1 00:00:00,320 --> 00:00:03,180 Righto, grab a coffee, because this one's a quick win. 2 00:00:03,960 --> 00:00:08,380 This is The Claude Code Changelog, brought to you by Jellypod AI. 3 00:00:09,500 --> 00:00:15,260 Anthropic shipped 2.1.316 on October 31st, right after 4 00:00:15,400 --> 00:00:22,120 2.1.314 on the 30th, and the one I'd change my setup for today is a new environment 5 00:00:22,180 --> 00:00:26,840 variable: CLAUDE_CODE_SUBAGENT_EFFORT. 6 00:00:27,702 --> 00:00:28,602 Here's the backstory. 7 00:00:29,542 --> 00:00:34,542 Back in 2.1.292 we got an effort parameter for subagents. 8 00:00:35,302 --> 00:00:40,102 Before that, and still today when nothing names an effort, a subagent inherited the 9 00:00:40,162 --> 00:00:41,722 effort of the session that spawned it. 10 00:00:42,522 --> 00:00:46,842 So if you bump your main session to high or xhigh for a nasty bug, 11 00:00:47,322 --> 00:00:50,462 every little helper you delegate to runs at that same depth. 12 00:00:51,062 --> 00:00:54,142 Searching files, drafting a test, running a check. 13 00:00:54,762 --> 00:00:56,982 All of them thinking as hard as the main thread. 14 00:00:58,115 --> 00:01:01,155 That's like hiring a brain surgeon to carry the groceries in. 15 00:01:01,895 --> 00:01:03,735 Lovely work, but a bit of overkill. 16 00:01:04,595 --> 00:01:10,355 And remember 2.1.311, where the ultrathink keyword bumps that turn's effort to 17 00:01:10,595 --> 00:01:10,975 xhigh. 18 00:01:11,735 --> 00:01:15,495 Pair that with inheritance and one keyword can ripple out to every worker 19 00:01:15,555 --> 00:01:16,035 underneath. 20 00:01:17,975 --> 00:01:22,115 Right, and the release note for 2.1.316 is pretty plain. 21 00:01:22,775 --> 00:01:27,475 CLAUDE_CODE_SUBAGENT_EFFORT sets the effort level subagents run at when nothing else 22 00:01:27,535 --> 00:01:29,835 names one, instead of inheriting the session's. 23 00:01:30,295 --> 00:01:34,275 So you can keep your interactive session at xhigh while background workers default 24 00:01:34,315 --> 00:01:35,775 to something cheaper and faster. 25 00:01:36,342 --> 00:01:37,542 Setup is two lines of effort. 26 00:01:37,902 --> 00:01:41,262 In your shell, export CLAUDE_CODE_SUBAGENT_EFFORT equals low. 27 00:01:41,662 --> 00:01:43,782 You can also use medium, high, or xhigh. 28 00:01:44,282 --> 00:01:48,142 Or make it stick per project or per user by putting it in settings.json, 29 00:01:48,442 --> 00:01:52,162 either in your home claude folder or in the dot claude folder inside the repo. 30 00:01:52,642 --> 00:01:56,202 Under the env key, you write CLAUDE_CODE_SUBAGENT_EFFORT as the key, 31 00:01:56,362 --> 00:01:57,602 and low as the value. 32 00:01:58,237 --> 00:02:00,637 The order matters here, so let me spell it out. 33 00:02:01,357 --> 00:02:07,197 First, if the subagent's own definition names an effort, in its YAML frontmatter or 34 00:02:07,257 --> 00:02:09,437 in the prompt that spawns it, that wins. 35 00:02:10,277 --> 00:02:13,537 Second, if that's silent, your new env var applies. 36 00:02:14,317 --> 00:02:19,037 Third, if neither is set, you get the old behaviour, and it inherits the parent's 37 00:02:19,137 --> 00:02:19,877 current effort. 38 00:02:20,417 --> 00:02:22,977 So nothing changes for you until you opt in. 39 00:02:23,693 --> 00:02:27,333 A practical pattern: set the variable to low or medium globally. 40 00:02:27,733 --> 00:02:30,853 Then, for the one reviewer agent where depth really pays off, 41 00:02:31,073 --> 00:02:35,933 say a security reviewer or an architecture critic, put an explicit effort of high in 42 00:02:35,973 --> 00:02:36,573 its frontmatter. 