loop-5plus1-architecture

Source Metadata

Original cangjie-skill frontmatter from the distillation run:

name: loop-5plus1-architecture
description: |
  设计或审计完整 Loop 系统的组件架构: 5 大组件 + 1 根脊柱。
  当用户需要搭建一个完整的 (而非最小可用的) loop 系统、或审计现有系统的组件完整性时使用。
  不适用于: 最小可用 loop 的设计、或单次任务的自动化。
  关键 trigger: "设计一个完整的 loop 系统"、"我的 loop 缺什么"、"loop 系统怎么组织"。
source_book: "Loop Engineering 视频合集"
source_chapter: 视频3 (小木头) / 视频4 (Idoos Money)
tags: [architecture, system-design, components, blueprint]
related_skills: [loop-three-elements, maker-checker, goal-verification]

5+1 Loop System Architecture — The Design Blueprint for a Complete Loop

R — Reading (Original Text)

"ados money broke it down into 5 pieces, plus a spine. The first piece is the heartbeat (scheduled trigger), the second is the work tree (isolated branches), the third is the skill (project rules), the fourth is the connector (MCP, connecting to real tools), and the fifth is the sub-agent (separating code writing and code review). Finally there's the spine — the memory that supports the entire loop system." — Xiaomutou (Video 3)

I — Interpretation (Self-Account)

Any complete Loop system can be broken down into 5 components + 1 spine:

1. Heartbeat: Scheduled trigger (cron), automatically starts the loop at the appointed time.
2. Work Tree: Isolated workspace, so multiple agents don't step on each other when running in parallel.
3. Skill: Project rules and constraints, written into SKILL.md, automatically read by each agent.
4. Connector: Connect to external tools via MCP (issue systems, databases, etc.).
5. Sub-agent: Separate production and review, to avoid self-production and self-checking.
6. Spine: Persistent memory layer, recording historical operations, attempt results, and to-do items.

The spine is key: Agents forget (limited context), so persistent memory must be established outside the conversation.

A1 — Past Application (Cases in the Book)

Case 1: Boris's multi-Loop system (Video 2)

• Heartbeat: Trigger the support loop every 30 minutes
• Work Tree: Independent branch per agent
• Skill: Project rules written into agents.md
• Connector: MCP connects to Intercom, Stripe, Supabase
• Sub-agent: Separate code writing and code review
• Spine: Shared signals folder

Case 2: Topic Selection Loop (Video 3)

• Heartbeat: cron every day at 8 AM
• Work Tree: Independent branch
• Skill: research + topic-score skills
• Connector: research API
• Sub-agent: topic-score as the rating agent
• Spine: inbox.md file

A2 — Future Trigger (Future Triggers)

1. When building a complete loop system: "I want to build a complete X automation system"
2. When auditing an existing system: "What components is my loop missing?"
3. When expanding from a single loop to multiple loops: "How do I make multiple loops work together?"
4. When promoting loops in a team: Use this architecture as the blueprint for "what a complete loop looks like"

Language Signals: "design a complete loop system", "what is my loop missing", "how should a loop system be organized", "multi-loop collaboration"

Differences from Adjacent Skills:

• loop-three-elements: The minimally viable unit (this skill is the complete system)
• maker-checker: The specific implementation of sub-agent division of labor (this skill is the overall architecture)
• loop-build-path: Build steps (this skill is a static structure)

E — Execution (executable steps)

Step 1: Audit the existing system

Check item by item against the 5+1 checklist:

□ 心跳: 有定时/事件触发吗?
□ Work Tree: 有隔离工作空间吗?
□ Skill: 规则写成了 SKILL.md 吗?
□ 连接器: 接入了外部工具吗?
□ 子智能体: 生产和审查分开了吗?
□ 脊柱: 有持久化记忆吗?

Step 2: Fill in missing components

• Missing heartbeat → add cron / webhook
• Missing Work Tree → add git worktree / separate directory
• Missing Skill → write the instructions as SKILL.md
• Missing connector → integrate MCP
• Missing sub-agent → split maker and checker
• Missing spine → add a state file / kanban

Step 3: Verify component collaboration

• Trigger → execute → verify → record → next trigger (complete closed loop)
• Does each component participate in the process normally?

B — Boundary

Scenarios where this skill should not be used:

1. Minimal viable loop: only need loop-three-elements and that's enough
2. Single task: no need for a complete architecture
3. Already has a stable system: a well-functioning loop does not need refactoring

The author's blind spots and limitations:

• The 5+1 architecture is Idoos Money's original framework and has not been widely validated
• The concept of "spine" is rather vague — it can be a file, database, or kanban, and the author did not provide specific selection guidance
• The video cases are all based on the Claude Code ecosystem; other tools may require different components

Adjacent but easily confused methodologies:

• Microservice architecture: "splitting independent services"; this skill is "splitting loop components"
• MVC pattern: "model-view-controller"; similar in structure but oriented toward loop systems