File size: 3,267 Bytes
549e098 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 | # Loop Gallery
The gallery is for real-world or realistic loop examples contributed by the community.
Use it to show how a loop works in practice, not just what resources describe it. A good gallery entry should be concrete enough that another builder can adapt the pattern without guessing the trigger, state, verification, or escalation rules.
## Contribution Format
Create one file per loop:
```text
gallery/<short-loop-name>.md
```
Use [`gallery/template.md`](template.md) as the starting point.
## Reference Entries
These entries are reference examples, not claimed production deployments. They show the level of specificity expected from future real or anonymized gallery contributions.
- [PR babysitter reference loop](pr-babysitter-reference.md)
- [CI repair reference loop](ci-repair-reference.md)
- [Docs drift reference loop](docs-drift-reference.md)
## Quality Bar
Gallery entries should include:
- the runtime or tool used;
- trigger or cadence;
- work intake source;
- agent roles or delegation pattern;
- workspace and permissions;
- verification gates;
- durable state artifact;
- retry budget and stop condition;
- escalation path;
- screenshots, PR links, issue links, trace IDs, or anonymized receipts when available;
- lessons learned.
Do not include secrets, private customer data, proprietary code, or internal links that readers cannot inspect. If a real example cannot be public, write an anonymized version and say which details were generalized.
## Minimum Useful Case Study
An entry is worth publishing once it answers all of these. If you cannot fill one in, say so explicitly rather than inventing it.
- [ ] **Objective**: the single outcome the loop optimizes for.
- [ ] **Trigger**: schedule, event, or manual bootstrap.
- [ ] **Intake**: where the loop finds work each run.
- [ ] **Runtime / tool**: which runtime and agent system ran it.
- [ ] **Agent roles**: who explores, acts, checks, and decides.
- [ ] **Verification**: the deterministic gate that decided done.
- [ ] **State**: what survived between runs, and where.
- [ ] **Budget**: the retry or runtime cap.
- [ ] **Escalation**: when and how a human took over.
- [ ] **Receipts**: public or anonymized evidence (PR, issue, trace, dashboard, commands).
- [ ] **Lesson learned**: what changed after the loop met reality.
## Safe Anonymization Checklist
Run this before publishing a real-world entry:
- [ ] Remove secrets, tokens, keys, and credentials.
- [ ] Remove private customer data and any PII.
- [ ] Remove internal URLs, hostnames, and dashboards readers cannot reach.
- [ ] Replace personal names with roles (for example "the on-call engineer", "the reviewer").
- [ ] Include only public or anonymized receipts; redact anything in a screenshot you would not paste in plain text.
- [ ] State which details were generalized so readers know what was changed.
## Suggested First Gallery Entries
- PR babysitter loop for a small open-source repository.
- CI repair loop that turns failed checks into small verified patches.
- Docs drift loop that compares CLI output with README examples.
- Deploy verifier loop that watches dashboards and escalates on threshold breaches.
- Feedback clusterer loop that groups issues or support tickets into themes.
|