Chapter 26 · Loop Engineering: From Prompts to Loops

Chapter goals: Understand the core philosophy of "stop prompting, design loops," master the five-building-blocks + Memory architecture, and评估 your own loops for production-readiness using Loop Ready Score.

26.1 Core Philosophy: From Prompts to Loops

Peter Steinberger once said a widelyquoted line:

"You shouldn't be prompting coding agents anymore. You should be designing loops that prompt your agents."

Boris Cherny (head of Claude Code) expressed a similar view:

"I don't prompt Claude anymore. I have loops running that prompt Claude and figuring out what to do. My job is to write loops."

This points to the shift in leverage point: from "crafting a single prompt" to "designing the control loop."

Single Prompt vs. Loop

Dimension	Single Prompt	Loop
What you提供	"What to do next"	"What the final state should look like"
What the agent does	Execute once	Loop until the goal is达成
Whojudges completion	You	Verifiable stop条件
Can you walk away?	No	Yes — input /goal and leave

26.2 Five Building Blocks + Memory

Loop Engineering拆解 all loops into five basic blocks, plus a Memory layer贯穿始终:

Block	Responsibility	Typical Implementation
Automations / Scheduling	定时 discovery + triage	cron, GitHub Actions, /loop command
Worktrees	Safe parallel execution	git worktree add, loop-worktree tool
Skills	Persistent project knowledge	SKILL.md, skills directory
Plugins & Connectors	Connect to real tools	MCP server, GitHub API, Slack bot
Sub-agents	maker/checker separation	Independent sub-agent execution + verification
+ Memory / State	Cross-session persistence	STATE.md, PROGRESS.md, git commits

Loop Anatomy Diagram

flowchart LR
    A[Schedule / Automation] --> B[Triage Skill]
    B --> C[Read + Write STATE / Memory]
    C --> D[Isolated Worktree]
    D --> E[Implementer Sub-agent]
    E --> F[Verifier Sub-agent<br/>tests + gates]
    F --> G[MCP / Git / Tickets]
    G --> H{Human Gate?}
    H -->|safe / allowlisted| I[Commit / PR / Action]
    H -->|risky / ambiguous| J[Escalate to human<br/>with full context]
    I --> A
    J --> A

26.3 Four Loop Types

Type	Trigger	Stop Condition	Typical Scenario
Turn-based	Manually input each prompt	Agent认为完成 or you打断	Small tasks, exploratory work
Goal-based	Given a goal	Independent evaluator confirms完成 or reaches上限	/goal complex tasks
Time-based	Scheduled timing	Manually stop or the task自然 exits	/loop monitoring, periodic checks
Event-driven	External events (PR opened, CI failed)	Exit after处理 or reach retry上限	CI/CD integration, responsive workflows

Distinguish /goal vs. /loop

Both带 "loop" but solve不同 problems:

• /goal: one大 task; loops until完成 (a marathon)
• /loop: one小动作;周期 repeated execution (an alarm clock)

26.4 Loop Ready Score

The official CLI provides a loop audit command that scans the repo and gives a 0–100 Loop Ready Score, measuring whether the loop具备 production-readiness. Audit dimensions include:

Dimension	Check Content
Skills	Whether a compact-format triage skill exists; whether the skill description is "boring but specific"
State	Whether the state file schema is明确; whether each round读旧写新
Maker/Checker Separation	Whether the implementer and verifier are独立
Budget	Whether there is a budget定义 and over-limit actions
Constraints	Whether path denylists, attempt上限, and other constraints are成文
Governance / Run log	Whether there is a run log and LOOP.md self-description

Chapter Summary

• The core of Loop Engineering is "design loops, not写 prompts";
• Five blocks (Automations/Worktrees/Skills/Plugins&Connectors/Sub-agents) + Memory form the完整架构;
• Four loop types correspond to different trigger and stop scenarios;
• Loop Ready Score provides a quantitative评估 of production-readiness.

🧪 随堂测验

点击你认为正确的选项。答错时会展示正确答案与原因解析。

1. Which phrase best概括izes the core philosophy of Loop Engineering?

2. Which of the following is NOT one of Loop Engineering's five building blocks?

3. What is the core difference between /goal and /loop?

4. When Loop Ready Score is below 60, what does the official建议?

🛠️ Hands-on Practice

1. Use npx @cobusgreyling/loop init . --pattern daily-triage --tool claude to initialize a Daily Triage loop; run loop doctor to查看 the health status.
2. Design a /goal loop for a simple task;明确 the goal, verification method, and stop condition.
3. Read the output of loop-audit;理解 each audit dimension; and改进 your loop针对 the lowest-scoring dimension.

After completing the exercises, proceed to Next chapter: Seven Production Loop Patterns.