like Claude Code can be utilized for all kinds of duties. The primary process you may consider is to really implement code.
Nevertheless, there are additionally so many different duties that you need to use them for. You should utilize them to investigate paperwork, navigate your pc, use the browser, carry out spreadsheet calculations, and so forth. My level is that coding brokers can principally be used to carry out all duties in your pc. The duty I’ll cowl at the moment is working end-to-end exams with Claude Code.
Finish-to-end exams are principally exams the place you make the coding agent open a browser and click on by it and take a look at its implementations utterly. In that method, an alternate testing scheme might be unit exams or integration exams.
end-to-end exams are nice as a result of they take a look at the appliance identical to how a consumer would use the appliance, which ensures that you’ll be able to correctly take a look at the appliance and uncover points that the consumer may expertise.
Why run end-to-end exams with Claude Code
To begin with, I’d wish to cowl why you must care about this subject. The explanation you must run end-to-end exams with Claude Code is solely that it’s an efficient method of testing your code implementations. After coding brokers grew to become extraordinarily highly effective, most notably after the discharge of Claude Opus 4.5, a brand new bottleneck that arose was testing the code that’s been applied.
Earlier than coding brokers, the bottleneck was principally producing the code as a result of it took an engineer a considerable amount of time to provide the code to both implement a bug repair or implement a brand new characteristic. Nevertheless, this has shifted now that we’ve got coding brokers as a result of they’re so efficient at implementing new code.
The brand new bottleneck now could be principally testing the code. And also you don’t wish to be manually testing all of the code that coding brokers implement, which is why you have to be utilizing end-to-end exams. That is principally making the coding agent absolutely take a look at its personal code earlier than you confirm the work.
Operating end-to-end exams is tremendous highly effective, and it principally provides the fashions a approach to confirm their very own work, which is extraordinarily highly effective and makes them extra prone to do one-shot implementations, which can prevent loads of time as an engineer.
Methods to run end-to-end exams with Claude Code
Now it’s time to debate easy methods to truly run end-to-end exams. The benefit of coding brokers is that you need to use them to additionally arrange end-to-end testing.
Briefly, all you’re gonna do to run end-to-end exams is principally to have a coding agent and provides that coding agent entry to the browser and in addition login entry to the instrument that you simply’re testing.
When you’ve accomplished this, you may merely ask the coding agent to carry out end-to-end exams earlier than contemplating a process is completed. This end-to-end testing is tremendous easy, however there are loads of methods to optimize end-to-end testing each to make it more practical and make the brokers higher at performing the end-to-end testing.
Thus, I’ll cowl some totally different strategies I exploit to run end-to-end testing with Claude Code that make my end-to-end testing more practical.
I’ll additionally talk about some totally different conditions the place you may wish to run end-to-end testing. For instance, you may wish to run it:
- After implementing a brand new characteristic
- after fixing a bug
- on a every day schedule to verify that every part within the utility is working as anticipated
My end-to-end testing immediate
To begin with, I’ll cowl the immediate I exploit to carry out end-to-end testing (and, basically, the immediate I exploit to make sure my coding brokers have accomplished their work efficiently).
Implement every part I requested for. Confirm it finish to finish by clicking by
the browser utilizing the Playwright MCP. It is not acceptable to check the
utility solely by integration exams. It's essential to truly click on
across the app. Proceed like this till it really works. Repair any points when you
encounter them then do an finish to finish take a look at once more. Run codex exec and run the
overview talent with codex and make him approve it and iterate till codex
has accredited it. When codex has accredited it, get this to dev, after which
take a look at finish to finish in dev once more. Proceed till its working fantastic in dev.That is the every day immediate I exploit to get stuff to dev. So primarily, I clarify a process to Claude Code. Then I run slash objective with the immediate that you simply see above.
To make it tremendous easy to put in writing out this full immediate each time, I exploit the FluidVoice transcription instrument, and I then say “full merge take a look at”. I then arrange phrase alternative in order that FluidVoice routinely swaps out the three phrases for the total immediate above. This makes it tremendous easy for me to continually immediate my brokers to run the total end-to-end take a look at.
Playwright MCP
To begin with, with Claude Code, they’ve a built-in Chrome MCP you need to use to present Claude entry to Chrome, and it will possibly thus entry the browser. Nevertheless, I personally desire the Playwright MCP as a result of I imagine that it makes the agent carry out higher when interacting with the browser. To put in the Playwright MCP, you may merely immediate Claude Code to put in it for you. Restart the Claude Code session, and it’ll have entry to the Playwright MCP, which you need to use to work together with the browser.
It’s additionally price noting that there are loads of instruments on the market that you need to use to work together with the browser, and that Playwright MCP is solely the instrument that I’m utilizing each day.
The /objective command
The slash objective command can be extraordinarily highly effective. Mainly, the slash objective command is a command that I’ve mentioned earlier than as nicely, however it’s a command the place you set a objective, and the agent continues till it achieves that objective. If the agent stops at any level, there’s an computerized hook that triggers, which prompts the agent to investigate if it achieved its objective. And if it achieved it, it stops, in fact, but when it didn’t obtain it, it continues till it achieves its objective. It’s principally an excellent approach to make the agent proceed working till it reaches a last goal.
Finish-to-end testing with /objective is extraordinarily highly effective. The reason being that you need to use the command to make sure the agent continues till efficiently implementing what you specify, however it additionally makes it a lot simpler for the agent to confirm its personal work and to make sure it truly achieved the objective.
Thus, combining end-to-end testing with the command is a really highly effective approach that you need to use to vastly improve the usefulness and productiveness of your coding brokers
Operating end-to-end testing with OpenClaw
The scenario I described above is after I run the total end-to-end testing after performing my very own implementations with Claude Code; nonetheless, you may additionally have loads of OpenClaw brokers working. This could additionally make use of the total end-to-end testing.
One particular factor you may arrange is to have an OpenClaw agent set off as soon as a day or a number of instances a day an end-to-end take a look at of your manufacturing utility. This testing ought to undergo the complete utility, use the totally different options, and ensure all of them work as anticipated, which can show you how to uncover loads of bugs.
Since you set off this on a schedule, you’ll discover loads of bugs earlier than customers uncover them, which is in fact one thing you must all the time attempt for.
You may as well set this up utilizing a cron job in your pc or the schedule performance in Claude Code. Nevertheless, I discovered that establishing an OpenClaw agent to carry out this end-to-end testing works very nicely. It could actually replace the testing scheme at any time when I replace the appliance and routinely repair the problems that the OpenClaw agent discovers.
Conclusion
On this article, I mentioned how one can run full end-to-end testing. That is principally a way the place you make a coding agent undergo the browser to check the implementations that it implements. Finish-to-end testing is extraordinarily highly effective as a result of it experiences the appliance identical to a consumer would, and it thus helps you uncover loads of bugs. I imagine end-to-end testing and, basically, verifiability is extraordinarily essential in relation to making essentially the most out of coding brokers, and it’s one thing you must all the time arrange in order that it’s easy to your coding agent to run full end-to-end exams. You also needs to set them up on a schedule so that you simply uncover points earlier than your customers do.
👉 My free eBook and Webinar:
🚀 10x Your Engineering with LLMs (Free 3-Day E mail Course)
📚 Get my free Imaginative and prescient Language Fashions book
💻 My webinar on Imaginative and prescient Language Fashions
👉 Discover me on socials:
💌 Substack
