Transcription
[music] Hi, welcome to another video.
So, Boris Churnney shared a thread on X about his favorite hidden and underutilized Claude code features. And this is pretty interesting because Boris has built Claude code. So, this is not random theory. This is basically a list of the things he actually uses in real life.
Now, I'm saying probably underutilized because Boris framed them that way. But Anthropic has not published a full public usage ranking for all of these features. So, take that part with a grain of salt. But if someone that close to the product is using these features all the time, then I think it is fair to say most people are not using Claude code to its full potential. So, let's get right into it.
The first thing is that Claude code is not just a terminal app. Boris says he writes a lot of code from the Claude mobile app on iOS and you can also use it on Android which is pretty amazing to be honest because a lot of people still think of Claude code as something that only lives inside one terminal window on a laptop and once you understand that the second hidden feature makes way more sense. You can actually move sessions between mobile web desktop and terminal. Boris called out dash-teleport or slashtleport to continue a cloud session on your machine and slash remote control to control a locally running session from your phone or the web. So this is kind of great because you can start somewhere convenient and continue somewhere powerful.
Now the next two are honestly some of the biggest ones in the whole thread and these are sloop and slash schedule. Boris says he uses these to automate work on an interval like babysitting pull requests, rebasing, collecting Slack feedback, sweeping missed review comments, and pruning stale PRs. This is where Claude Code stops feeling like a chat tool and starts feeling more like an automated co-worker. And the important idea here is that you can turn repeatable workflows into skills plus loops. So instead of manually checking the same stuff every 30 minutes, you just let Claude keep doing it. This is pretty amazing for sure, especially if you're paying for Claude code already and only using it in one shot mode.
Now, let's talk about hooks because this is another one that most people do not use enough. Boris says you can use hooks to deterministically run logic at different points in the agent life cycle. So, for example, you can autoload context at session start, log every bash command before a tool runs, route permission prompts to somewhere else for approval, or even poke cla to continue when it stops, which is literally giving Claude code programmable behavior around the edges. And that is super powerful.
Now, this next one is slightly adjacent, but it is still very interesting. Boris also mentioned dispatching co-work which he says the most important tip for using Claude code is to give Claude a way to verify its own output. And for front-end work, he personally uses the Chrome extension because it tends to work more reliably than some other similar setups. That makes a lot of sense because if an AI cannot see what it built, then it is basically guessing. Along the same lines, he also points out that the Claw desktop app can automatically start web servers and test them in a built-in browser. So for web work, this is pretty awesome. You write code, launch the app, let Claude inspect the result, and then iterate until things actually look good instead of just compiling.
Now, another hidden feature a lot of people do not think about is session forking. Boris says, "If you want to fork an existing session, you can use slash branch inside the session or from the command line, you can resume a session and add dash-fork session." This is really good because sometimes the AI is going in a useful direction, but you also want to test another path without destroying your original context. And while the agent is working, Boris also likes to use /btw for side queries. This is such a good quality of life feature. You can ask a quick question, get a quick answer, and not pollute the main line of work, which is also good because a lot of AI sessions get worse when you keep mixing unrelated thoughts into the same thread.
Now, let's come to parallel work because this is where things get really interesting. Boris specifically calls out git work trees and this is a really good option for sure if you do lots of parallel work in the same repo. You can have multiple claudids running in different work trees, each isolated, each working on a separate problem and they are not stepping on each other's toes. And if that is not enough, he also mentions/batch to fan out massive change sets. The way he describes it, Claude basically interviews you first and then fans the work out to as many workree agents as it takes. So for large migrations or repetitive codebasewide changes, this is pretty amazing to be honest.
Now for scripted or SDKdriven usage, Boris also calls out the dash-bear flag. By default, Claude looks for local.Claude files, settings, and MCPS when it starts. But for non-interactive usage, d-bear can speed things up a lot by skipping some of that automatic loading, which is pretty great if you're doing programmatic runs and care about startup overhead. He also mentions d--add dear which lets claude see and access more than one folder. This is really good for multi-reo work because a lot of real projects are not sitting inside one neat little repository anymore. So instead of constantly switching context, you can let one claude session understand the broader setup. And then there is d- agent which lets you give clawed code a custom system prompt and tool set through custom agents in theclaw agents folder. I like this a lot because it means you can create specialized agents for readonly analysis, migrations, testing, documentation, or whatever else you want. This is one of those features that looks small at first, but it actually changes how structured your workflows can become.
Finally, Boris mentions /Voice. And this one is honestly super underrated because a lot of people still think serious coding has to be done only by typing, but Boris says he does a lot of his coding by speaking to Claude code like a single terminal chatbot. While power users are using it like a whole operating environment. Mobile, web, desktop, remote control, hooks, loops, branching, work trees, verification, custom agents, voice, all of it works together. And that is why this thread matters. It is not just a bunch of hidden tricks. It is basically a map of how someone deeply involved with claude code actually uses the product daytoday. So if you are paying for claude code and you're only using simple prompts in one terminal, you are probably leaving a ton of value on the table. Overall, it's pretty cool.
Anyway, let me know your thoughts in the comments. If you like this video, consider donating through the super thanks option or becoming a member by clicking the join button. Also [music] give this video a thumbs up and subscribe to my channel. I'll see you in the next one. Until then, bye. [music] >> [music]