1 00:00:00,380 --> 00:00:05,140 Right, so picture Claude Code running a build, and instead of waiting for it, 2 00:00:05,560 --> 00:00:10,400 it just types sleep sixty, checks, types sleep sixty again. 3 00:00:10,960 --> 00:00:14,640 That's a bit like standing at the oven door asking the roast if it's done yet. 4 00:00:15,560 --> 00:00:20,100 This is The Claude Code Changelog, brought to you by Jellypod AI, 5 00:00:20,640 --> 00:00:27,560 and today it's version 2.1.310, released October 23rd, and the fix for exactly 6 00:00:27,600 --> 00:00:28,160 that habit. 7 00:00:29,218 --> 00:00:30,038 Here's what changed. 8 00:00:30,538 --> 00:00:34,738 The Monitor tool, and plugin monitors, are now enabled on Amazon Bedrock, 9 00:00:34,918 --> 00:00:37,118 Google Vertex, and Microsoft Foundry. 10 00:00:37,618 --> 00:00:41,238 They also now work when telemetry or nonessential traffic is disabled, 11 00:00:41,618 --> 00:00:44,978 so that's DISABLE_TELEMETRY set to 1, or 12 00:00:45,138 --> 00:00:49,578 CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC set to 1. 13 00:00:50,258 --> 00:00:53,798 Per the changelog, long foreground sleeps are now blocked in those same environments 14 00:00:53,858 --> 00:00:54,078 too. 15 00:00:54,258 --> 00:00:57,238 So Claude can't fall back on a long sleep in Bash there anymore. 16 00:00:57,923 --> 00:01:00,463 And that lines up with last episode's story, James. 17 00:01:00,943 --> 00:01:07,623 In 2.1.309 we talked about features that simply weren't there on Bedrock and 18 00:01:07,683 --> 00:01:10,123 Vertex, ultrareview being the example. 19 00:01:10,903 --> 00:01:12,903 Enterprise setups kept getting the leftovers. 20 00:01:13,403 --> 00:01:14,903 This one closes a real gap. 21 00:01:15,550 --> 00:01:16,010 It does. 22 00:01:16,390 --> 00:01:21,790 Monitor has been around since 2.1.98, but on third party clouds it was restricted, 23 00:01:22,190 --> 00:01:24,350 and it broke if those privacy flags were set. 24 00:01:25,010 --> 00:01:29,790 So in exactly the locked down environments, Claude fell back to crude loops to watch 25 00:01:29,850 --> 00:01:33,030 dev servers, builds, test runners, and log files. 26 00:01:33,730 --> 00:01:34,530 Two costs there. 27 00:01:35,190 --> 00:01:39,210 Every check is another Bash call, so you burn context on the same question over and 28 00:01:39,290 --> 00:01:39,570 over. 29 00:01:40,090 --> 00:01:44,390 And a long foreground sleep freezes your turn while nothing useful happens. 30 00:01:45,028 --> 00:01:47,508 So what does Monitor actually do instead? 31 00:01:48,188 --> 00:01:53,568 According to the tools reference, it runs a command in the background and feeds each 32 00:01:53,668 --> 00:01:57,688 output line back to Claude, so it can react to log entries, 33 00:01:57,888 --> 00:02:02,028 file changes, or polled status in the middle of the conversation. 34 00:02:02,788 --> 00:02:07,048 It can also open a WebSocket and treat each incoming message as an event. 35 00:02:07,708 --> 00:02:09,688 The point is that it's interrupt driven. 36 00:02:10,128 --> 00:02:15,168 Claude only wakes up, and only spends tokens, when something actually shows up. 37 00:02:15,923 --> 00:02:18,783 Using it is mostly a matter of how you phrase the prompt. 38 00:02:19,263 --> 00:02:21,143 Describe the event, not the wait. 39 00:02:21,723 --> 00:02:27,443 For example: run npm run dev and monitor localhost:3000 until the server is ready, 40 00:02:27,783 --> 00:02:28,943 then run the smoke tests. 41 00:02:29,603 --> 00:02:34,923 Or: tail server.log in the background and tell me the moment a 5xx error shows up. 42 00:02:35,643 --> 00:02:39,643 Or: run the test suite in the background and alert me if anything fails. 43 00:02:40,283 --> 00:02:43,563 Each of those names a condition, and that's what Monitor is built around. 