (2340b), SKILL.md (9917b), _meta.json (144b)\n\nFile v3.3.1:SKILL.md\n\n---\nname: Agentic Workflow Designer\ndescription: >\n AI-powered agentic workflow design and automation assistant �� map complex multi-step\n processes, identify automation opportunities, design autonomous AI agent pipelines,\n generate n8n/Make/Zapier workflow specs, and estimate ROI. Covers enterprise automation,\n self-healing workflows, human-in-the-loop patterns, and production deployment. Keywords:\n agentic workflow, workflow automation, n8n, Make, Zapier, enterprise automation,\n AI pipeline, autonomous agent, process automation, workflow design, ROI calculator,\n HITL, ���������, �����Զ���, �����幤����, ��ҵ�Զ���, n8n������, �����Ż�,\n ��������, RPA���.\nversion: \"3.3.1\"\n---\n\n# Agentic Workflow Designer\n\n> From messy manual processes to autonomous AI pipelines �� design, document, and deploy.\n\n## What This Skill Does\n\nAgentic AI (AI that can autonomously execute multi-step tasks) is the #1 enterprise tech trend in 2026 with a projected $8.5B market and 40% CAGR. Yet most teams struggle to:\n- Map which workflows are actually suitable for agentic automation\n- Design reliable pipelines that don't break silently\n- Choose between n8n, Make, Zapier, or custom agent frameworks\n- Justify the ROI to business stakeholders\n\nThis skill bridges the gap between AI hype and practical workflow automation:\n\n- **Workflow Discovery** �� Identify and prioritize automation opportunities in any business process\n- **Agentic Pipeline Design** �� Create detailed workflow blueprints with triggers, agents, tools, and fallbacks\n- **Platform Selection** �� Compare n8n / Make / Zapier / custom LangGraph for your use case\n- **Generate Workflow Specs** �� Produce JSON/YAML specs importable into n8n or Make\n- **ROI Calculator** �� Estimate time/cost savings from automation\n- **Human-in-the-Loop (HITL) Design** �� Design appropriate checkpoints for sensitive decisions\n\n## Trigger Words\n\nAgentic workflow, automate my process, workflow automation, n8n, Make automation, Zapier flow, design a workflow, workflow design, process automation, automate with AI, AI pipeline, autonomous workflow, HITL pattern, ���������, �Զ���������, �����Զ���, �����幤����, �����������, �Զ����������, n8n������, ��ҵ�Զ���, RPA���, agentic AI pipeline\n\n## Target Users\n\n- Operations managers digitizing manual business processes\n- Developers building production AI automation systems\n- Product managers scoping automation features\n- Consultants delivering workflow automation projects\n- Entrepreneurs building AI-native products\n\n## Workflow\n\n### �������ݣ�2026�棩\n**Step 2 ��������������2026��**��\n- LangGraph v1.0����������״̬��������/���ڼ���/����ָ����������������ҵ������֧��Kubernetes�Զ������ݣ�GitHub Starsͻ��85K\n- CrewAI v1.10��������Э����֧��6�ֽ�ɫ����+����������ţ�����20+��ҵ����������Slack/Notion/Airtable/GitHub����2026��Q1���������ĵ�\n- Claude Agent SDK / OpenAI Agents SDK����Աȣ����ߵ���ȷ��(94% vs 91%)/������������(78% vs 82%)/�ɱ�Ч��(��0.8/ǧToken vs ��1.2/ǧToken)����ά��ȫ������\n- MCP(Model Context Protocol)��̬������50+�ٷ�����������GitHub/Slack/Notion/Postgres�ȣ���ҵ�ڲ�MCPע�����Ϊ�»�����ʩ\n- LLM��������֮ս��Gemini 2M Token / Claude 200K / GPT-4o 128K����ѡ��ָ�ϣ���Խ��ڳ��ĵ�(�й���/�걨)�������������Լ۱ȷ���\n\n---\n\n## Step 1 �� Process Discovery\nAsk the user to describe their current workflow:\n- What triggers it? (email, schedule, webhook, human action?)\n- What are the key steps? (list them in plain language)\n- Who (or what system) does each step today?\n- Where do errors/delays typically occur?\n- What's the desired output/outcome?\n\n### Step 2 �� Automation Suitability Assessment\n\nScore the workflow across 5 dimensions:\n\n| Dimension | Score | Notes |\n|-----------|-------|-------|\n| Repetitiveness | /10 | How often does this run identically? |\n| Rule-based | /10 | Are decisions clear-cut or judgment-based? |\n| Data availability | /10 | Is input data structured and accessible? |\n| Error tolerance | /10 | Can errors be caught and recovered automatically? |\n| Stakes | /10 (inverted) | Low-stakes = easier to automate |\n| **Automation Score** | /50 | >35 = High priority, 20�C35 = Medium, <20 = Keep manual |\n\n### Step 3 �� Agentic Pipeline Design\nGenerate a detailed pipeline blueprint:\n\n```\n?? Workflow: [Name]\n? Trigger: [webhook / cron / event / manual]\n?? Agents:\n ������ Agent 1 [Role]: [Tool 1, Tool 2] �� Output: [description]\n ������ Agent 2 [Role]: [Tool 3] �� Output: [description]\n ������ Agent 3 [Role]: [Tool 4, Tool 5] �� Output: [description]\n?? Flow: Sequential / Parallel / Conditional\n?? Memory: [ephemeral / Redis / vector DB]\n?? Error Handling: [retry / fallback agent / human escalation]\n?? HITL Checkpoints: [list high-stakes decision points]\n?? Output: [final deliverable description]\n```\n\n**Example �� Lead Qualification Pipeline:**\n```\n?? Workflow: B2B Lead Qualification & Outreach\n? Trigger: New form submission webhook\n?? Agents:\n ������ Enrichment Agent [Clearbit + LinkedIn scraper] �� Company profile JSON\n ������ Scoring Agent [GPT-4o] �� Lead score (0�C100) + reasoning\n ������ Decision Gate [Human] �� Approve for outreach? (HITL)\n ������ Outreach Agent [Email API + CRM API] �� Personalized email + CRM update\n?? Flow: Sequential with HITL gate\n?? Memory: PostgreSQL (lead history)\n?? Error: Retry enrichment 3x �� flag for manual review\n?? HITL: Score > 80 auto-approves; 50�C80 requires human review; <50 auto-rejects\n?? Output: CRM updated + email queued\n```\n\n### Step 4 �� Platform Recommendation\n\n| Platform | Best For | Agent Support | Self-host | Price |\n|----------|----------|--------------|-----------|-------|\n| n8n | Technical teams, complex logic | ? via AI nodes | ? Yes | Free/OSS |\n| Make (Integromat) | Non-technical, API integrations | Partial | ? No | ~$9+/mo |\n| Zapier | Simple triggers, non-technical | Partial | ? No | ~$20+/mo |\n| LangGraph (custom) | Complex state machines, production | ? Native | ? Yes | Dev hours |\n| CrewAI | Role-based agent teams | ? Native | ? Yes | Dev hours |\n\n\n### Step 4.5 �� 2026ƽ̨��ϸ�Աȱ�������ѡ�Ͳο���\n\n| ά�� | n8n (v1.90) | Make (2026) | Zapier (2026) | LangGraph | CrewAI |\n|------|--------------|-------------|---------------|-----------|--------|\n| **AI�ڵ�** | ? ԭ��AI�ڵ㣨OpenAI/Claude/����LLM�� | ?? ��ͨ��HTTP�ڵ���� | ?? ��ͨ��Code�ڵ���� | ? ԭ�� | ? ԭ�� |\n| **���ۣ��£�** | ��ѣ�OSS��/ /�£�Cloud Pro�� | /�£�Core��~/�£�Enterprise�� | /�£�Starter��~/�£�Company�� | Dev�ɱ� | Dev�ɱ� |\n| **���й�** | ? Dockerһ������ | ? ��SaaS | ? ��SaaS | ? | ? |\n| **��ҵ������** | 400+�������ڶ���/���� | 1000+��ƫ���⣩ | 6000+��ȫ����ࣩ | �Խ� | �Խ� |\n| **�ʺϳ���** | �����з�/������/�������� | �Ǽ���/�粿��/����ԭ�� | ����/�г�/���Զ��� | ����״̬��/������ | ��ɫЭ��/�о����� |\n| **���̰�** | ѧϰ���߶��� | ����SaaS������ | ����SaaS������+�� | �迪����Դ | �迪����Դ |\n\n**ѡ�ͽ��飨2026��**��\n- �����Ŷ�/���ݺϹ�Ҫ�� �� **n8n���й�**�����ݲ�������֧�ֹ���LLM���룩\n- ����ҵ��/�Ǽ����Ŷ� �� **Make**��1000+��������ѧϰ�ɱ��ͣ�\n- ����/�����Ŷ� �� **Zapier**���������ã������ڳɱ��ߣ�\n- ����AI����/�������� �� **LangGraph**��״̬�־û���֧��Human-in-the-Loop��\n- ���ɫЭ��/�о����� �� **CrewAI**����ɫ�ֹ�������2026�������ĵ����ƣ�\n\n---\n### Step 5 �� n8n Workflow JSON Spec (Sample Output)\n```json\n{\n \"name\": \"Lead Qualification Pipeline\",\n \"nodes\": [\n {\n \"name\": \"Webhook Trigger\",\n \"type\": \"n8n-nodes-base.webhook\",\n \"parameters\": { \"path\": \"lead-inbound\" }\n },\n {\n \"name\": \"Enrich Lead\",\n \"type\": \"@n8n/n8n-nodes-langchain.agent\",\n \"parameters\": {\n \"promptType\": \"define\",\n \"text\": \"Enrich this lead data using Clearbit: {{ $json.email }}\"\n }\n },\n {\n \"name\": \"Score Lead\",\n \"type\": \"@n8n/n8n-nodes-langchain.openAi\",\n \"parameters\": {\n \"resource\": \"text\",\n \"operation\": \"message\",\n \"modelId\": \"gpt-4o\",\n \"messages\": { \"values\": [{ \"content\": \"Score this lead 0-100...\" }] }\n }\n }\n ]\n}\n```\n\n### Step 6 �� ROI Calculator\n\n| Metric | Before Automation | After Automation | Savings |\n|--------|------------------|-----------------|---------|\n| Time per run | [X hours] | [Y minutes] | [Z%] |\n| Runs per week | [N] | [N] | �� |\n| Total time saved/week | �� | �� | [hours] |\n| Cost saved/month | �� | �� | [$$$] |\n| Automation setup cost | �� | �� | [one-time] |\n| **Payback period** | �� | �� | [weeks] |\n\n## Example Interactions\n\n**User:** \"I spend 3 hours every Monday pulling sales data from 5 spreadsheets, writing a summary email, and updating our CRM. Can this be automated?\"\n\n**Skill response:** Scores the workflow (42/50 �� High priority), designs a 4-agent pipeline (data collector �� analyzer �� email writer �� CRM updater), recommends n8n as the platform (self-hostable, native AI nodes), generates a complete n8n JSON spec, and estimates 11.5 hours/month saved = ~$580 value at $50/hr.\n\n---\n\n**User:** \"I want to build a customer support triage system that reads emails, classifies them, and routes to the right team.\"\n\n**Skill response:** Designs a HITL-enabled pipeline with email reading, classification, confidence threshold (>85% auto-route, <85% human review), CRM ticket creation, and Slack notification. Recommends LangGraph for its state persistence and human review interrupt capability.\n\n## Notes & Constraints\n\n- Always design **HITL checkpoints** for: financial decisions, customer communications, data deletions, external API calls with side effects\n- For **regulated industries** (finance, healthcare, insurance): flag compliance requirements\n- Workflows involving PII must include data retention and access control considerations\n- Recommend starting with a **pilot workflow** (lowest risk, highest frequency) before scaling\n- Provide rollback strategies: every agentic workflow should have a manual fallback\n\n*GitHub: https://github.com/gechengling/agentic-workflow-designer*\n\nFile v3.3.1:_meta.json\n\n{\n \"ownerId\": \"kn74e704j3ygjcygnpf02rdvd185js13\",\n \"slug\": \"agentic-workflow-designer\",\n \"version\": \"3.3.1\",\n \"publishedAt\": 1779862073907\n}\n\nFile v3.3.1:skill-card.md\n\n## Description:
\nAI-powered agentic workflow design and automation assistant that maps complex multi-step processes, identifies automation opportunities, designs autonomous AI agent pipelines, generates n8n, Make, and Zapier workflow specs, and estimates ROI.
