AI Agent Workflow & Skill Architecture Guide — turn expert work into measurable, reusable agent systems. Covers workflow discovery, skill decomposition, harness design, evaluation, human escalation, observability, cost control, and multi-agent orchestration. Includes a measured marketing-delivery case: 15 people × 3–4 weeks reduced to one strategist + AI in 5 days. By Gingiris.
インストール方法を見るインストールする前に、エージェントに与えられる指示の中身を確認できます。
适用于：把研究、营销、运营、分析、内容生产等高认知任务，改造成可测量、可纠错、可复用的 Agent 工作流。
只有同时满足以下多数条件才进入自动化：
如果任务低频、目标持续变化、没有验收口径，先做人工 SOP，不要先搭多 Agent。
选择 10–30 个近期真实任务，记录人工基线：
| 指标 | 定义 |
|---|---|
| 任务成功率 | 首次交付通过验收的任务数 / 总任务数 |
| 一次通过率 | 无返工即通过的任务数 / 总任务数 |
| 周期 | 从收到完整输入到可交付输出的 elapsed time |
| 人工工时 | 研究、制作、复核、返工所花人时 |
| 单次成本 | 模型、工具、数据和人工复核成本之和 |
| 重试率 | 发生工具重试或整段重做的任务占比 |
| 严重错误率 | 错误发布、错误付款、数据泄露等高风险事件占比 |
没有这张基线表，就只能证明 Agent “能跑”，不能证明工作流变好了。
先画业务链路，再按可验收结果拆 skill：
需求澄清 → 数据获取 → 证据整理 → 分析 → 产出 → 质检 → 人工批准 → 交付 → 反馈沉淀
每个 skill 至少包含：
name: competitor-evidence-pack
input_contract:
required: [product, market, competitors, time_window]
output_contract:
required: [claims, source_urls, captured_at, confidence, unknowns]
tools:
allow: [search, fetch]
deny: [publish, delete, payment]
acceptance:
- every material claim has a source
- source capture time is recorded
- unknown facts are labeled, not guessed
escalate_when:
- authenticated source is inaccessible
- sources conflict on a decision-critical fact
优先做单一职责 skill。只有当步骤间存在清晰依赖时，才增加 orchestrator。
Prompt 只描述一次交互；harness 管理长期运行环境。至少包含五层：
一次失败的正确处理方式不是无限加提示词，而是：记录失败类型 → 判断是数据、工具、推理还是验收问题 → 修改对应层 → 用旧样例集回归。
| 模式 | 适用情况 | 主要风险 |
|---|---|---|
| 顺序 | 后一步严格依赖前一步输出 | 上游错误级联 |
| 并行 | 多个独立来源或方案可同时产生 | 合并冲突、重复成本 |
| 路由 | 不同任务应调用不同专长 | 分类错误 |
| 主管—执行者 | 任务可拆成多个独立子任务 | 主管成为瓶颈 |
| 评审—修订 | 输出有明确 rubric，可迭代改进 | 无界循环、成本失控 |
默认从单 Agent + 多 skill 开始。只有观测数据证明吞吐或专长隔离确实需要并发，才升级为多 Agent。
以下动作默认需要人工批准：
连续任务不要重复索取同一授权；记录授权对象、范围和有效期。权限不足时返回缺失项和恢复路径，不要假装完成。
按四阶段推进：
每次版本变更比较同一批任务的成功率、周期、人工工时、成本和严重错误率。任一安全指标恶化，回滚到上一稳定版本。
来源：Gingiris 飞书会议纪要《AI agent实践落地困境与规模化尝试分析》，2026-05-10。以下是会议中的经验陈述，不是独立审计或随机对照实验。
它支持“把专家判断编码为 skill + harness，可以减少交付周期和边际人工”的判断；它不证明所有行业都能获得相同幅度，也没有披露统一口径下的错误率、模型成本和客户长期留存。因此复用时必须重新建立本团队基线，并补测质量、重试和严重错误率。
每次工作流评审必须交付：
Built by Gingiris. Historical figures are labeled as reported case evidence; do not present them as guaranteed outcomes.
まだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
Launch your AI product to global attention — the playbook behind Manus, Devin, and AFFiNE's breakout launches. Covers AI-specific GTM strategy, hype cycle management, waitlist tactics, and multi-market rollout for maximum day-one impact.
日本語の概要は準備中です。原文の説明を表示しています。
First AI Product Launch — Zero to Live in 30 Days. Step-by-step guide for shipping your first AI product from idea validation to paying customers. Includes day-by-day timeline and real examples. By @WeiYipei.
日本語の概要は準備中です。原文の説明を表示しています。
Beginner-friendly App Store Optimization guide covering keyword research, screenshot design, rating management, and A/B testing for iOS and Android. By @WeiYipei — practical ASO for indie developers and small teams.
日本語の概要は準備中です。原文の説明を表示しています。
Complete B2B marketing pipeline combining LinkedIn content, cold email sequences, and webinar funnels. Designed for SaaS founders doing $0–$1M ARR who need predictable lead generation. By @WeiYipei.
日本語の概要は準備中です。原文の説明を表示しています。
You launched an ambassador program. 3 applications. One ghosted after week 2. Should you lower the bar? Offer more perks? Spam influencers? This gives you the community & ambassador operations SOP with volunteer, paid tutorial creator and code-contributor tracks — from pre-launch checklist to recruitment to tiered management to retention to governance. Built from Notion (20M users, real ambassador interviews), AFFiNE (60K stars), Asana, and ClickUp programs. By @WeiYipei. 🇨🇳 你搞了个大使计划。来了3个申请。一个第二周就消失了。该降门槛？加福利？群发KOL？这份SOP给你从启动前置条件到招募到留存到治理的完整社区/大使运营方法论。基于 Notion（2000万用户，大使真实访谈）、AFFiNE（60K stars）、Asana、ClickUp 实战案例。 🇯🇵 アンバサダープログラムを立ち上げた。応募3件。1人は2週間で消えた。基準を下げる？特典を増やす？このSOPは、事前チェックリストから採用、段階管理、リテンション、ガバナンスまでの完全なコミュニティ＆アンバサダー運営フレームワークを提供します。Notion（2000万ユーザー）、AFFiNE（60K stars）の実例から構築。 🇰🇷 앰배서더 프로그램을 시작했습니다. 지원 3건. 하나는 2주 만에 사라졌습니다. 기준을 낮출까요? 이 SOP는 사전 체크리스트부터 모집, 단계별 관리, 리텐션, 거버넌스까지 완전한 커뮤니티 & 앰배서더 운영 프레임워크를 제공합니다. Notion(2000만 사용자), AFFiNE(60K stars) 실전 사례에서 구축. Triggers: "ambassador program" | "community building" | "community management" | "DevRel" | "developer relations" | "community growth" | "大使计划" | "社区运营" | "社区搭建" | "コミュニティ運営" | "アンバサダー" | "커뮤니티 운영" | "앰배서더" | "open source community" | "user retention"
Build and grow developer and user communities from scratch — covering ambassador programs, community-led growth (CLG), event operations, content strategy, and health metrics. Grounded in real playbooks from Notion, Lovable, AFFiNE, ClickUp, Asana, and Lark. Use when: building a community for an open-source project, SaaS product, or brand; recruiting and managing ambassadors or community leaders; planning community events (online/offline meetups, AMAs, hackathons); designing ambassador tier/points systems; setting community health KPIs; launching a Build-in-Public strategy; setting up Discord, Slack, or Lark communities; running developer relations (DevRel); or when the user asks about community operations, ambassador programs, CLG, user advocacy, KOC/KOL cultivation, UGC flywheels, community monetization, or any variation of "how do I grow my community".
日本語の概要は準備中です。原文の説明を表示しています。