1 00:00:00,198 --> 00:00:03,440 Alright, imagine running a command, walking away to grab a coffee, 2 00:00:03,480 --> 00:00:07,360 and coming back to find your AI assistant didn't just write the code--it actually 3 00:00:07,424 --> 00:00:12,440 spun up a completely isolated workspace, ran the tests, committed the changes, 4 00:00:12,480 --> 00:00:16,560 pushed them to your repo, and opened a clean draft pull request for you. 5 00:00:16,640 --> 00:00:23,200 That is exactly what is happening in the brand new Claude Code v2.1.198 release, 6 00:00:23,220 --> 00:00:25,280 brought to you by Jellypod AI. 7 00:00:25,440 --> 00:00:27,520 No more baby-sitting the terminal. 8 00:00:28,023 --> 00:00:29,455 Oh, fair dinkum, James. 9 00:00:29,495 --> 00:00:33,815 That is absolutely massive for anyone sick of watching the terminal spinner like a 10 00:00:33,895 --> 00:00:34,295 hawk. 11 00:00:34,535 --> 00:00:39,575 I mean, we've had background agents before, but usually, they just stop. 12 00:00:39,687 --> 00:00:43,415 They get to the end of the work, and then they sit there, twiddling their virtual 13 00:00:43,484 --> 00:00:47,655 thumbs, waiting for you to type yes or hit enter to approve the push. 14 00:00:47,915 --> 00:00:49,015 Now? 15 00:00:49,122 --> 00:00:50,855 It's completely hands-off. 16 00:00:51,697 --> 00:00:51,857 Exactly. 17 00:00:51,997 --> 00:00:54,417 Let's break down how this actually works under the hood. 18 00:00:54,529 --> 00:00:59,377 When you run claude agents now, the CLI spawns a parallel background run. 19 00:00:59,497 --> 00:01:03,457 But instead of messing with your active working directory--which would be a total 20 00:01:03,505 --> 00:01:07,377 nightmare if you're trying to write code at the same time--it leverages a git 21 00:01:07,441 --> 00:01:08,097 worktree. 22 00:01:08,183 --> 00:01:12,977 Specifically, it isolates everything under .claude/worktrees/. 23 00:01:13,057 --> 00:01:17,057 So, you can keep hacking away in your main directory, while Claude is in its own 24 00:01:17,157 --> 00:01:20,017 sandbox, totally free from code stomping. 25 00:01:20,676 --> 00:01:25,268 Yeah, because there is nothing worse than an AI overwriting the file you literally 26 00:01:25,300 --> 00:01:27,108 just spent twenty minutes refactoring. 27 00:01:27,348 --> 00:01:30,388 That git worktree trick is brilliant. 28 00:01:30,448 --> 00:01:32,628 But, uh, what happens when it's done? 29 00:01:32,788 --> 00:01:35,108 You mentioned it opens a draft PR? 30 00:01:35,536 --> 00:01:39,963 Right, so once the agent finishes its task, it automatically commits, 31 00:01:40,009 --> 00:01:42,603 pushes to your remote, and raises a draft PR. 32 00:01:42,743 --> 00:01:45,643 And you don't have to guess when that happens because they've added system 33 00:01:45,692 --> 00:01:46,763 notification hooks. 34 00:01:46,863 --> 00:01:51,003 You get agent_needs_input if Claude gets stuck or blocked on something, 35 00:01:51,043 --> 00:01:53,843 and agent_completed when the draft PR is ready. 36 00:01:53,883 --> 00:01:58,243 Plus, if you're using Focus Mode on macOS or Windows, these notifications are 37 00:01:58,283 --> 00:02:02,923 beautifully grouped into a single, clean count so your desktop doesn't look like a 38 00:02:02,923 --> 00:02:02,923 spam inbox. 39 00:02:03,455 --> 00:02:04,282 Right, right. 40 00:02:04,422 --> 00:02:06,042 But wait a second, James. 41 00:02:06,309 --> 00:02:10,682 If this thing is running headless in the background, how does it handle credentials? 42 00:02:10,842 --> 00:02:15,242 Like, if my local terminal relies on custom environment variables, 43 00:02:15,274 --> 00:02:17,482 does the background supervisor inherit those? 44 00:02:18,258 --> 00:02:20,565 Ah, that is the major gotcha with this release. 45 00:02:20,685 --> 00:02:26,005 The background supervisor process does *not* inherit gateway environment variables, 46 00:02:26,053 --> 00:02:31,605 like ANTHROPIC_BASE_URL or special proxy settings, directly from your active shell. 47 00:02:31,745 --> 00:02:37,525 You absolutely must define them inside .claude/settings.json in an explicit env 48 00:02:37,578 --> 00:02:37,925 block. 49 00:02:38,045 --> 00:02:41,365 And because it's running headless, you've got to make sure your git remote 50 00:02:41,418 --> 00:02:46,885 credentials--whether you're using SSH keys or HTTPS helper tools--are fully 51 00:02:46,936 --> 00:02:51,045 configured and accessible globally, or the push will fail silently. 52 00:02:51,832 --> 00:02:53,027 Ah, that makes sense. 53 00:02:53,203 --> 00:02:56,627 Flat out like a lizard drinking if you forget that setting, 54 00:02:56,654 --> 00:02:56,787 eh? 55 00:02:56,967 --> 00:03:00,627 You'll be wondering why your headless agent is throwing auth errors. 56 00:03:00,807 --> 00:03:04,787 But speaking of authentication, they've actually cleaned up another massive pain 57 00:03:04,840 --> 00:03:05,907 point in this version. 58 00:03:06,047 --> 00:03:10,547 You know that annoying loop where your AWS STS token expires, 59 00:03:10,567 --> 00:03:14,627 and Claude suddenly screams "Please run slash login" mid-task? 60 00:03:15,352 --> 00:03:16,525 Oh, it's the worst. 61 00:03:16,685 --> 00:03:18,365 It completely kills the flow. 62 00:03:18,786 --> 00:03:19,692 Well, they fixed it! 63 00:03:19,832 --> 00:03:25,212 Now, awsAuthRefresh runs completely automatically the second those STS tokens 64 00:03:25,281 --> 00:03:26,012 expire. 65 00:03:26,092 --> 00:03:28,732 It just silently refreshes behind the scenes. 66 00:03:28,839 --> 00:03:33,372 On top of that, the built-in Explore agent now actually inherits your main session's 67 00:03:33,425 --> 00:03:37,852 model--though it's capped at Opus--instead of just defaulting to Haiku every single 68 00:03:37,916 --> 00:03:38,572 time. 69 00:03:38,652 --> 00:03:42,812 And if you have extended thinking configs turned on, those subagents inherit them 70 00:03:42,892 --> 00:03:43,452 too. 71 00:03:44,318 --> 00:03:44,473 Yes! 72 00:03:44,493 --> 00:03:48,473 And they finally added automatic exponential backoff for transient mid-response 73 00:03:48,530 --> 00:03:50,233 errors, like ECONNRESET. 74 00:03:50,323 --> 00:03:54,473 Instead of just crashing your entire multi-step agent run because of a one-second 75 00:03:54,523 --> 00:03:58,793 network blip, Claude Code will now pause, retry, and keep going. 76 00:03:58,926 --> 00:04:02,713 It makes the entire background automation loop incredibly robust. 77 00:04:02,926 --> 00:04:09,593 If you haven't run npm i -g @anthropic-ai/claude-code to grab 78 00:04:09,793 --> 00:04:15,753 v2.1.198 yet, do it today and let those background agents do the heavy lifting.