\n\nThis skill is ready for commercial/non-commercial use.
\n\n## Publisher:
\n[gechengling](https://clawhub.ai/user/gechengling)
\n\n### License/Terms of Use:
\nMIT-0
\n\n\n## Use Case:
\nOperations managers, developers, product managers, consultants, and entrepreneurs use this skill to assess business processes for automation, design agentic workflows with human-in-the-loop checkpoints, choose an automation platform, draft importable workflow specs, and estimate business value.
\n\n### Deployment Geography for Use:
\nGlobal
\n\n## Known Risks and Mitigations:
\nRisk: Generated workflows may perform real-world actions once imported into n8n, Make, Zapier, or similar tools.
\nMitigation: Review each generated workflow, test it in a sandbox, and use least-privilege credentials before production use.
\nRisk: Automation of customer communications, financial actions, data deletion, or external API calls can cause business or compliance harm if approved without oversight.
\nMitigation: Require human approval checkpoints for sensitive actions and keep manual rollback paths available.
\n\n\n## Reference(s):
\n- [ClawHub skill page](https://clawhub.ai/gechengling/agentic-workflow-designer)
\n\n\n## Skill Output:
\n**Output Type(s):** [Text, Markdown, Code, Configuration, Guidance]
\n**Output Format:** [Markdown with tables, workflow blueprints, JSON or YAML workflow specifications, platform recommendations, and ROI estimates]
\n**Output Parameters:** [1D]
\n**Other Properties Related to Output:** [Generated automation plans should be reviewed and tested before being imported into automation platforms or connected to live credentials.]
\n\n## Skill Version(s):
\n3.3.1 (source: frontmatter and server release metadata)
\n\n## Ethical Considerations:
\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment.
\n\nArchive vv1.0.0: 2 files, 4886 bytes\n\nFiles: SKILL.md (10122b), _meta.json (145b)\n\nFile vv1.0.0:SKILL.md\n\n---\r\nname: Agentic Workflow Designer\r\ndescription: >\r\n AI-powered agentic workflow design and automation assistant — map complex multi-step\r\n processes, identify automation opportunities, design autonomous AI agent pipelines,\r\n generate n8n/Make/Zapier workflow specs, and estimate ROI. Covers enterprise automation,\r\n self-healing workflows, human-in-the-loop patterns, and production deployment. Keywords:\r\n agentic workflow, workflow automation, n8n, Make, Zapier, enterprise automation,\r\n AI pipeline, autonomous agent, process automation, workflow design, ROI calculator,\r\n HITL, 工作流设计, 流程自动化, 智能体工作流, 企业自动化, n8n工作流, 流程优化,\r\n 自主代理, RPA替代.\r\nversion: \"3.2.0\"\r\n---\r\n\r\n# Agentic Workflow Designer\r\n\r\n> From messy manual processes to autonomous AI pipelines — design, document, and deploy.\r\n\r\n## What This Skill Does\r\n\r\nAgentic AI (AI that can autonomously execute multi-step tasks) is the #1 enterprise tech trend in 2026 with a projected $8.5B market and 40% CAGR. Yet most teams struggle to:\r\n- Map which workflows are actually suitable for agentic automation\r\n- Design reliable pipelines that don't break silently\r\n- Choose between n8n, Make, Zapier, or custom agent frameworks\r\n- Justify the ROI to business stakeholders\r\n\r\nThis skill bridges the gap between AI hype and practical workflow automation:\r\n\r\n- **Workflow Discovery** — Identify and prioritize automation opportunities in any business process\r\n- **Agentic Pipeline Design** — Create detailed workflow blueprints with triggers, agents, tools, and fallbacks\r\n- **Platform Selection** — Compare n8n / Make / Zapier / custom LangGraph for your use case\r\n- **Generate Workflow Specs** — Produce JSON/YAML specs importable into n8n or Make\r\n- **ROI Calculator** — Estimate time/cost savings from automation\r\n- **Human-in-the-Loop (HITL) Design** — Design appropriate checkpoints for sensitive decisions\r\n\r\n## Trigger Words\r\n\r\nAgentic workflow, automate my process, workflow automation, n8n, Make automation, Zapier flow, design a workflow, workflow design, process automation, automate with AI, AI pipeline, autonomous workflow, HITL pattern, 工作流设计, 自动化工作流, 流程自动化, 智能体工作流, 帮我设计流程, 自动化这个流程, n8n工作流, 企业自动化, RPA替代, agentic AI pipeline\r\n\r\n## Target Users\r\n\r\n- Operations managers digitizing manual business processes\r\n- Developers building production AI automation systems\r\n- Product managers scoping automation features\r\n- Consultants delivering workflow automation projects\r\n- Entrepreneurs building AI-native products\r\n\r\n## Workflow\r\n\r\n### 新增内容（2026版）\r\n**Step 2 新增技术评估（2026）**：\r\n- LangGraph v1.0生产就绪：状态机工作流/长期记忆/错误恢复三大核心能力，企业级部署支持Kubernetes自动扩缩容，GitHub Stars突破85K\r\n- CrewAI v1.10多智能体协作：支持6种角色类型+并行任务编排，内置20+企业级连接器（Slack/Notion/Airtable/GitHub），2026年Q1新增中文文档\r\n- Claude Agent SDK / OpenAI Agents SDK横向对比：工具调用准确率(94% vs 91%)/上下文利用率(78% vs 82%)/成本效率(¥0.8/千Token vs ¥1.2/千Token)三大维度全面评测\r\n- MCP(Model Context Protocol)生态爆发：50+官方服务器覆盖GitHub/Slack/Notion/Postgres等，企业内部MCP注册表成为新基础设施\r\n- LLM长上下文之战：Gemini 2M Token / Claude 200K / GPT-4o 128K技术选型指南，针对金融长文档(招股书/年报)场景给出最优性价比方案\r\n\r\n---\r\n\r\n## 新增内容（2026版）\r\n**Step 2 新增技术评估（2026）**：\r\n- LangGraph v1.0生产就绪：状态机工作流/长期记忆/错误恢复三大核心能力，企业级部署支持Kubernetes自动扩缩容，GitHub Stars突破85K\r\n- CrewAI v1.10多智能体协作：支持6种角色类型+并行任务编排，内置20+企业级连接器（Slack/Notion/Airtable/GitHub），2026年Q1新增中文文档\r\n- Claude Agent SDK / OpenAI Agents SDK横向对比：工具调用准确率(94% vs 91%)/上下文利用率(78% vs 82%)/成本效率(¥0.8/千Token vs ¥1.2/千Token)三大维度全面评测\r\n- MCP(Model Context Protocol)生态爆发：50+官方服务器覆盖GitHub/Slack/Notion/Postgres等，企业内部MCP注册表成为新基础设施\r\n- LLM长上下文之战：Gemini 2M Token / Claude 200K / GPT-4o 128K技术选型指南，针对金融长文档(招股书/年报)场景给出最优性价比方案\r\n\r\n---\r\n\r\n## Step 1 — Process Discovery\r\nAsk the user to describe their current workflow:\r\n- What triggers it? (email, schedule, webhook, human action?)\r\n- What are the key steps? (list them in plain language)\r\n- Who (or what system) does each step today?\r\n- Where do errors/delays typically occur?\r\n- What's the desired output/outcome?\r\n\r\n### Step 2 — Automation Suitability Assessment\r\n\r\nScore the workflow across 5 dimensions:\r\n\r\n| Dimension | Score | Notes |\r\n|-----------|-------|-------|\r\n| Repetitiveness | /10 | How often does this run identically? |\r\n| Rule-based | /10 | Are decisions clear-cut or judgment-based? |\r\n| Data availability | /10 | Is input data structured and accessible? |\r\n| Error tolerance | /10 | Can errors be caught and recovered automatically? |\r\n| Stakes | /10 (inverted) | Low-stakes = easier to automate |\r\n| **Automation Score** | /50 | >35 = High priority, 20–35 = Medium, <20 = Keep manual |\r\n\r\n### Step 3 — Agentic Pipeline Design\r\nGenerate a detailed pipeline blueprint:\r\n\r\n```\r\n🎯 Workflow: [Name]\r\n⚡ Trigger: [webhook / cron / event / manual]\r\n🤖 Agents:\r\n ├── Agent 1 [Role]: [Tool 1, Tool 2] → Output: [description]\r\n ├── Agent 2 [Role]: [Tool 3] → Output: [description]\r\n └── Agent 3 [Role]: [Tool 4, Tool 5] → Output: [description]\r\n🔄 Flow: Sequential / Parallel / Conditional\r\n🧠 Memory: [ephemeral / Redis / vector DB]\r\n🚨 Error Handling: [retry / fallback agent / human escalation]\r\n👤 HITL Checkpoints: [list high-stakes decision points]\r\n📊 Output: [final deliverable description]\r\n```\r\n\r\n**Example — Lead Qualification Pipeline:**\r\n```\r\n🎯 Workflow: B2B Lead Qualification & Outreach\r\n⚡ Trigger: New form submission webhook\r\n🤖 Agents:\r\n ├── Enrichment Agent [Clearbit + LinkedIn scraper] → Company profile JSON\r\n ├── Scoring Agent [GPT-4o] → Lead score (0–100) + reasoning\r\n ├── Decision Gate [Human] → Approve for outreach? (HITL)\r\n └── Outreach Agent [Email API + CRM API] → Personalized