1 00:00:00,120 --> 00:00:01,820 Brought to you by Jellypod AI. 2 00:00:02,160 --> 00:00:06,480 Have you ever actually opened your project's CLAUDE dot md file after six months of 3 00:00:06,560 --> 00:00:07,200 active building? 4 00:00:07,740 --> 00:00:11,620 It is like walking into an attic full of old holiday decorations you forgot you 5 00:00:11,700 --> 00:00:11,960 owned. 6 00:00:13,233 --> 00:00:14,173 Oh, spot on mate. 7 00:00:14,673 --> 00:00:17,613 You find all these bizarre prompt hacks from last year, right? 8 00:00:18,093 --> 00:00:22,213 Like, yelling in all caps, or telling the model to take a deep breath before 9 00:00:22,293 --> 00:00:22,693 answering. 10 00:00:23,253 --> 00:00:25,273 And you just keep piling them on top of each other. 11 00:00:25,333 --> 00:00:25,813 Right! 12 00:00:25,873 --> 00:00:29,733 But all that accumulated cruft actually degrades model attention. 13 00:00:29,893 --> 00:00:35,093 It eats up precious context tokens and confuses newer models that do not need those 14 00:00:35,153 --> 00:00:36,293 old tricks anyway. 15 00:00:36,386 --> 00:00:41,333 Which brings us to the brand new Claude Code version two point one point two hundred 16 00:00:41,390 --> 00:00:42,613 eighty three release. 17 00:00:42,845 --> 00:00:46,465 Yeah, Anthropic just dropped a really handy diagnostic tool in this release, 18 00:00:46,505 --> 00:00:46,845 didn't they? 19 00:00:47,385 --> 00:00:48,805 Slash doctor prompt audit. 20 00:00:49,405 --> 00:00:51,605 Or you can run slash checkup prompt audit. 21 00:00:52,033 --> 00:00:52,633 Exactly. 22 00:00:53,033 --> 00:00:54,813 It parses through your entire workspace. 23 00:00:55,393 --> 00:00:59,333 We are talking CLAUDE dot md files, custom skills, subagents, 24 00:00:59,773 --> 00:01:02,633 commands in dot claude slash rules, all of it. 25 00:01:03,233 --> 00:01:06,453 And it searches specifically for prompting patterns that were written for older 26 00:01:06,533 --> 00:01:06,913 models. 27 00:01:06,958 --> 00:01:11,678 Wait, so it, it, it actually flags stuff like broken file paths and dead CLI 28 00:01:11,722 --> 00:01:12,478 commands too? 29 00:01:12,500 --> 00:01:13,060 Yes! 30 00:01:13,273 --> 00:01:18,420 Stale paths, commands that do not exist anymore, and contradictory rules. 31 00:01:18,516 --> 00:01:23,220 Like if one instruction file tells Claude to always use typescript and another file 32 00:01:23,247 --> 00:01:26,260 in a subfolder says stick to plain JavaScript. 33 00:01:27,512 --> 00:01:28,472 That is brilliant. 34 00:01:28,832 --> 00:01:33,212 I can't tell you how many times I have chased my own tail because two instruction 35 00:01:33,312 --> 00:01:36,152 files were quietly arguing with each other in the background. 36 00:01:37,092 --> 00:01:40,412 So, um, how does the audit report actually present all this? 37 00:01:40,458 --> 00:01:45,498 When you execute slash doctor prompt audit, it prioritizes the big stuff right at 38 00:01:45,538 --> 00:01:46,298 the top. 39 00:01:46,378 --> 00:01:50,938 The report leads with stale paths, stale commands, and contradicting instruction 40 00:01:51,018 --> 00:01:51,578 files. 41 00:01:51,678 --> 00:01:56,378 But here is the smart part, it preserves documented thinking keywords that newer 42 00:01:56,418 --> 00:01:58,218 reasoning models still rely on. 43 00:01:58,250 --> 00:02:02,010 Ah, okay, so it is not just nuking keywords blindly. 44 00:02:02,170 --> 00:02:06,490 It gives you a clean remediation list without automatically deleting your files 45 00:02:06,530 --> 00:02:07,530 without permission. 46 00:02:07,542 --> 00:02:08,182 Precisely. 47 00:02:08,209 --> 00:02:13,462 It tells you what is outdated, why it is problematic, and leaves the actual cleanup 48 00:02:13,524 --> 00:02:15,222 decision in your hands. 49 00:02:15,450 --> 00:02:16,530 That is a relief. 50 00:02:17,110 --> 00:02:20,590 No rogue agent wiping out custom rules while you grab a coffee. 51 00:02:21,830 --> 00:02:25,610 Now, what else came in two point one point two hundred eighty three? 52 00:02:26,690 --> 00:02:31,010 I heard there were some nice enterprise and terminal quality of life updates. 53 00:02:31,125 --> 00:02:31,605 Yeah! 54 00:02:31,712 --> 00:02:36,165 If you run an enterprise LLM gateway, there is a new environment variable setting, 55 00:02:36,218 --> 00:02:38,805 CLAUDE CODE GATEWAY HINT HEADERS equals one. 56 00:02:38,949 --> 00:02:43,285 That injects an x claude code prompt id header into request hint headers. 57 00:02:43,292 --> 00:02:47,772 Wait, so the gateway can group all the underlying API requests that serve a single 58 00:02:47,836 --> 00:02:49,372 user prompt together? 59 00:02:49,375 --> 00:02:50,095 You got it. 60 00:02:50,202 --> 00:02:55,775 Makes tracking, logging, and cost attribution across multi turn tool calls way 61 00:02:55,832 --> 00:02:57,375 easier for infrastructure teams. 62 00:02:57,375 --> 00:02:59,935 Man, that is huge for platform engineers. 63 00:03:00,055 --> 00:03:04,415 And on the terminal UI side, they finally fixed list navigation! 64 00:03:04,957 --> 00:03:07,017 Oh man, the UI pickers! 65 00:03:07,477 --> 00:03:13,517 Commands like slash mcp, slash tasks, and slash rewind now support page keys, 66 00:03:13,897 --> 00:03:16,317 mouse wheel scrolling, and direct clicks. 67 00:03:16,723 --> 00:03:17,223 Finally! 68 00:03:17,603 --> 00:03:22,403 No more arrow keying through fifty tasks one by one like it is nineteen ninety five. 69 00:03:23,083 --> 00:03:27,303 A proper scroll wheel support in the terminal feels like magic. 70 00:03:27,417 --> 00:03:30,377 So if your prompt files have been feeling a bit bloated lately, 71 00:03:30,437 --> 00:03:35,297 run slash doctor prompt audit on your codebase today and clear out the cobwebs. 72 00:03:35,353 --> 00:03:36,457 Good chatting Lachlan! 73 00:03:36,458 --> 00:03:37,498 Yeah, good chat mate. 74 00:03:37,538 --> 00:03:37,898 See ya.