43 00:02:37,133 --> 00:02:40,413 Everything else stays cheap by default, and the agent that earns it keeps the deep 44 00:02:40,473 --> 00:02:40,853 reasoning. 45 00:02:41,513 --> 00:02:42,133 That's the whole trick. 46 00:02:42,453 --> 00:02:44,213 Default low, override where it counts. 47 00:02:44,750 --> 00:02:45,730 Two caveats. 48 00:02:46,310 --> 00:02:47,750 One, it's only a default. 49 00:02:48,250 --> 00:02:53,330 Explicit frontmatter still beats it, so if some agent in your repo says high, 50 00:02:53,710 --> 00:02:55,430 the variable won't tone it down. 51 00:02:56,310 --> 00:02:59,030 Two, it's about effort, not model choice. 52 00:02:59,470 --> 00:03:03,310 It doesn't replace CLAUDE_CODE_WORKFLOW_SUBAGENT_MODEL or 53 00:03:03,470 --> 00:03:05,730 CLAUDE_CODE_SUBAGENT_MODEL_FORCE. 54 00:03:06,250 --> 00:03:08,010 Those pick which model runs. 55 00:03:08,470 --> 00:03:10,270 This picks how hard it thinks. 56 00:03:10,993 --> 00:03:16,533 Also in 2.1.316, and related: CLAUDE_CODE_DISABLE_ULTRATHINK. 57 00:03:17,073 --> 00:03:19,873 Set it to 1 and the ultrathink keyword is turned off. 58 00:03:20,413 --> 00:03:25,133 That's handy when you want effort pinned, like a cost capped CI job or a shared 59 00:03:25,233 --> 00:03:29,333 workspace where you don't want a stray word in a prompt spiking a turn to xhigh. 60 00:03:30,053 --> 00:03:34,833 And in VS Code, ultrathink now runs its turn at xhigh and shows in rainbow in the 61 00:03:34,873 --> 00:03:36,373 chat box, same as the terminal. 62 00:03:37,252 --> 00:03:38,372 Now the smaller bits. 63 00:03:39,072 --> 00:03:45,812 In 2.1.314, there's a new setting, permissions.blockWritesOutsideWorkingDirectories. 64 00:03:46,432 --> 00:03:52,152 It makes Write, Edit, and NotebookEdit refuse paths outside your declared working 65 00:03:52,192 --> 00:03:57,452 directories in every permission mode, even bypassPermissions or dontAsk. 66 00:03:58,052 --> 00:04:03,032 But mind the catch: it covers the file tools only, not raw Bash commands. 67 00:04:03,867 --> 00:04:10,747 Also 2.1.314: in VS Code and Desktop, sessions over 200K tokens that sit idle 68 00:04:10,787 --> 00:04:15,067 get summarized, and that summary is used on your next message if the prompt cache 69 00:04:15,107 --> 00:04:15,767 has expired. 70 00:04:16,447 --> 00:04:22,187 And 2.1.316 fixes shell commands being left running on macOS and Linux when Claude 71 00:04:22,266 --> 00:04:25,387 Code exits on its own, like when credentials are rejected. 72 00:04:26,067 --> 00:04:30,407 It also fixes bash aliases with line breaks, and functions starting with a dash, 73 00:04:30,507 --> 00:04:31,967 getting garbled in the Bash tool. 74 00:04:32,600 --> 00:04:33,900 One date to put in your calendar. 75 00:04:34,400 --> 00:04:39,820 Fast mode for Opus 4.8 is deprecated as of October 30, 2026, 76 00:04:40,340 --> 00:04:42,500 and it's removed on November 30th. 77 00:04:43,100 --> 00:04:46,940 After that, fast mode requests on Opus 4.8 return an error. 78 00:04:47,400 --> 00:04:50,520 They do not fall back to standard Opus 4.8. 79 00:04:51,040 --> 00:04:54,060 To keep the speedup, migrate to Opus 5.5. 80 00:04:54,940 --> 00:04:56,300 So your homework is small. 81 00:04:56,720 --> 00:05:02,980 Add CLAUDE_CODE_SUBAGENT_EFFORT to your settings, name the agents that deserve more, 82 00:05:03,340 --> 00:05:07,440 and check whether anything of yours still leans on fast mode with Opus 4.8. 83 00:05:07,920 --> 00:05:09,180 That's the changelog for today.