Chapter 27 · Seven Production Loop Patterns
Chapter goals: Master the seven most practical production loop patterns in the Loop Engineering ecosystem; understand each pattern's节奏, required skills, and token cost; select appropriately for real projects.
27.1 Pattern Overview
There are seven officially认证 production loop patterns,排列 by complexity and token cost ascending:
| Pattern | Cadence | Autonomy Level | Token Cost |
|---|---|---|---|
| Daily Triage | 1 day – 2 hours | L1 | Low |
| PR Babysitter | 5 – 15 minutes | L1 | High |
| CI Sweeper | 5 – 15 minutes | L2 (cautious) | Very High |
| Dependency Sweeper | 6 hours – 1 day | L2 | Medium |
| Changelog Drafter | 1 day or tag | L1 | Low |
| Post-Merge Cleanup | 1 day – 6 hours | L1 | Low |
| Issue Triage | 2 hours – 1 day | L1 | Low |
27.2 Daily Triage
Goal
Auto-generate a priority-sortedissue清单 at the start of each day, letting the team know what to do today.
Required Skills
loop-triage: reads CI, issues, commits, chat; produces priority findingsminimal-fix(optional, Phase 2): drafts a minimal fix for明显 failures- Reviewer sub-agent or skill (optional):验证 the proposed fix
State Management
Use STATE.md as the记忆 backbone:
# Loop State — Project X
Last run: 2026-06-09 08:15 UTC
## High Priority
- [ ] #1241 — flaky test in auth flow (CI red on main)
Loop action: Opened worktree. Fix proposed. Waiting for human PR review.
## Watch List
- PR #1238 open 4 days with no activity.Recommended Cadence
- Morning triage:
/loop 1d - Active sprintperiod:
/loop 2h - Team without TUI: GitHub Action cron
0 8 * * 1-5
27.3 PR Babysitter
Goal
Auto-track PRs after opening;remind stale PRs;自动 comment or标记 when necessary.
Cadence and Cost
- Check every 5–15 minutes
- Token cost is高 (frequent LLM calls to judge状态)
- Recommend L1 only observation; no auto operations
Key Constraints
- Don't auto-close or merge PRs
- Don't replace the human reviewer's judgment
- Only remind; only mark; only summarize
27.4 CI Sweeper
Goal
Auto-diagnose the cause when CI fails;尝试修复;开 new PR.
Risks and Constraints
- Token cost is极高 (involves code generation + test running)
- Must start at L2: first have a stable L1 triage record for 1–2 weeks
- Must have an独立 verifier (verification sub-agent)
- Must have an attempt上限 (e.g. 3 times)
Safety Boundaries
# loop-budget.md
max_tokens_per_run: 50000
max_attempts: 3
denylist_paths:
- "config/prod-secret*"
- ".env"27.5 Dependency Sweeper
Goal
定期 check dependency security vulnerabilities and outdated packages; auto开 PR to upgrade.
Cadence
- Every 6 hours to 1 day
- Token cost: medium
Pattern Characteristics
- Suitable for L2: can auto-upgrad非破坏性 dependencies -破坏性 upgrades need人工 confirmation
27.6 Changelog Drafter
Goal
Auto-generate a CHANGELOG.md draft based on git commits.
Cadence
- Manually triggered (at tag time) or once daily
Characteristics
- Token cost低
- L1 is enough: only draft; don't改 CHANGELOG
- Human最终 approval
27.7 Post-Merge Cleanup
Goal
Auto-clean branches, delete临时 files, update docs after a PR is merged.
Cadence
- Once every 1 day – 6 hours
Characteristics
- Token cost低
- L1: only cleanup; don't touch business code
27.8 Issue Triage
Goal
When a new Issue arrives, auto-classify,标记 priority, assign labels.
Cadence
- Every 2 hours – 1 day
Characteristics
- L1: only建议; don't auto-assign
- Can结合 project历史 issue to learn classification patterns
27.9 Pattern Selection Quick Reference
| Your Scenario | Recommended Pattern |
|---|---|
| Team doesn't know what to do每天 | Daily Triage |
| PRs经常石沉大海 | PR Babysitter |
| CI经常挂且没人修 | CI Sweeper (L2+) |
| Dependency vulnerabilities堆积 | Dependency Sweeper |
| CHANGELOG从不更新 | Changelog Drafter |
| Branches are a mess aftermerge | Post-Merge Cleanup |
| Issue labels are混乱 | Issue Triage |
Chapter Summary
- Seven patterns各有定位; ranging from L1 to L2;
- Token cost ranges from low to very high; selection需权衡价值与成本;
- Safety principle: prove you can see accurately at L1 first; then consider L2 allowing hands-on;
- An interactive pattern picker is available to辅助 decision-making.
🧪 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 loop pattern has the highest token cost?
2. Regarding CI Sweeper safety requirements, which is NOT必须的?
3. At which autonomy level is the Issue Triage pattern通常建议 to stay?
4. Which of the following is NOT in the recommended cadence for Daily Triage?
🛠️ Hands-on Practice
- Use
npx @cobusgreyling/loop init . --pattern daily-triage --tool claudeto initialize a Daily Triage loop; observe the generated file structure. - Design a CI Sweeper loop (L1 reporting phase only) for a simple GitHub Actions CI failure scenario; write your STATE.md template.
- Compare the token cost difference between the Changelog Drafter and Issue Triage patterns;解释 why the former is lower.
After completing the exercises, proceed to Next chapter: Autonomy Levels and Loop Ready Scoring.