email + CRM update\r\n🔄 Flow: Sequential with HITL gate\r\n🧠 Memory: PostgreSQL (lead history)\r\n🚨 Error: Retry enrichment 3x → flag for manual review\r\n👤 HITL: Score > 80 auto-approves; 50–80 requires human review; <50 auto-rejects\r\n📊 Output: CRM updated + email queued\r\n```\r\n\r\n### Step 4 — Platform Recommendation\r\n\r\n| Platform | Best For | Agent Support | Self-host | Price |\r\n|----------|----------|--------------|-----------|-------|\r\n| n8n | Technical teams, complex logic | ✅ via AI nodes | ✅ Yes | Free/OSS |\r\n| Make (Integromat) | Non-technical, API integrations | Partial | ❌ No | ~$9+/mo |\r\n| Zapier | Simple triggers, non-technical | Partial | ❌ No | ~$20+/mo |\r\n| LangGraph (custom) | Complex state machines, production | ✅ Native | ✅ Yes | Dev hours |\r\n| CrewAI | Role-based agent teams | ✅ Native | ✅ Yes | Dev hours |\r\n\r\n### Step 5 — n8n Workflow JSON Spec (Sample Output)\r\n```json\r\n{\r\n \"name\": \"Lead Qualification Pipeline\",\r\n \"nodes\": [\r\n {\r\n \"name\": \"Webhook Trigger\",\r\n \"type\": \"n8n-nodes-base.webhook\",\r\n \"parameters\": { \"path\": \"lead-inbound\" }\r\n },\r\n {\r\n \"name\": \"Enrich Lead\",\r\n \"type\": \"@n8n/n8n-nodes-langchain.agent\",\r\n \"parameters\": {\r\n \"promptType\": \"define\",\r\n \"text\": \"Enrich this lead data using Clearbit: {{ $json.email }}\"\r\n }\r\n },\r\n {\r\n \"name\": \"Score Lead\",\r\n \"type\": \"@n8n/n8n-nodes-langchain.openAi\",\r\n \"parameters\": {\r\n \"resource\": \"text\",\r\n \"operation\": \"message\",\r\n \"modelId\": \"gpt-4o\",\r\n \"messages\": { \"values\": [{ \"content\": \"Score this lead 0-100...\" }] }\r\n }\r\n }\r\n ]\r\n}\r\n```\r\n\r\n### Step 6 — ROI Calculator\r\n\r\n| Metric | Before Automation | After Automation | Savings |\r\n|--------|------------------|-----------------|---------|\r\n| Time per run | [X hours] | [Y minutes] | [Z%] |\r\n| Runs per week | [N] | [N] | — |\r\n| Total time saved/week | — | — | [hours] |\r\n| Cost saved/month | — | — | [$$$] |\r\n| Automation setup cost | — | — | [one-time] |\r\n| **Payback period** | — | — | [weeks] |\r\n\r\n## Example Interactions\r\n\r\n**User:** \"I spend 3 hours every Monday pulling sales data from 5 spreadsheets, writing a summary email, and updating our CRM. Can this be automated?\"\r\n\r\n**Skill response:** Scores the workflow (42/50 — High priority), designs a 4-agent pipeline (data collector → analyzer → email writer → CRM updater), recommends n8n as the platform (self-hostable, native AI nodes), generates a complete n8n JSON spec, and estimates 11.5 hours/month saved = ~$580 value at $50/hr.\r\n\r\n---\r\n\r\n**User:** \"I want to build a customer support triage system that reads emails, classifies them, and routes to the right team.\"\r\n\r\n**Skill response:** Designs a HITL-enabled pipeline with email reading, classification, confidence threshold (>85% auto-route, <85% human review), CRM ticket creation, and Slack notification. Recommends LangGraph for its state persistence and human review interrupt capability.\r\n\r\n## Notes & Constraints\r\n\r\n- Always design **HITL checkpoints** for: financial decisions, customer communications, data deletions, external API calls with side effects\r\n- For **regulated industries** (finance, healthcare, insurance): flag compliance requirements\r\n- Workflows involving PII must include data retention and access control considerations\r\n- Recommend starting with a **pilot workflow** (lowest risk, highest frequency) before scaling\r\n- Provide rollback strategies: every agentic workflow should have a manual fallback\n\nFile vv1.0.0:_meta.json\n\n{\n \"ownerId\": \"kn74e704j3ygjcygnpf02rdvd185js13\",\n \"slug\": \"agentic-workflow-designer\",\n \"version\": \"v1.0.0\",\n \"publishedAt\": 1779091952052\n}\n\nArchive v3.3.0: 2 files, 4886 bytes\n\nFiles: SKILL.md (10122b), _meta.json (144b)\n\nFile v3.3.0:SKILL.md\n\n---\r\nname: Agentic Workflow Designer\r\ndescription: >\r\n AI-powered agentic workflow design and automation assistant — map complex multi-step\r\n processes, identify automation opportunities, design autonomous AI agent pipelines,\r\n generate n8n/Make/Zapier workflow specs, and estimate ROI. Covers enterprise automation,\r\n self-healing workflows, human-in-the-loop patterns, and production deployment. Keywords:\r\n agentic workflow, workflow automation, n8n, Make, Zapier, enterprise automation,\r\n AI pipeline, autonomous agent, process automation, workflow design, ROI calculator,\r\n HITL, 工作流设计, 流程自动化, 智能体工作流, 企业自动化, n8n工作流, 流程优化,\r\n 自主代理, RPA替代.\r\nversion: \"3.2.0\"\r\n---\r\n\r\n# Agentic Workflow Designer\r\n\r\n> From messy manual processes to autonomous AI pipelines — design, document, and deploy.\r\n\r\n## What This Skill Does\r\n\r\nAgentic AI (AI that can autonomously execute multi-step tasks) is the #1 enterprise tech trend in 2026 with a projected $8.5B market and 40% CAGR. Yet most teams struggle to:\r\n- Map which workflows are actually suitable for agentic automation\r\n- Design reliable pipelines that don't break silently\r\n- Choose between n8n, Make, Zapier, or custom agent frameworks\r\n- Justify the ROI to business stakeholders\r\n\r\nThis skill bridges the gap between AI hype and practical workflow automation:\r\n\r\n- **Workflow Discovery** — Identify and prioritize automation opportunities in any business process\r\n- **Agentic Pipeline Design** — Create detailed workflow blueprints with triggers, agents, tools, and fallbacks\r\n- **Platform Selection** — Compare n8n / Make / Zapier / custom LangGraph for your use case\r\n- **Generate Workflow Specs** — Produce JSON/YAML specs importable into n8n or Make\r\n- **ROI Calculator** — Estimate time/cost savings from automation\r\n- **Human-in-the-Loop (HITL) Design** — Design appropriate checkpoints for sensitive decisions\r\n\r\n## Trigger Words\r\n\r\nAgentic workflow, automate my process, workflow automation, n8n, Make automation, Zapier flow, design a workflow, workflow design, process automation, automate with AI, AI pipeline, autonomous workflow, HITL pattern, 工作流设计, 自动化工作流, 流程自动化, 智能体工作流, 帮我设计流程, 自动化这个流程, n8n工作流, 企业自动化, RPA替代, agentic AI pipeline\r\n\r\n## Target Users\r\n\r\n- Operations managers digitizing manual business processes\r\n- Developers building production AI automation systems\r\n- Product managers scoping automation features\r\n- Consultants delivering workflow automation projects\r\n- Entrepreneurs building AI-native products\r\n\r\n## Workflow\r\n\r\n### 新增内容（2026版）\r\n**Step 2 新增技术评估（2026）**：\r\n- LangGraph v1.0生产就绪：状态机工作流/长期记忆/错误恢复三大核心能力，企业级部署支持Kubernetes自动扩缩容，GitHub Stars突破85K\r\n- CrewAI v1.10多智能体协作：支持6种角色类型+并行任务编排，内置20+企业级连接器（Slack/Notion/Airtable/GitHub），2026年Q1新增中文文档\r\n- Claude Agent SDK / OpenAI Agents SDK横向对比：工具调用准确率(94% vs 91%)/上下文利用率(78% vs 82%)/成本效率(¥0.8/千Token vs ¥1.2/千Token)三大维度全面评测\r\n- MCP(Model Context Protocol)生态爆发：50+官方服务器覆盖GitHub/Slack/Notion/Postgres等，企业内部MCP注册表成为新基础设施\r\n- LLM长上下文之战：Gemini 2M Token / Claude 200K / GPT-4o 128K技术选型指南，针对金融长文档(招股书/年报)场景给出最优性价比方案\r\n\r\n---\r\n\r\n## 新增内容（2026版）\r\n**Step 2 新增技术评估（2026）**：\r\n- LangGraph v1.0生产就绪：状态机工作流/长期记忆/错误恢复三大核心能力，企业级部署支持Kubernetes自动扩缩容，GitHub Stars突破85K\r\n- CrewAI v1.10多智能体协作：支持6种角色类型+并行任务编排，内置20+企业级连接器（Slack/Notion/Airtable/GitHub），2026年Q1新增中文文档\r\n- Claude Agent SDK / OpenAI Agents SDK横向对比：工具调用准确率(94% vs 91%)/上下文利用率(78% vs 82%)/成本效率(¥0.8/千Token vs ¥1.2/千Token)三大维度全面评测\r\n- MCP(Model Context Protocol)生态爆发：50+官方服务器覆盖GitHub/Slack/Notion/Postgres等，企业内部MCP注册表成为新基础设施\r\n- LLM长上下文之战：Gemini 2M Token / Claude 200K / GPT-4o 128K技术选型指南，针对金融长文档(招股书/年报)场景给出最优性价比方案\r\n\r\n---\r\n\r\n## Step 1 — Process Discovery\r\nAsk the user to describe their current workflow:\r\n- What triggers it? (email, schedule, webhook, human action?)