1 00:00:00,000 --> 00:00:00,800 G'day everyone! 2 00:00:01,156 --> 00:00:02,086 Welcome to the show. 3 00:00:02,286 --> 00:00:06,007 I'm Lachlan Reed, and as always, I'm joined by James Turner. 4 00:00:06,114 --> 00:00:10,327 Today we are diving straight into some serious terminal wizardry, 5 00:00:10,367 --> 00:00:15,767 but first, a massive shoutout to Jellypod for making this daily show a reality. 6 00:00:15,947 --> 00:00:22,247 Now, James, I was poking around the v2.1.169 release of Claude Code, 7 00:00:22,267 --> 00:00:26,487 and they have quietly handed us the keys to complete headless automation. 8 00:00:27,000 --> 00:00:29,320 Oh, the updates to `claude agents` are massive! 9 00:00:30,029 --> 00:00:34,120 Especially if you've ever tried to script Claude in a CI/CD pipeline or a background 10 00:00:34,200 --> 00:00:35,360 cron job. 11 00:00:35,423 --> 00:00:39,880 Before this release, running Claude as a background agent was a bit of a black box. 12 00:00:39,920 --> 00:00:44,840 But now, they've upgraded `claude agents --json` with two crucial additions: the 13 00:00:44,880 --> 00:00:47,800 `--all` flag and the `waitingFor` field. 14 00:00:48,000 --> 00:00:50,960 That `waitingFor` field is the real gold nugget here. 15 00:00:51,800 --> 00:00:55,120 See, previously, if your background agent hit a point where it needed user 16 00:00:55,164 --> 00:00:59,760 permission-like, say, running a sketchy terminal command or writing to a protected 17 00:00:59,808 --> 00:01:02,600 directory-the script would just hang. 18 00:01:02,710 --> 00:01:05,840 You'd be staring at a blank terminal like a stunned mullet. 19 00:01:06,100 --> 00:01:10,480 Now, the JSON output explicitly tells your orchestrator script, 20 00:01:10,600 --> 00:01:13,840 "Hey, I'm blocked because I'm waiting on a permission prompt." 21 00:01:14,000 --> 00:01:17,680 Right, it literally exposes the state machine of the agent. 22 00:01:17,840 --> 00:01:24,480 So, if you run `claude agents --json --all`, you get a structured list of every 23 00:01:24,549 --> 00:01:25,920 active background agent. 24 00:01:26,107 --> 00:01:31,120 If one of them is stuck, the `waitingFor` key in that JSON payload will populate 25 00:01:31,152 --> 00:01:33,840 with something like `USER_APPROVAL`. 26 00:01:33,872 --> 00:01:37,920 Your wrapper script can parse that and trigger an alert, or even programmatically 27 00:01:37,970 --> 00:01:40,080 approve it if you've built an auto-decide layer. 28 00:01:41,000 --> 00:01:41,960 Exactly! 29 00:01:42,665 --> 00:01:45,240 You'd just pipe that output straight into `jq`. 30 00:01:45,370 --> 00:01:52,040 Imagine a bash loop running every five seconds, doing `claude agents --json 31 00:01:52,413 --> 00:01:59,400 --all | jq '.[] | select(.status == "paused" 32 00:01:59,620 --> 00:02:05,507 and .waitingFor == "USER_APPROVAL")'`. 33 00:02:05,524 --> 00:02:10,680 If that returns a hit, you can fire off a Slack webhook or a PagerDuty alert. 34 00:02:10,947 --> 00:02:17,080 It turns Claude from an interactive CLI tool into a fully queryable system daemon. 35 00:02:18,000 --> 00:02:20,720 Which brings us to the security elephant in the room. 36 00:02:21,502 --> 00:02:24,640 If you're running these agents headlessly or letting them loose on a server, 37 00:02:24,800 --> 00:02:27,200 you do not want them running wild. 38 00:02:27,344 --> 00:02:31,600 That's why the introduction of the `CLAUDE_CODE_DISABLE_BUNDLED_SKILLS` environment 39 00:02:31,644 --> 00:02:36,880 variable-or the `disableBundledSkills` config setting-is a massive win for 40 00:02:36,931 --> 00:02:38,240 enterprise sandboxing. 41 00:02:39,000 --> 00:02:39,840 Spot on. 42 00:02:39,933 --> 00:02:43,080 It's all about restricting the blast radius. 43 00:02:43,112 --> 00:02:48,280 When you set `CLAUDE_CODE_DISABLE_BUNDLED_SKILLS=true`, you are essentially 44 00:02:48,344 --> 00:02:53,280 stripping Claude of its default slash commands and built-in workflows. 45 00:02:53,400 --> 00:02:59,320 No more `/search`, no more automatic git commits, and no more executing bundled bash 46 00:02:59,380 --> 00:03:01,800 scripts that you haven't explicitly whitelisted. 47 00:03:02,000 --> 00:03:04,080 It's the ultimate lock-down mode. 48 00:03:04,547 --> 00:03:09,040 If you disable those bundled skills, Claude is forced to rely solely on the explicit 49 00:03:09,107 --> 00:03:14,160 tools you expose to it via your own custom API schemas or runtime environments. 50 00:03:14,320 --> 00:03:18,480 It prevents the model from trying to be clever and using a default workflow that 51 00:03:18,507 --> 00:03:21,040 might bypass your custom security proxies. 52 00:03:22,000 --> 00:03:24,960 Yeah, but there is a massive caveat here, mate. 53 00:03:26,054 --> 00:03:30,721 If you strip all those default capabilities, you're also stripping its legs. 54 00:03:30,988 --> 00:03:35,761 If your automation scripts rely on Claude using its standard filesystem search or 55 00:03:35,841 --> 00:03:41,681 file editing skills to refactor code, those will fail silently or throw errors 56 00:03:41,761 --> 00:03:45,681 because the model suddenly doesn't know how to call those slash commands anymore. 57 00:03:45,861 --> 00:03:50,401 You've gotta be prepared to build your own custom tool definitions to replace what 58 00:03:50,441 --> 00:03:51,201 you took away. 59 00:03:52,000 --> 00:03:56,640 Exactly, you have to find that sweet spot between absolute security and actual 60 00:03:56,710 --> 00:03:57,201 utility. 61 00:03:57,827 --> 00:04:02,000 If you're going to sandbox it, you need to be highly intentional about the specific 62 00:04:02,057 --> 00:04:03,600 subset of tools you expose. 63 00:04:03,787 --> 00:04:08,480 It's a classic trade-off: headless automation gives you incredible leverage, 64 00:04:08,520 --> 00:04:11,200 but only if you design the guardrails properly. 65 00:04:12,000 --> 00:04:12,880 Well said, James. 66 00:04:13,067 --> 00:04:16,880 It's a whole new paradigm for local agent execution. 67 00:04:17,040 --> 00:04:22,080 Let us know if you've started running Claude Code in your deployment pipelines yet. 68 00:04:22,256 --> 00:04:24,960 That's all for today's quick tech teardown. 69 00:04:25,067 --> 00:04:26,239 Catch you all tomorrow! 70 00:04:27,000 --> 00:04:27,400 See ya!