How to turn your manual checks into skills, so Claude closes its own feedback loop.
Most agentic coding：https://claude.com/blog/introduction-to-agentic-coding sessions follow a loop: you ask for a change, Claude gathers context, takes action, verifies the results, and if needed, loops back to gather additional context.
Verification is how agents check their work before responding. Claude already does some of this from observing the deterministic signals in your codebase, including type checkers, linters, tests, and runtime errors. Whatever Claude can't infer becomes the steps you take to manually check a feature.
These manual steps, however, can be transformed into verification loops. In Claude Code：https://claude.com/product/claude-code, a verification loop is an iterative process where Claude checks and attempts to fix the work.
In this article, we cover the most common types of verification loops and show you what we use inside Anthropic. Then we’ll show how to encode the manual checks you already do as skills, so Claude can close its own feedback loop and you can work on something else while it iterates.
Before diving into designing custom verification loops, it can be helpful to understand the built-in support Claude has for a number of different verification loops. Common features and approaches include:
When you have an existing project and you find yourself making the same small corrections every time Claude implements a new feature for you, it’s time to turn those steps into your own custom verification loop. The first step is to write down everything that you find yourself doing every time
The same goes if you're starting a new project and need to figure out how the project should behave. Write the best-practices version in plain English, the way you'd hand it to a new teammate on day one.
If you're struggling to articulate the verification check itself, ask Claude for best practices first and edit from there. Your version probably differs on a few specific points, and those differences are exactly what you want to capture.
Pro tip : The check doesn't have to be qualitative to belong here. "Reject any migration that drops a column without a backfill step" is a deterministic rule no generic linter will catch but a project-specific one will. Anything you keep having to enforce by hand as a manual check qualifies for capture as a loop.
The most common way to encode repetitive steps into a verification loop is to write it as a skill：https://claude.com/blog/complete-guide-to-building-skills-for-claude, and the fastest way to create a skill is to install the skill-creator plugin and let Claude interview you:
You can also hand-write a skill by dropping a markdown file in .claude/skills/ inside your project. The simplest possible verification skill is a few lines of frontmatter plus a body:
The full schema and the philosophy behind it are in our complete guide to building skills：https://claude.com/blog/complete-guide-to-building-skills-for-claude.
The next thing to determine will be how the verification loop kicks off: standalone, embedded, chained, or tied to PR.
You invoke it deliberately, after the artifact exists. A standalone skill earns its place for cross-cutting checks that don't apply every time: a pre-commit security scan, a pre-PR accessibility audit, license-header verification across a repo. Anything you want available across many workflows but don't want firing on every code change.
The cost is that each invocation is still a turn you have to remember to take. The signal that you've outgrown standalone is when you're running it after every change. At that point, the procedure has earned a permanent home: embed it or chain it.
Fires automatically as part of the producing skill. The check belongs to one specific workflow, and the workflow now runs it without you asking.
The simplest version is a one-line append to the producing skill's body:
Verify the embed works by invoking the skill on a fresh task and confirming the new step runs as part of the output. If it doesn't, the skill's description or earlier instructions aren't pulling the appended check in.
Embedded only works on skills you can edit: ones you wrote yourself, or ones installed at a project level where the SKILL.md file is under your control. Built-in skills and plugin-managed skills (the kind that get overwritten on update) are off-limits for this pattern; for those, chain instead.
Skip embedded for checks that span workflows; those want standalone, so you can invoke them from any context.
One skill calls another at its end, and several verified handoffs run end-to-end.
Members of Anthropic's Claude Code team use this pattern in their day-to-day: /code-review hunts for bugs, /simplify cleans up the diff, a /verify skill confirms end-to-end behavior, and a custom /design skill checks against guidelines in a DESIGN.md file if the change touched UI.
Chaining is also how you add verification to a skill you can't modify: build a custom wrapper skill that invokes the original, then invokes your verification skill, as depicted below:
What started as a habit ("I always run /verify after /simplify") becomes a contract ("/simplify always runs /verify when it finishes"). The chain runs the whole dev cycle on its own. You only step in when something escalates back to you.
You can skip chaining when the steps are independent enough that you sometimes want to run one without the others; chaining trades flexibility for automation. Chained verification loops can increase token spend, so it's best to test these loops before deploying them broadly.
Once the chain is solid for your own changes, the same procedure can run on every PR. A teammate's change passes the same gates yours did, whether they remembered to invoke the chain or not. The infrastructure is the same kind of thing as the chain you already wrote, one step further along: the same skills, the same rubrics, the same standards, applied without depending on the author's diligence.
This is where verification stops being personal infrastructure and becomes team infrastructure：https://claude.com/blog/how-claude-code-works-in-large-codebases-best-practices-and-where-to-start. The check you wrote down to save yourself two minutes a week is now saving everyone two minutes a week, on every change. Hold off on PR-wide gates while the chain is still in flux; every adjustment becomes a team-visible event.
Once you have the process down, you’re ready to expand your loop engineering. The verification loop creation process is consistent, no matter what you’re automating or in what environment:
The more you can encode for Claude to follow, the more often Claude's response will land closer to what you want on the very first try. The corrections you no longer have to fiddle with now free up your attention for the individual and exclusive work that no skill can write down for you.
Get started with verification loops in Claude Code ：https://www.anthropic.com/product/claude-code.
This article was written by Delba de Oliviera, a member of the Claude Code team.
Explore more product news and best practices for teams building with Claude.
Product updates, how-tos, community spotlights, and more. Delivered monthly to your inbox.