\r\n- What are the key steps? (list them in plain language)\r\n- Who (or what system) does each step today?\r\n- Where do errors/delays typically occur?\r\n- What's the desired output/outcome?\r\n\r\n### Step 2 — Automation Suitability Assessment\r\n\r\nScore the workflow across 5 dimensions:\r\n\r\n| Dimension | Score | Notes |\r\n|-----------|-------|-------|\r\n| Repetitiveness | /10 | How often does this run identically? |\r\n| Rule-based | /10 | Are decisions clear-cut or judgment-based? |\r\n| Data availability | /10 | Is input data structured and accessible? |\r\n| Error tolerance | /10 | Can errors be caught and recovered automatically? |\r\n| Stakes | /10 (inverted) | Low-stakes = easier to automate |\r\n| **Automation Score** | /50 | >35 = High priority, 20–35 = Medium, <20 = Keep manual |\r\n\r\n### Step 3 — Agentic Pipeline Design\r\nGenerate a detailed pipeline blueprint:\r\n\r\n```\r\n🎯 Workflow: [Name]\r\n⚡ Trigger: [webhook / cron / event / manual]\r\n🤖 Agents:\r\n ├── Agent 1 [Role]: [Tool 1, Tool 2] → Output: [description]\r\n ├── Agent 2 [Role]: [Tool 3] → Output: [description]\r\n └── Agent 3 [Role]: [Tool 4, Tool 5] → Output: [description]\r\n🔄 Flow: Sequential / Parallel / Conditional\r\n🧠 Memory: [ephemeral / Redis / vector DB]\r\n🚨 Error Handling: [retry / fallback agent / human escalation]\r\n👤 HITL Checkpoints: [list high-stakes decision points]\r\n📊 Output: [final deliverable description]\r\n```\r\n\r\n**Example — Lead Qualification Pipeline:**\r\n```\r\n🎯 Workflow: B2B Lead Qualification & Outreach\r\n⚡ Trigger: New form submission webhook\r\n🤖 Agents:\r\n ├── Enrichment Agent [Clearbit + LinkedIn scraper] → Company profile JSON\r\n ├── Scoring Agent [GPT-4o] → Lead score (0–100) + reasoning\r\n ├── Decision Gate [Human] → Approve for outreach? (HITL)\r\n └── Outreach Agent [Email API + CRM API] → Personalized email + CRM update\r\n🔄 Flow: Sequential with HITL gate\r\n🧠 Memory: PostgreSQL (lead history)\r\n🚨 Error: Retry enrichment 3x → flag for manual review\r\n👤 HITL: Score > 80 auto-approves; 50–80 requires human review; <50 auto-rejects\r\n📊 Output: CRM updated + email queued\r\n```\r\n\r\n### Step 4 — Platform Recommendation\r\n\r\n| Platform | Best For | Agent Support | Self-host | Price |\r\n|----------|----------|--------------|-----------|-------|\r\n| n8n | Technical teams, complex logic | ✅ via AI nodes | ✅ Yes | Free/OSS |\r\n| Make (Integromat) | Non-technical, API integrations | Partial | ❌ No | ~$9+/mo |\r\n| Zapier | Simple triggers, non-technical | Partial | ❌ No | ~$20+/mo |\r\n| LangGraph (custom) | Complex state machines, production | ✅ Native | ✅ Yes | Dev hours |\r\n| CrewAI | Role-based agent teams | ✅ Native | ✅ Yes | Dev hours |\r\n\r\n### Step 5 — n8n Workflow JSON Spec (Sample Output)\r\n```json\r\n{\r\n \"name\": \"Lead Qualification Pipeline\",\r\n \"nodes\": [\r\n {\r\n \"name\": \"Webhook Trigger\",\r\n \"type\": \"n8n-nodes-base.webhook\",\r\n \"parameters\": { \"path\": \"lead-inbound\" }\r\n },\r\n {\r\n \"name\": \"Enrich Lead\",\r\n \"type\": \"@n8n/n8n-nodes-langchain.agent\",\r\n \"parameters\": {\r\n \"promptType\": \"define\",\r\n \"text\": \"Enrich this lead data using Clearbit: {{ $json.email }}\"\r\n }\r\n },\r\n {\r\n \"name\": \"Score Lead\",\r\n \"type\": \"@n8n/n8n-nodes-langchain.openAi\",\r\n \"parameters\": {\r\n \"resource\": \"text\",\r\n \"operation\": \"message\",\r\n \"modelId\": \"gpt-4o\",\r\n \"messages\": { \"values\": [{ \"content\": \"Score this lead 0-100...\" }] }\r\n }\r\n }\r\n ]\r\n}\r\n```\r\n\r\n### Step 6 — ROI Calculator\r\n\r\n| Metric | Before Automation | After Automation | Savings |\r\n|--------|------------------|-----------------|---------|\r\n| Time per run | [X hours] | [Y minutes] | [Z%] |\r\n| Runs per week | [N] | [N] | — |\r\n| Total time saved/week | — | — | [hours] |\r\n| Cost saved/month | — | — | [$$$] |\r\n| Automation setup cost | — | — | [one-time] |\r\n| **Payback period** | — | — | [weeks] |\r\n\r\n## Example Interactions\r\n\r\n**User:** \"I spend 3 hours every Monday pulling sales data from 5 spreadsheets, writing a summary email, and updating our CRM. Can this be automated?\"\r\n\r\n**Skill response:** Scores the workflow (42/50 — High priority), designs a 4-agent pipeline (data collector → analyzer → email writer → CRM updater), recommends n8n as the platform (self-hostable, native AI nodes), generates a complete n8n JSON spec, and estimates 11.5 hours/month saved = ~$580 value at $50/hr.\r\n\r\n---\r\n\r\n**User:** \"I want to build a customer support triage system that reads emails, classifies them, and routes to the right team.\"\r\n\r\n**Skill response:** Designs a HITL-enabled pipeline with email reading, classification, confidence threshold (>85% auto-route, <85% human review), CRM ticket creation, and Slack notification. Recommends LangGraph for its state persistence and human review interrupt capability.\r\n\r\n## Notes & Constraints\r\n\r\n- Always design **HITL checkpoints** for: financial decisions, customer communications, data deletions, external API calls with side effects\r\n- For **regulated industries** (finance, healthcare, insurance): flag compliance requirements\r\n- Workflows involving PII must include data retention and access control considerations\r\n- Recommend starting with a **pilot workflow** (lowest risk, highest frequency) before scaling\r\n- Provide rollback strategies: every agentic workflow should have a manual fallback\n\nFile v3.3.0:_meta.json\n\n{\n \"ownerId\": \"kn74e704j3ygjcygnpf02rdvd185js13\",\n \"slug\": \"agentic-workflow-designer\",\n \"version\": \"3.3.0\",\n \"publishedAt\": 1778909227042\n}\n\nArchive v3.2.0: 2 files, 4886 bytes\n\nFiles: SKILL.md (10122b), _meta.json (144b)\n\nFile v3.2.0:SKILL.md\n\n---\r\nname: Agentic Workflow Designer\r\ndescription: >\r\n AI-powered agentic workflow design and automation assistant — map complex multi-step\r\n processes, identify automation opportunities, design autonomous AI agent pipelines,\r\n generate n8n/Make/Zapier workflow specs, and estimate ROI. Covers enterprise automation,\r\n self-healing workflows, human-in-the-loop patterns, and production deployment. Keywords:\r\n agentic workflow, workflow automation, n8n, Make, Zapier, enterprise automation,\r\n AI pipeline, autonomous agent, process automation, workflow design, ROI calculator,\r\n HITL, 工作流设计, 流程自动化, 智能体工作流, 企业自动化, n8n工作流, 流程优化,\r\n 自主代理, RPA替代.\r\nversion: \"3.0.0\"\r\n---\r\n\r\n# Agentic Workflow Designer\r\n\r\n> From messy manual processes to autonomous AI pipelines — design, document, and deploy.