44 00:02:44,177 --> 00:02:49,997 If you use the self pacing slash loop command, that one leans on the Monitor tool to 45 00:02:50,057 --> 00:02:54,117 wait for status changes without polling, so it benefits from this as well. 46 00:02:54,957 --> 00:02:55,857 Now, permissions. 47 00:02:56,557 --> 00:03:01,737 Monitor is marked as requiring permission in manual mode, so the first time Claude 48 00:03:01,877 --> 00:03:06,237 spawns a watcher, you'll get a prompt showing the exact command it's about to run in 49 00:03:06,277 --> 00:03:06,897 the background. 50 00:03:07,517 --> 00:03:10,337 In auto mode, the classifier makes that call. 51 00:03:11,277 --> 00:03:14,837 And for teams, you can allow or deny it in your settings. 52 00:03:15,317 --> 00:03:21,837 Put the tool name Monitor in permissions.allow or permissions.deny in 53 00:03:21,857 --> 00:03:24,357 .claude/settings.json. 54 00:03:25,057 --> 00:03:30,477 The docs also list Bash command patterns, like Bash with npm run star, 55 00:03:30,917 --> 00:03:33,157 as applying to both Bash and Monitor. 56 00:03:33,942 --> 00:03:34,782 Now the caveats. 57 00:03:35,282 --> 00:03:39,742 First, anything you wrote that says sleep thirty seconds and check again is going to 58 00:03:39,782 --> 00:03:41,162 hit a wall in these environments. 59 00:03:41,682 --> 00:03:43,902 That includes custom skills and saved prompts. 60 00:03:44,482 --> 00:03:48,302 Rewrite them to name the event: watch the build log until it completes. 61 00:03:49,042 --> 00:03:51,682 Second, a monitor is a real background process. 62 00:03:52,202 --> 00:03:54,922 A forgotten one keeps consuming memory or CPU. 63 00:03:55,522 --> 00:03:59,122 Use slash tasks to see what's running, and the TaskStop tool, 64 00:03:59,342 --> 00:04:01,482 or slash tasks itself, to end it. 65 00:04:02,262 --> 00:04:05,222 Third, those permission prompts will show up in manual mode, 66 00:04:05,462 --> 00:04:08,022 so expect a few more approvals the first time through. 67 00:04:08,677 --> 00:04:11,877 Quick round of smaller things in 2.1.310. 68 00:04:12,517 --> 00:04:16,157 If you're behind a gateway, like LiteLLM or a corporate proxy, 69 00:04:16,557 --> 00:04:21,197 that rejects the effort beta value or mid conversation system messages, 70 00:04:21,677 --> 00:04:24,877 Claude Code now retries without them, instead of just failing. 71 00:04:25,637 --> 00:04:27,177 There's a compaction fix too. 72 00:04:27,797 --> 00:04:32,877 After a Prompt is too long error, compaction used to keep too much dense text, 73 00:04:33,257 --> 00:04:38,317 think base64 or SVG, so the very next request could fail again. 74 00:04:39,017 --> 00:04:39,517 That's fixed. 75 00:04:40,092 --> 00:04:44,972 Security wise, file read permission prompts now show where a symbolic link leads, 76 00:04:45,232 --> 00:04:47,372 so you can see the real target before you approve. 77 00:04:48,012 --> 00:04:52,512 On Linux, slash sandbox and the startup warning now name Ubuntu's AppArmor 78 00:04:52,552 --> 00:04:56,812 restriction on user namespaces when it stops bubblewrap from starting the sandbox, 79 00:04:57,292 --> 00:04:58,412 and they tell you how to lift it. 80 00:04:59,072 --> 00:05:02,912 Auto mode also now tells you when blocking reads outside your working directories 81 00:05:02,952 --> 00:05:04,972 will make shell commands ask for permission. 82 00:05:05,572 --> 00:05:09,752 And for admins on Enterprise, cloud sessions and Remote Control permissions can now 83 00:05:09,792 --> 00:05:13,652 be assigned through custom roles, so Owners can control each by role. 84 00:05:14,360 --> 00:05:20,500 So the takeaway: if you're on Bedrock, Vertex, Foundry, or running with telemetry 85 00:05:20,580 --> 00:05:27,220 off, update to 2.1.310, then go find any prompt of yours that says 86 00:05:27,480 --> 00:05:30,480 sleep, and swap it for the event you actually care about. 87 00:05:31,320 --> 00:05:31,940 Cheers for listening.