Hi Claude! Could you help me develop a unique voice for an audience? If you need more information from me, ask me 1-2 key questions right away. If you think I should upload any documents that would help you do a better job, let me know. You can use the tools you have access to— like Google Drive, web search, etc.—if they’ll help you better accomplish this task. Do not use analysis tool. Please keep your responses friendly, brief and conversational. Please execute the task as soon as you can—an artifact would be great if it makes sense. If using an artifact, consider what kind of artifact (interactive, visual, checklist, etc.) might be most helpful for this specific task. Thanks for your help!
Hi Claude! Could you improve my writing style? If you need more information from me, ask me 1-2 key questions right away. If you think I should upload any documents that would help you do a better job, let me know. You can use the tools you have access to— like Google Drive, web search, etc.—if they’ll help you better accomplish this task. Do not use analysis tool. Please keep your responses friendly, brief and conversational. Please execute the task as soon as you can—an artifact would be great if it makes sense. If using an artifact, consider what kind of artifact (interactive, visual, checklist, etc.) might be most helpful for this specific task. Thanks for your help!
Hi Claude! Could you brainstorm creative ideas? If you need more information from me, ask me 1-2 key questions right away. If you think I should upload any documents that would help you do a better job, let me know. You can use the tools you have access to— like Google Drive, web search, etc.—if they’ll help you better accomplish this task. Do not use analysis tool. Please keep your responses friendly, brief and conversational. Please execute the task as soon as you can—an artifact would be great if it makes sense. If using an artifact, consider what kind of artifact (interactive, visual, checklist, etc.) might be most helpful for this specific task. Thanks for your help!
Hi Claude! Could you explain a complex topic simply? If you need more information from me, ask me 1-2 key questions right away. If you think I should upload any documents that would help you do a better job, let me know. You can use the tools you have access to— like Google Drive, web search, etc.—if they’ll help you better accomplish this task. Do not use analysis tool. Please keep your responses friendly, brief and conversational. Please execute the task as soon as you can—an artifact would be great if it makes sense. If using an artifact, consider what kind of artifact (interactive, visual, checklist, etc.) might be most helpful for this specific task. Thanks for your help!
Hi Claude! Could you help me make sense of these ideas? If you need more information from me, ask me 1-2 key questions right away. If you think I should upload any documents that would help you do a better job, let me know. You can use the tools you have access to— like Google Drive, web search, etc.—if they’ll help you better accomplish this task. Do not use analysis tool. Please keep your responses friendly, brief and conversational. Please execute the task as soon as you can—an artifact would be great if it makes sense. If using an artifact, consider what kind of artifact (interactive, visual, checklist, etc.) might be most helpful for this specific task. Thanks for your help!
Hi Claude! Could you prepare for an exam or interview? If you need more information from me, ask me 1-2 key questions right away. If you think I should upload any documents that would help you do a better job, let me know. You can use the tools you have access to— like Google Drive, web search, etc.—if they’ll help you better accomplish this task. Do not use analysis tool. Please keep your responses friendly, brief and conversational. Please execute the task as soon as you can—an artifact would be great if it makes sense. If using an artifact, consider what kind of artifact (interactive, visual, checklist, etc.) might be most helpful for this specific task. Thanks for your help!
Hi Claude! Could you explain a programming concept? If you need more information from me, ask me 1-2 key questions right away. If you think I should upload any documents that would help you do a better job, let me know. You can use the tools you have access to— like Google Drive, web search, etc.—if they’ll help you better accomplish this task. Do not use analysis tool. Please keep your responses friendly, brief and conversational. Please execute the task as soon as you can—an artifact would be great if it makes sense. If using an artifact, consider what kind of artifact (interactive, visual, checklist, etc.) might be most helpful for this specific task. Thanks for your help!
Hi Claude! Could you look over my code and give me tips? If you need more information from me, ask me 1-2 key questions right away. If you think I should upload any documents that would help you do a better job, let me know. You can use the tools you have access to— like Google Drive, web search, etc.—if they’ll help you better accomplish this task. Do not use analysis tool. Please keep your responses friendly, brief and conversational. Please execute the task as soon as you can—an artifact would be great if it makes sense. If using an artifact, consider what kind of artifact (interactive, visual, checklist, etc.) might be most helpful for this specific task. Thanks for your help!
Hi Claude! Could you vibe code with me? If you need more information from me, ask me 1-2 key questions right away. If you think I should upload any documents that would help you do a better job, let me know. You can use the tools you have access to— like Google Drive, web search, etc.—if they’ll help you better accomplish this task. Do not use analysis tool. Please keep your responses friendly, brief and conversational. Please execute the task as soon as you can—an artifact would be great if it makes sense. If using an artifact, consider what kind of artifact (interactive, visual, checklist, etc.) might be most helpful for this specific task. Thanks for your help!
Hi Claude! Could you write grant proposals? If you need more information from me, ask me 1-2 key questions right away. If you think I should upload any documents that would help you do a better job, let me know. You can use the tools you have access to — like Google Drive, web search, etc. — if they’ll help you better accomplish this task. Do not use analysis tool. Please keep your responses friendly, brief and conversational. Please execute the task as soon as you can - an artifact would be great if it makes sense. If using an artifact, consider what kind of artifact (interactive, visual, checklist, etc.) might be most helpful for this specific task. Thanks for your help!
情报判断
Aioga 编辑摘要
Aioga 编辑摘要：Anthropic 介绍了如何在 Claude Code 中将手动检查转化为验证循环，让 Claude 自动检查并修复工作成果。 Aioga 将其归入「技巧观点」方向，重点关注它对真实使用和行业竞争的影响。
扫码直达文章详情