\r\n\r\n## What This Skill Does\r\n\r\nAgentic AI (AI that can autonomously execute multi-step tasks) is the #1 enterprise tech trend in 2026 with a projected $8.5B market and 40% CAGR. Yet most teams struggle to:\r\n- Map which workflows are actually suitable for agentic automation\r\n- Design reliable pipelines that don't break silently\r\n- Choose between n8n, Make, Zapier, or custom agent frameworks\r\n- Justify the ROI to business stakeholders\r\n\r\nThis skill bridges the gap between AI hype and practical workflow automation:\r\n\r\n- **Workflow Discovery** — Identify and prioritize automation opportunities in any business process\r\n- **Agentic Pipeline Design** — Create detailed workflow blueprints with triggers, agents, tools, and fallbacks\r\n- **Platform Selection** — Compare n8n / Make / Zapier / custom LangGraph for your use case\r\n- **Generate Workflow Specs** — Produce JSON/YAML specs importable into n8n or Make\r\n- **ROI Calculator** — Estimate time/cost savings from automation\r\n- **Human-in-the-Loop (HITL) Design** — Design appropriate checkpoints for sensitive decisions\r\n\r\n## Trigger Words\r\n\r\nAgentic workflow, automate my process, workflow automation, n8n, Make automation, Zapier flow, design a workflow, workflow design, process automation, automate with AI, AI pipeline, autonomous workflow, HITL pattern, 工作流设计, 自动化工作流, 流程自动化, 智能体工作流, 帮我设计流程, 自动化这个流程, n8n工作流, 企业自动化, RPA替代, agentic AI pipeline\r\n\r\n## Target Users\r\n\r\n- Operations managers digitizing manual business processes\r\n- Developers building production AI automation systems\r\n- Product managers scoping automation features\r\n- Consultants delivering workflow automation projects\r\n- Entrepreneurs building AI-native products\r\n\r\n## Workflow\r\n\r\n### 新增内容（2026版）\r\n**Step 2 新增技术评估（2026）**：\r\n- LangGraph v1.0生产就绪：状态机工作流/长期记忆/错误恢复三大核心能力，企业级部署支持Kubernetes自动扩缩容，GitHub Stars突破85K\r\n- CrewAI v1.10多智能体协作：支持6种角色类型+并行任务编排，内置20+企业级连接器（Slack/Notion/Airtable/GitHub），2026年Q1新增中文文档\r\n- Claude Agent SDK / OpenAI Agents SDK横向对比：工具调用准确率(94% vs 91%)/上下文利用率(78% vs 82%)/成本效率(¥0.8/千Token vs ¥1.2/千Token)三大维度全面评测\r\n- MCP(Model Context Protocol)生态爆发：50+官方服务器覆盖GitHub/Slack/Notion/Postgres等，企业内部MCP注册表成为新基础设施\r\n- LLM长上下文之战：Gemini 2M Token / Claude 200K / GPT-4o 128K技术选型指南，针对金融长文档(招股书/年报)场景给出最优性价比方案\r\n\r\n---\r\n\r\n## 新增内容（2026版）\r\n**Step 2 新增技术评估（2026）**：\r\n- LangGraph v1.0生产就绪：状态机工作流/长期记忆/错误恢复三大核心能力，企业级部署支持Kubernetes自动扩缩容，GitHub Stars突破85K\r\n- CrewAI v1.10多智能体协作：支持6种角色类型+并行任务编排，内置20+企业级连接器（Slack/Notion/Airtable/GitHub），2026年Q1新增中文文档\r\n- Claude Agent SDK / OpenAI Agents SDK横向对比：工具调用准确率(94% vs 91%)/上下文利用率(78% vs 82%)/成本效率(¥0.8/千Token vs ¥1.2/千Token)三大维度全面评测\r\n- MCP(Model Context Protocol)生态爆发：50+官方服务器覆盖GitHub/Slack/Notion/Postgres等，企业内部MCP注册表成为新基础设施\r\n- LLM长上下文之战：Gemini 2M Token / Claude 200K / GPT-4o 128K技术选型指南，针对金融长文档(招股书/年报)场景给出最优性价比方案\r\n\r\n---\r\n\r\n## Step 1 — Process Discovery\r\nAsk the user to describe their current workflow:\r\n- What triggers it? (email, schedule, webhook, human action?)\r\n- What are the key steps? (list them in plain language)\r\n- Who (or what system) does each step today?\r\n- Where do errors/delays typically occur?\r\n- What's the desired output/outcome?\r\n\r\n### Step 2 — Automation Suitability Assessment\r\n\r\nScore the workflow across 5 dimensions:\r\n\r\n| Dimension | Score | Notes |\r\n|-----------|-------|-------|\r\n| Repetitiveness | /10 | How often does this run identically? |\r\n| Rule-based | /10 | Are decisions clear-cut or judgment-based? |\r\n| Data availability | /10 | Is input data structured and accessible? |\r\n| Error tolerance | /10 | Can errors be caught and recovered automatically? |\r\n| Stakes | /10 (inverted) | Low-stakes = easier to automate |\r\n| **Automation Score** | /50 | >35 = High priority, 20–35 = Medium, <20 = Keep manual |\r\n\r\n### Step 3 — Agentic Pipeline Design\r\nGenerate a detailed pipeline blueprint:\r\n\r\n```\r\n🎯 Workflow: [Name]\r\n⚡ Trigger: [webhook / cron / event / manual]\r\n🤖 Agents:\r\n ├── Agent 1 [Role]: [Tool 1, Tool 2] → Output: [description]\r\n ├── Agent 2 [Role]: [Tool 3] → Output: [description]\r\n └── Agent 3 [Role]: [Tool 4, Tool 5] → Output: [description]\r\n🔄 Flow: Sequential / Parallel / Conditional\r\n🧠 Memory: [ephemeral / Redis / vector DB]\r\n🚨 Error Handling: [retry / fallback agent / human escalation]\r\n👤 HITL Checkpoints: [list high-stakes decision points]\r\n📊 Output: [final deliverable description]\r\n```\r\n\r\n**Example — Lead Qualification Pipeline:**\r\n```\r\n🎯 Workflow: B2B Lead Qualification & Outreach\r\n⚡ Trigger: New form submission webhook\r\n🤖 Agents:\r\n ├── Enrichment Agent [Clearbit + LinkedIn scraper] → Company profile JSON\r\n ├── Scoring Agent [GPT-4o] → Lead score (0–100) + reasoning\r\n ├── Decision Gate [Human] → Approve for outreach? (HITL)\r\n └── Outreach Agent [Email API + CRM API] → Personalized email + CRM update\r\n🔄 Flow: Sequential with HITL gate\r\n🧠 Memory: PostgreSQL (lead history)\r\n🚨 Error: Retry enrichment 3x → flag for manual review\r\n👤 HITL: Score > 80 auto-approves; 50–80 requires human review; <50 auto-rejects\r\n📊 Output: CRM updated + email queued\r\n```\r\n\r\n### Step 4 — Platform Recommendation\r\n\r\n| Platform | Best For | Agent Support | Self-host | Price |\r\n|----------|----------|--------------|-----------|-------|\r\n| n8n | Technical teams, complex logic | ✅ via AI nodes | ✅ Yes | Free/OSS |\r\n| Make (Integromat) | Non-technical, API integrations | Partial | ❌ No | ~$9+/mo |\r\n| Zapier | Simple triggers, non-technical | Partial | ❌ No | ~$20+/mo |\r\n| LangGraph (custom) | Complex state machines, production | ✅ Native | ✅ Yes | Dev hours |\r\n| CrewAI | Role-based agent teams | ✅ Native | ✅ Yes | Dev hours |\r\n\r\n### Step 5 — n8n Workflow JSON Spec (Sample Output)\r\n```json\r\n{\r\n \"name\": \"Lead Qualification Pipeline\",\r\n \"nodes\": [\r\n {\r\n \"name\": \"Webhook Trigger\",\r\n \"type\": \"n8n-nodes-base.webhook\",\r\n \"parameters\": { \"path\": \"lead-inbound\" }\r\n },\r\n {\r\n \"name\": \"Enrich Lead\",\r\n \"type\": \"@n8n/n8n-nodes-langchain.agent\",\r\n \"parameters\": {\r\n \"promptType\": \"define\",\r\n \"text\": \"Enrich this lead data using Clearbit: {{ $json.email }}\"\r\n }\r\n },\r\n {\r\n \"name\": \"Score Lead\",\r\n \"type\": \"@n8n/n8n-nodes-langchain.openAi\",\r\n \"parameters\": {\r\n \"resource\": \"text\",\r\n \"operation\": \"message\",\r\n \"modelId\": \"gpt-4o\",\r\n \"messages\": { \"values\": [{ \"content\": \"Score this lead 0-100...\" }] }\r\n }\r\n }\r\n ]\r\n}\r\n```\r\n\r\n### Step 6 — ROI Calculator\r\n\r\n| Metric | Before Automation | After Automation | Savings |\r\n|--------|------------------|-----------------|---------|\r\n| Time per run | [X hours] | [Y minutes] | [Z%] |\r\n| Runs per week | [N] | [N] | — |\r\n| Total time saved/week | — | — | [hours] |\r\n| Cost saved/month | — | — | [$$$] |\r\n| Automation setup cost | — | — | [one-time] |\r\n| **Payback period** | — | — | [weeks] |\r\n\r\n## Example Interactions\r\n\r\n**User:** \"I spend 3 hours every Monday pulling sales data from 5 spreadsheets, writing a summary email, and updating our CRM. Can this be automated?\"\r\n\r\n**Skill response:** Scores the workflow (42/50 — High priority), designs a 4-agent pipeline (data collector → analyzer → email writer → CRM updater), recommends n8n as the platform (self-hostable, native AI nodes), generates a complete n8n JSON spec, and estimates 11.5 hours/month saved = ~$580 value at $50/hr.\r\n\r\n---\r\n\r\n**User:** \"I want to build a customer support triage system that reads emails, classifies them, and routes to the right team.\"\r\n\r\n**Skill response:** Designs a HITL-enabled pipeline with email reading, classification, confidence threshold (>85% auto-route, <85% human review), CRM ticket creation, and Slack notification. Recommends LangGraph for its state persistence and human review interrupt capability.