我推荐这个网址： https://t.co/qCvjAtWbU2 收录了几十个大神的 loop engineering的思路 https://t.co/1CmoCwF3Yb

loops! | Pre-built agent loops

Source: https://loops.elorm.xyz/

Stop prompting. Startlooping.

Copy closed-loop workflows for coding agents. Each loop includes triggers, feedback gates, and exit conditions so agents self-pace until the job is done.

Trending loop

Ship PR Until Green

Implement on a branch, run tests, push, open a PR, wait for CI, and loop until checks pass and the PR is ready to merge.

View Start the “Ship PR Until Green” loop.

1.4k copies

Featured loops

Battle-tested loops with copy-paste kickoffs and optional install bundles.

Start the “Ship PR Until Green” loop. Goal: PR is open with all CI checks passing Max iterations: 10 Between iterations run: gh pr checks Exit when: all PR checks are success Step 1: Implement the change, test locally, push, open PR, and fix CI until green. Self-pace this loop. After each iteration, run the check command, read the output, and only continue if the exit condition is not met. Stop when the exit condition passes or max iterations is reached. Give a short status update each pass.

Claude CodeCursorCodex

pr

ci

ship

continuous-claude

Install and run the “Pre-Commit Guard” loop. Goal: block git commits when tests are failing. Between iterations run: npm test Exit when: tests exit 0 before each commit. Step 1: Before any git commit, run tests. Fix failures before committing.

CursorClaude Code

hooks

testing

git

pre-commit

Install and run the “Post-Edit Test Guard” loop. Goal: after each batch of file edits, related tests must pass before continuing. Between iterations run: npm test -- --findRelatedTests <edited files> Exit when: related tests exit 0. Step 1: After edits, run related tests. If they fail, fix before making more changes.

CursorClaude Code

Start the “Independent Verifier Pass” loop. Goal: build, lint, and tests pass under independent verification Max iterations: 8 Between iterations run: npm run build && npm run lint && npm test Exit when: all verifier commands exit 0 Step 1: Run build, lint, and tests as a verifier. Trust only command output, not prior claims. Self-pace this loop. After each iteration, run the check command, read the output, and only continue if the exit condition is not met. Stop when the exit condition passes or max iterations is reached. Give a short status update each pass.

Claude CodeCursor

verification

testing

letitloop

/loop 5m Start the “CI Failure Watcher” loop. Goal: latest CI run on this branch is green. Max iterations: 12. Between iterations run: gh run list --branch $(git branch --show-current) --limit 1 Exit when: latest run conclusion is success. Step 1: Check CI status. If failed, read logs, fix root cause, verify locally, and push if needed.

CursorClaude Code

/loop 15m Start the “Deploy Verification Loop”. Goal: all post-deploy health and smoke endpoints return success. Max iterations: 8. Between iterations run: curl -fsS <your-health-url> Exit when: every configured endpoint succeeds. Step 1: Hit health/smoke URLs. If any fail, inspect deploy logs and fix or escalate.

CodexCursorClaude Code

Browse all loops

烟花老师 (@teach_fireworks): Loop Engineering 精华文章汇总!

2026 年 Agent 开始聚焦在长任务后，重点慢慢变成了：

如何设计一个能够持续思考、执行、观察、验证和演进的循环系统？

从 Codex 到 Claude Code，从 OpenHands 到各种 Coding Agent。

业余项目和生产级系统之间最大的差距是Harness 工程，包括 Loop。

Agent