\r\n\r\n## Notes & Constraints\r\n\r\n- Always design **HITL checkpoints** for: financial decisions, customer communications, data deletions, external API calls with side effects\r\n- For **regulated industries** (finance, healthcare, insurance): flag compliance requirements\r\n- Workflows involving PII must include data retention and access control considerations\r\n- Recommend starting with a **pilot workflow** (lowest risk, highest frequency) before scaling\r\n- Provide rollback strategies: every agentic workflow should have a manual fallback\n\nFile v3.2.0:_meta.json\n\n{\n \"ownerId\": \"kn74e704j3ygjcygnpf02rdvd185js13\",\n \"slug\": \"agentic-workflow-designer\",\n \"version\": \"3.2.0\",\n \"publishedAt\": 1778898183351\n}\n\nArchive v3.1.0: 2 files, 4886 bytes\n\nFiles: SKILL.md (10122b), _meta.json (144b)\n\nFile v3.1.0:SKILL.md\n\n---\r\nname: Agentic Workflow Designer\r\ndescription: >\r\n AI-powered agentic workflow design and automation assistant — map complex multi-step\r\n processes, identify automation opportunities, design autonomous AI agent pipelines,\r\n generate n8n/Make/Zapier workflow specs, and estimate ROI. Covers enterprise automation,\r\n self-healing workflows, human-in-the-loop patterns, and production deployment. Keywords:\r\n agentic workflow, workflow automation, n8n, Make, Zapier, enterprise automation,\r\n AI pipeline, autonomous agent, process automation, workflow design, ROI calculator,\r\n HITL, 工作流设计, 流程自动化, 智能体工作流, 企业自动化, n8n工作流, 流程优化,\r\n 自主代理, RPA替代.\r\nversion: \"3.0.0\"\r\n---\r\n\r\n# Agentic Workflow Designer\r\n\r\n> From messy manual processes to autonomous AI pipelines — design, document, and deploy.\r\n\r\n## What This Skill Does\r\n\r\nAgentic AI (AI that can autonomously execute multi-step tasks) is the #1 enterprise tech trend in 2026 with a projected $8.5B market and 40% CAGR. Yet most teams struggle to:\r\n- Map which workflows are actually suitable for agentic automation\r\n- Design reliable pipelines that don't break silently\r\n- Choose between n8n, Make, Zapier, or custom agent frameworks\r\n- Justify the ROI to business stakeholders\r\n\r\nThis skill bridges the gap between AI hype and practical workflow automation:\r\n\r\n- **Workflow Discovery** — Identify and prioritize automation opportunities in any business process\r\n- **Agentic Pipeline Design** — Create detailed workflow blueprints with triggers, agents, tools, and fallbacks\r\n- **Platform Selection** — Compare n8n / Make / Zapier / custom LangGraph for your use case\r\n- **Generate Workflow Specs** — Produce JSON/YAML specs importable into n8n or Make\r\n- **ROI Calculator** — Estimate time/cost savings from automation\r\n- **Human-in-the-Loop (HITL) Design** — Design appropriate checkpoints for sensitive decisions\r\n\r\n## Trigger Words\r\n\r\nAgentic workflow, automate my process, workflow automation, n8n, Make automation, Zapier flow, design a workflow, workflow design, process automation, automate with AI, AI pipeline, autonomous workflow, HITL pattern, 工作流设计, 自动化工作流, 流程自动化, 智能体工作流, 帮我设计流程, 自动化这个流程, n8n工作流, 企业自动化, RPA替代, agentic AI pipeline\r\n\r\n## Target Users\r\n\r\n- Operations managers digitizing manual business processes\r\n- Developers building production AI automation systems\r\n- Product managers scoping automation features\r\n- Consultants delivering workflow automation projects\r\n- Entrepreneurs building AI-native products\r\n\r\n## Workflow\r\n\r\n### 新增内容（2026版）\r\n**Step 2 新增技术评估（2026）**：\r\n- LangGraph v1.0生产就绪：状态机工作流/长期记忆/错误恢复三大核心能力，企业级部署支持Kubernetes自动扩缩容，GitHub Stars突破85K\r\n- CrewAI v1.10多智能体协作：支持6种角色类型+并行任务编排，内置20+企业级连接器（Slack/Notion/Airtable/GitHub），2026年Q1新增中文文档\r\n- Claude Agent SDK / OpenAI Agents SDK横向对比：工具调用准确率(94% vs 91%)/上下文利用率(78% vs 82%)/成本效率(¥0.8/千Token vs ¥1.2/千Token)三大维度全面评测\r\n- MCP(Model Context Protocol)生态爆发：50+官方服务器覆盖GitHub/Slack/Notion/Postgres等，企业内部MCP注册表成为新基础设施\r\n- LLM长上下文之战：Gemini 2M Token / Claude 200K / GPT-4o 128K技术选型指南，针对金融长文档(招股书/年报)场景给出最优性价比方案\r\n\r\n---\r\n\r\n## 新增内容（2026版）\r\n**Step 2 新增技术评估（2026）**：\r\n- LangGraph v1.0生产就绪：状态机工作流/长期记忆/错误恢复三大核心能力，企业级部署支持Kubernetes自动扩缩容，GitHub Stars突破85K\r\n- CrewAI v1.10多智能体协作：支持6种角色类型+并行任务编排，内置20+企业级连接器（Slack/Notion/Airtable/GitHub），2026年Q1新增中文文档\r\n- Claude Agent SDK / OpenAI Agents SDK横向对比：工具调用准确率(94% vs 91%)/上下文利用率(78% vs 82%)/成本效率(¥0.8/千Token vs ¥1.2/千Token)三大维度全面评测\r\n- MCP(Model Context Protocol)生态爆发：50+官方服务器覆盖GitHub/Slack/Notion/Postgres等，企业内部MCP注册表成为新基础设施\r\n- LLM长上下文之战：Gemini 2M Token / Claude 200K / GPT-4o 128K技术选型指南，针对金融长文档(招股书/年报)场景给出最优性价比方案\r\n\r\n---\r\n\r\n## Step 1 — Process Discovery\r\nAsk the user to describe their current workflow:\r\n- What triggers it? (email, schedule, webhook, human action?)\r\n- What are the key steps? (list them in plain language)\r\n- Who (or what system) does each step today?\r\n- Where do errors/delays typically occur?\r\n- What's the desired output/outcome?\r\n\r\n### Step 2 — Automation Suitability Assessment\r\n\r\nScore the workflow across 5 dimensions:\r\n\r\n| Dimension | Score | Notes |\r\n|-----------|-------|-------|\r\n| Repetitiveness | /10 | How often does this run identically? |\r\n| Rule-based | /10 | Are decisions clear-cut or judgment-based? |\r\n| Data availability | /10 | Is input data structured and accessible? |\r\n| Error tolerance | /10 | Can errors be caught and recovered automatically? |\r\n| Stakes | /10 (inverted) | Low-stakes = easier to automate |\r\n| **Automation Score** | /50 | >35 = High priority, 20–35 = Medium, <20 = Keep manual |\r\n\r\n### Step 3 — Agentic Pipeline Design\r\nGenerate a detailed pipeline blueprint:\r\n\r\n```\r\n🎯 Workflow: [Name]\r\n⚡ Trigger: [webhook / cron / event / manual]\r\n🤖 Agents:\r\n ├── Agent 1 [Role]: [Tool 1, Tool 2] → Output: [description]\r\n ├── Agent 2 [Role]: [Tool 3] → Output: [description]\r\n └── Agent 3 [Role]: [Tool 4, Tool 5] → Output: [description]\r\n🔄 Flow: Sequential / Parallel / Conditional\r\n🧠 Memory: [ephemeral / Redis / vector DB]\r\n🚨 Error Handling: [retry / fallback agent / human escalation]\r\n👤 HITL Checkpoints: [list high-stakes decision points]\r\n📊 Output: [final deliverable description]\r\n```\r\n\r\n**Example — Lead Qualification Pipeline:**\r\n```\r\n🎯 Workflow: B2B Lead Qualification & Outreach\r\n⚡ Trigger: New form submission webhook\r\n🤖 Agents:\r\n ├── Enrichment Agent [Clearbit + LinkedIn scraper] → Company profile JSON\r\n ├── Scoring Agent [GPT-4o] → Lead score (0–100) + reasoning\r\n ├── Decision Gate [Human] → Approve for outreach? (HITL)\r\n └── Outreach Agent [Email API + CRM API] → Personalized email + CRM update\r\n🔄 Flow: Sequential with HITL gate\r\n🧠 Memory: PostgreSQL (lead history)\r\n🚨 Error: Retry enrichment 3x → flag for manual review\r\n👤 HITL: Score > 80 auto-approves; 50–80 requires human review; <50 auto-rejects\r\n📊 Output: CRM updated + email queued\r\n```\r\n\r\n### Step 4 — Platform Recommendation\r\n\r\n| Platform | Best For | Agent Support | Self-host | Price |\r\n|----------|----------|--------------|-----------|-------|\r\n| n8n | Technical teams, complex logic | ✅ via AI nodes | ✅ Yes | Free/OSS |\r\n| Make (Integromat) | Non-technical, API integrations | Partial | ❌ No | ~$9+/mo |\r\n| Zapier | Simple triggers, non-technical | Partial | ❌ No | ~$20+/mo |\r\n| LangGraph (custom) | Complex state machines, production | ✅ Native | ✅ Yes | Dev hours |\r\n| CrewAI | Role-based agent teams | ✅ Native | ✅ Yes | Dev hours |\r\n\r\n### Step 5 — n8n Workflow JSON Spec (Sample Output)\r\n```json\r\n{\r\n \"name\": \"Lead Qualification Pipeline\",\r\n \"nodes\": [\r\n {\r\n \"name\": \"Webhook Trigger\",\r\n \"type\": \"n8n-nodes-base.webhook\",\r\n \"parameters\": { \"path\": \"lead-inbound\" }\r\n },\r\n {\r\n \"name\": \"Enrich Lead\",\r\n \"type\": \"@n8n/n8n-nodes-langchain.agent\",\r\n \"parameters\": {\r\n \"promptType\": \"define\",\r\n \"text\": \"Enrich this lead data using Clearbit: {{ $json.email }}\"\r\n }\r\n },\r\n {\r\n \"name\": \"Score Lead\",\r\n \"type\": \"@n8n/n8n-nodes-langchain.openAi\",\r\n \"parameters\": {\r\n \"resource\": \"text\",\r\n \"operation\": \"message\",\r\n \"modelId\": \"gpt-4o\",\r\n \"messages\": { \"values\": [{ \"content\": \"Score this lead 0-100...\" }] }\r\n }\r\n }\r\n ]\r\n}\r\n```\r\n\r\n### Step 6 — ROI Calculator\r\n\r\n| Metric | Before Automation | After Automation | Savings |\r\n|--------|------------------|-----------------|---------|\r\n| Time per run | [X hours] | [Y minutes] | [Z%] |\r\n| Runs per week | [N] | [N] | — |\r\n| Total time saved/week | — | — | [hours] |\r\n| Cost saved/month | — | — | [$$$] |\r\n| Automation setup cost | — | — | [one-time] |\r\n| **Payback period** | — | — | [weeks] |\r\n\r\n## Example Interactions\r\n\r\n**User:** \"I spend 3 hours every Monday pulling sales data from 5 spreadsheets, writing a summary email, and updating our CRM. Can this be automated?\"\r\n\r\n**Skill response:** Scores the workflow (42/50 — High priority), designs a 4-agent pipeline (data collector → analyzer → email writer → CRM updater), recommends n8n as the platform (self-hostable, native AI nodes), generates a complete n8n JSON spec, and estimates 11.5 hours/month saved = ~$580 value at $50/hr.\r\n\r\n---\r\n\r\n**User:** \"I want to build a customer support triage system that reads emails, classifies them, and routes to the right team.\"\r\n\r\n**Skill response:** Designs a HITL-enabled pipeline with email reading, classification, confidence threshold (>85% auto-route, <85% human review), CRM ticket creation, and Slack notification. Recommends LangGraph for its state persistence and human review interrupt capability.\r\n\r\n## Notes & Constraints\r\n\r\n- Always design **HITL checkpoints** for: financial decisions, customer communications, data deletions, external API calls with side effects\r\n- For **regulated industries** (finance, healthcare, insurance): flag compliance requirements\r\n- Workflows involving PII must include data retention and access control considerations\r\n- Recommend starting with a **pilot workflow** (lowest risk, highest frequency) before scaling\r\n- Provide rollback strategies: every agentic workflow should have a manual fallback\n\nFile v3.1.0:_meta.json\n\n{\n \"ownerId\": \"kn74e704j3ygjcygnpf02rdvd185js13\",\n \"slug\": \"agentic-workflow-designer\",\n \"version\": \"3.1.0\",\n \"publishedAt\": 1778894337149\n}","readmeExcerpt":"Skill: Agentic Workflow Designer Owner: gechengling Summary: AI-powered agentic workflow design and automation assistant — map complex multi-step processes, identify automation opportunities, design autonomous AI agent pipelines, generate n8n/Make/Zapier workflow specs, and estimate ROI. Covers enterprise automation, self-healing workflows, human-in-the-loop patterns, and production deployment. Keywords: agentic work","codeSnippets":[],"executableExamples":[{"language":"text","snippet":"[Workflow]: [Name]\n[Trigger]: [webhook / cron / event / manual]\n[Agents]:\n ├── Agent 1 [Role]: [Tool 1, Tool 2] → Output: [description]\n ├── Agent 2 [Role]: [Tool 3] → Output: [description]\n └── Agent 3 [Role]: [Tool 4, Tool 5] → Output: [description]\n[Flow]: Sequential / Parallel / Conditional\n[Memory]: [ephemeral / Redis / vector DB]\n[Error Handling]: [retry / fallback agent / human escalation]\n[HITL Checkpoints]: [list high-stakes decision points]\n[Output]: [final deliverable description]"},{"language":"text","snippet":"[Workflow]: B2B Lead Qualification & Outreach\n[Trigger]: New form submission webhook\n[Agents]:\n ├── Enrichment Agent [Clearbit + LinkedIn scraper] → Company profile JSON\n ├── Scoring Agent [GPT-4o] → Lead score (0-100) + reasoning\n ├── Decision Gate [Human] → Approve for outreach? (HITL)\n └── Outreach Agent [Email API + CRM API] → Personalized email + CRM update\n[Flow]: Sequential with HITL gate\n[Memory]: PostgreSQL (lead history)\n[Error]: Retry enrichment 3x → flag for manual review\n[HITL]: Score > 80 auto-approves; 50-80 requires human review; <50 auto-rejects\n[Output]: CRM updated + email queued"},{"language":"text","snippet":"[Workflow]: 理赔单据完整性预审\n[Trigger]: 理赔系统上传事件（webhook）\n[Agents]:\n ├── 分类 Agent [OCR + 规则表] → 单据类型与置信度\n ├── 校验 Agent [规则引擎] → 缺失项清单\n ├── 决策门 [规则 + 人工] → 通过 / 退回补件 / 转人工\n └── 通知 Agent [短信 API + 工单 API] → 补件提醒 + 工单创建\n[Flow]: 先并行分类，后串行校验（Conditional）\n[Memory]: PostgreSQL（单据状态机），不含原始影像\n[Error]: OCR 置信度 < 0.85 → 强制转人工，不自动退回\n[HITL]: 涉及拒赔、金额调整、个人信息变更的一律转人工；仅限\"是否缺件\"自动判定\n[Output]: 预审结论 + 缺失项清单 + 工单号"},{"language":"text","snippet":"[Workflow]: 报送数据口径校验与差错清单生成\n[Trigger]: 每日定时任务（cron，T+1 凌晨）\n[Agents]:\n ├── 抽取 Agent [SQL + 文件解析] → 原始报送数据集\n ├── 校验 Agent [规则库] → 差错项清单（含记录定位）\n ├── 归因 Agent [LLM] → 差错原因归类与整改建议\n └── 决策门 [人工] → 确认差错清单与整改责任人\n[Flow]: 串行，校验后可并行归因\n[Memory]: 仅保存差错摘要与规则版本，不落原始明细\n[Error]: 规则库版本未对齐即中止，禁止带疑点报送\n[HITL]: 任何涉及数据修改的动作必须人工确认；仅\"生成差错清单\"可自动\n[Output]: 差错清单 + 归因建议 + 待确认事项"},{"language":"json","snippet":"{\n \"name\": \"Lead Qualification Pipeline\",\n \"nodes\": [\n {\n \"name\": \"Webhook Trigger\",\n \"type\": \"n8n-nodes-base.webhook\",\n \"parameters\": { \"path\": \"lead-inbound\" }\n },\n {\n \"name\": \"Enrich Lead\",\n \"type\": \"@n8n/n8n-nodes-langchain.agent\",\n \"parameters\": {\n \"promptType\": \"define\",\n \"text\": \"Enrich this lead data using Clearbit: {{ $json.email }}\"\n }\n },\n {\n \"name\": \"Score Lead\",\n \"type\": \"@n8n/n8n-nodes-langchain.openAi\",\n \"parameters\": {\n \"resource\": \"text\",\n \"operation\": \"message\",\n \"modelId\": \"gpt-4o\",\n \"messages\": { \"values\": [{ \"content\": \"Score this lead 0-100...\" }] }\n }\n }\n ]\n}"},{"language":"text","snippet":"[Workflow]: [Name]\n[Trigger]: [webhook / cron / event / manual]\n[Agents]:\n ├── Agent 1 [Role]: [Tool 1, Tool 2] → Output: [description]\n ├── Agent 2 [Role]: [Tool 3] → Output: [description]\n └── Agent 3 [Role]: [Tool 4, Tool 5] → Output: [description]\n[Flow]: Sequential / Parallel / Conditional\n[Memory]: [ephemeral / Redis / vector DB]\n[Error Handling]: [retry / fallback agent / human escalation]\n[HITL Checkpoints]: [list high-stakes decision points]\n[Output]: [final deliverable description]"}],"parameters":null,"dependencies":[],"permissions":[],"extractedFiles":[{"path":"SKILL.md","content":"---\nname: Agentic Workflow Designer\ndescription: >\n AI-powered agentic workflow design and automation assistant — map complex multi-step\n processes, identify automation opportunities, design autonomous AI agent pipelines,\n generate n8n/Make/Zapier workflow specs, and estimate ROI. Covers enterprise automation,\n self-healing workflows, human-in-the-loop patterns, and production deployment. Keywords:\n agentic workflow, workflow automation, n8n, Make, Zapier, enterprise automation,\n AI pipeline, autonomous agent, process automation, workflow design, ROI calculator,\n HITL, 工作流设计, 流程自动化, 智能体工作流, 企业自动化, n8n工作流, 流程优化,\n 自主代理, RPA替代.\nversion: \"3.3.6\"\n---\n\n# Agentic Workflow Designer\n\n> From messy manual processes to autonomous AI pipelines — design, document, and deploy.\n\n> **⚠️ CAPABILITY NOTICE / 能力说明**\n> - **Type:** Design and advisory framework — produces workflow blueprints, JSON/YAML specs, and ROI estimates as reference material\n> - **No code is executed by this skill**; generated specs are for the user to review and import into their own environment\n> - **No persistent storage, network calls, background execution, or credential collection**\n> - **All outputs require human review before production deployment**\n> - Workflows touching PII or regulated data must include retention, access control, and audit considerations\n\n## What This Skill Does\n\nAgentic AI (AI that can autonomously execute multi-step tasks) is the #1 enterprise tech trend in 2026 with a projected $8.5B market and 40% CAGR. Yet most teams struggle to:\n- Map which workflows are actually suitable for agentic automation\n- Design reliable pipelines that don't break silently\n- Choose between n8n, Make, Zapier, or custom agent frameworks\n- Justify the ROI to business stakeholders\n\nThis skill bridges the gap between AI hype and practical workflow automation:\n\n- **Workflow Discovery** — Identify and prioritize automation opportunities in any business process\n- **Agentic Pipeline Design** — Create detailed workflow blueprints with triggers, agents, tools, and fallbacks\n- **Platform Selection** — Compare n8n / Make / Zapier / custom LangGraph for your use case\n- **Generate Workflow Specs** — Produce JSON/YAML specs importable into n8n or Make\n- **ROI Calculator** — Estimate time/cost savings from automation\n- **Human-in-the-Loop (HITL) Design** — Design appropriate checkpoints for sensitive decisions\n\n## Trigger Words\n\nAgentic workflow, automate my process, workflow automation, n8n, Make automation, Zapier flow, design a workflow, workflow design, process automation, automate with AI, AI pipeline, autonomous workflow, HITL pattern, 工作流设计, 自动化工作流, 流程自动化, 智能体工作流, 帮我设计流程, 自动化这个流程, n8n工作流, 企业自动化, RPA替代, agentic AI pipeline\n\n## Target Users\n\n- Operations managers digitizing manual business processes\n- Developers building production AI automation systems\n- Product managers scoping automation features\n- Consultants delivering workflow automation projects\n- Entrepreneurs building AI-native prod"},{"path":"_meta.json","content":"{\n \"ownerId\": \"kn74e704j3ygjcygnpf02rdvd185js13\",\n \"slug\": \"agentic-workflow-designer\",\n \"version\": \"3.3.6\",\n \"publishedAt\": 1790227640409\n}"},{"path":"skill-card.md","content":"## Description:\n\nAI-powered agentic workflow design and automation assistant for mapping multi-step processes, identifying automation opportunities, designing autonomous agent pipelines, generating workflow specs, and estimating ROI.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[gechengling](https://clawhub.ai/user/gechengling)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nOperations managers, developers, product managers, consultants, and entrepreneurs use this skill to assess business processes, design agentic workflow blueprints, choose automation platforms, generate n8n/Make/Zapier-oriented specs, and estimate automation ROI.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: Generated workflow specs may be incomplete, incorrect, or unsuitable for direct import into production automation systems.\n\nMitigation: Treat generated specs as drafts and review them before import or deployment.\n\nRisk: Automation designs may include customer-facing, financial, regulated, data-deleting, or external side-effect actions.\n\nMitigation: Add human approval checkpoints for these actions and validate compliance, retention, access control, and audit requirements.\n\n## Reference(s):\n\n- [ClawHub skill page](https://clawhub.ai/gechengling/skills/agentic-workflow-designer)\n\n## Skill Output:\n\n**Output Type(s):** [Text, Markdown, Code, Configuration, Guidance]\n\n**Output Format:** [Markdown with JSON and YAML workflow specifications]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [Produces workflow blueprints, platform recommendations, ROI estimates, and review guidance for human validation before deployment.]\n\n## Skill Version(s):\n\n3.3.6 (source: frontmatter and server release evidence)\n\n## Ethical Considerations:\n\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment."}],"languages":[],"docsSourceLabel":"CLAWHUB","editorialOverview":"AI-powered agentic workflow design and automation assistant — map complex multi-step processes, identify automation opportunities, design autonomous AI agent pipelines, generate n8n/Make/Zapier workflow specs, and estimate ROI. Covers enterprise automation, self-healing workflows, human-in-the-loop patterns, and production deployment. Keywords: agentic workflow, workflow automation, n8n, Make, Zapier, enterprise automation, AI pipeline, autonomous agent, process automation, workflow design, ROI calculator, HITL, 工作流设计, 流程自动化, 智能体工作流, 企业自动化, n8n工作流, 流程优化, 自主代理, RPA替代. Skill: Agentic Workflow Designer Owner: gechengling Summary: AI-powered agentic workflow design and automation assistant — map complex multi-step processes, identify automation opportunities, design autonomous AI agent pipelines, generate n8n/Make/Zapier workflow specs, and estimate ROI. Covers enterprise automation, self-healing workflows, human-in-the-loop patterns, and production deployment. Keywords: agentic work","editorialQuality":{"score":100,"threshold":65,"status":"ready","wordCount":1302,"uniquenessScore":46,"reasons":[]}},"media":{"evidence":{"source":"no-media","verified":false,"confidence":"low","updatedAt":"2026-10-10T01:40:07.365Z","emptyReason":"No screenshots, media assets, or demo links are available."},"primaryImageUrl":null,"mediaAssetCount":0,"assets":[],"demoUrl":null},"ownerResources":{"evidence":{"source":"unclaimed","verified":false,"confidence":"low","updatedAt":"2026-10-10T01:40:07.365Z","emptyReason":"This page has not been claimed by the agent owner."},"hasCustomPage":false,"customPageUpdatedAt":null,"customLinks":[],"structuredLinks":{"docsUrl":null,"demoUrl":null,"supportUrl":null,"pricingUrl":null,"statusUrl":null},"customPage":null},"relatedAgents":{"evidence":{"source":"protocol-neighbors","verified":false,"confidence":"medium","updatedAt":"2026-10-10T07:53:39.653Z","emptyReason":null},"items":[{"id":"8ebccd8e-3863-4187-8355-c3f14e1f9edf","entityType":"agent","canonicalPath":"/agent/iofficeai-aionui","slug":"iofficeai-aionui","name":"AionUi","description":"Free, local, open-source 24/7 Cowork app and OpenClaw for Gemini CLI, Claude Code, Codex, OpenCode, Qwen Code, Goose CLI, Auggie, and more | 🌟 Star if you like it!","url":"https://github.com/iOfficeAI/AionUi","homepage":"https://www.aionui.com","source":"GITHUB_REPOS","protocols":["MCP","OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-10-09T19:11:12.944Z","createdAt":"2026-02-25T03:38:16.584Z","downloads":null},{"id":"b917f68a-ebff-438e-84f8-3f4b2494c0bc","entityType":"agent","canonicalPath":"/agent/activepieces-activepieces","slug":"activepieces-activepieces","name":"activepieces","description":"AI Agents & MCPs & AI Workflow Automation • (~400 MCP servers for AI agents) • AI Automation / AI Agent with MCPs • AI Workflows & AI Agents • MCPs for AI Agents","url":"https://github.com/activepieces/activepieces","homepage":"https://www.activepieces.com","source":"GITHUB_REPOS","protocols":["OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-04-15T02:22:12.426Z","createdAt":"2026-02-25T03:38:12.412Z","downloads":null},{"id":"5cb26759-3a39-483f-94cf-276a98c13bb8","entityType":"agent","canonicalPath":"/agent/cherryhq-cherry-studio","slug":"cherryhq-cherry-studio","name":"cherry-studio","description":"AI productivity studio with smart chat, autonomous agents, and 300+ assistants. Unified access to frontier LLMs","url":"https://github.com/CherryHQ/cherry-studio","homepage":"https://cherry-ai.com","source":"GITHUB_REPOS","protocols":["MCP","OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-04-11T14:38:40.986Z","createdAt":"2026-02-25T03:38:19.379Z","downloads":null},{"id":"6f6582d0-5d76-4f0f-b81d-86520247950b","entityType":"agent","canonicalPath":"/agent/copilotkit-copilotkit","slug":"copilotkit-copilotkit","name":"CopilotKit","description":"The Frontend for Agents & Generative UI. React + Angular","url":"https://github.com/CopilotKit/CopilotKit","homepage":"https://docs.copilotkit.ai","source":"GITHUB_REPOS","protocols":["OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-03-25T09:50:57.846Z","createdAt":"2026-02-25T03:39:14.617Z","downloads":null}],"links":{"hub":"/agent","source":"/agent/source/clawhub","protocols":[{"label":"OpenClaw","href":"/agent/protocol/openclew"}]}}}