1
00:00:00,000 --> 00:00:03,280
So, I was updating the CSS on a client's
login page last night,

2
00:00:03,320 --> 00:00:07,440
right in my local environment, and instead
of squinting at those external contrast

3
00:00:07,497 --> 00:00:10,720
calculators, I just ran the new /dataviz
skill.

4
00:00:10,880 --> 00:00:16,480
This is the new built-in tool that dropped
in Claude Code v2.1.198,

5
00:00:16,640 --> 00:00:22,320
and mate, it programmatically validates
your color palettes against the WCAG 2.1

6
00:00:22,364 --> 00:00:24,400
compliance standards right in the
terminal.

7
00:00:24,693 --> 00:00:29,760
No more manual copying of hex codes into
web forms, it just scans the local files

8
00:00:29,780 --> 00:00:32,800
and flags the contrast ratios that'll fail
accessibility.

9
00:00:33,240 --> 00:00:38,480
Brought to you by Jellypod AI, this tool
is honestly a massive lifesaver for digital

10
00:00:38,528 --> 00:00:41,120
designers who just want to keep their
heads down in the code.

11
00:00:41,609 --> 00:00:42,927
That is wild!

12
00:00:43,114 --> 00:00:48,287
So /dataviz actually parses your local
theme files, calculates the precise luminance

13
00:00:48,349 --> 00:00:53,887
contrast ratio-like the standard 4.5-to-1
ratio for body text-and highlights the

14
00:00:53,940 --> 00:00:55,287
failures directly?

15
00:00:55,359 --> 00:00:58,127
That saves so much back-and-forth context
switching.

16
00:00:58,418 --> 00:00:59,218
Exactly!

17
00:00:59,245 --> 00:01:03,138
It-it-it literally just spits out a neat
little table of your foreground and

18
00:01:03,174 --> 00:01:04,258
background pairings.

19
00:01:04,391 --> 00:01:08,738
If your brand's neon green on white is
sitting at a terrible 2.1-to-1,

20
00:01:08,818 --> 00:01:11,538
it tells you right there before you even
commit the code.

21
00:01:11,658 --> 00:01:13,698
But wait-[excited] it gets even better.

22
00:01:13,858 --> 00:01:20,578
If you update to the absolute freshest
version, v2.1.199, which literally just

23
00:01:20,635 --> 00:01:24,098
landed, they unlocked stacked
slash-skills.

24
00:01:24,198 --> 00:01:27,858
You can now chain up to five custom or
built-in tools together.

25
00:01:27,951 --> 00:01:34,338
Think about typing /lint /test and then
your custom build skill all in one single,

26
00:01:34,378 --> 00:01:35,538
beautiful command line.

27
00:01:35,638 --> 00:01:37,858
The tool execution is fully pipelined.

28
00:01:38,435 --> 00:01:39,038
Oh, wow.

29
00:01:39,278 --> 00:01:44,318
Chaining up to five slash-commands in a
single run is massive for local automation.

30
00:01:44,665 --> 00:01:48,318
So, instead of running /lint, waiting for
Claude to finish,

31
00:01:48,338 --> 00:01:52,638
and then manually typing /test to verify
your changes, you can just stack them.

32
00:01:52,838 --> 00:01:55,598
But wait, what-what-what happens if the
linter fails?

33
00:01:55,694 --> 00:01:59,358
Does the stack halt, or does it keep
executing downstream tools?

34
00:01:59,648 --> 00:02:03,168
Ah, well, [clears throat] it depends on
how you configure the execution,

35
00:02:03,208 --> 00:02:07,488
but generally, the pipeline is smart
enough to stop if a critical test breaks the

36
00:02:07,555 --> 00:02:07,968
chain.

37
00:02:08,088 --> 00:02:14,208
And speaking of things breaking, v2.1.199
also introduced some serious resilience

38
00:02:14,251 --> 00:02:15,568
improvements under the hood.

39
00:02:15,688 --> 00:02:19,568
You know when you're on a patchy Wi-Fi
connection, or maybe behind a corporate

40
00:02:19,661 --> 00:02:24,608
proxy, and your terminal tool just goes
into an infinite loop of TLS retries,

41
00:02:24,658 --> 00:02:28,528
burning your API tokens and rate limits
while doing absolutely nothing?

42
00:02:29,482 --> 00:02:32,997
Oh, the TLS retry burn is the absolute
worst.

43
00:02:33,117 --> 00:02:36,837
You just watch your terminal hang for
three minutes while it secretly drains your

44
00:02:36,890 --> 00:02:38,037
quota behind the scenes.

45
00:02:39,134 --> 00:02:40,190
Yeah, it's a shocker.

46
00:02:40,366 --> 00:02:44,990
Well, now Claude Code stops that TLS retry
burn immediately.

47
00:02:45,150 --> 00:02:50,190
If it detects a handshake or proxy issue,
it halts the loop and spits out instant,

48
00:02:50,236 --> 00:02:54,830
highly specific environment-variable
diagnostics to tell you exactly which cert or

49
00:02:54,910 --> 00:02:56,910
proxy config is blocking the pipe.

50
00:02:57,086 --> 00:03:01,550
Plus, if the network does drop
mid-response, they've added a mechanism that

51
00:03:01,598 --> 00:03:03,790
preserves your partial code streaming.

52
00:03:04,003 --> 00:03:08,670
So, if Claude was halfway through writing
a 200-line React component and your

53
00:03:08,706 --> 00:03:12,030
connection blips, you don't lose the whole
stream and have to start over from

54
00:03:12,100 --> 00:03:12,830
scratch.

55
00:03:13,608 --> 00:03:15,608
That is incredibly practical.

56
00:03:15,784 --> 00:03:19,928
Preserving that partial stream instead of
throwing a generic network error and

57
00:03:19,992 --> 00:03:23,368
discarding the buffer is a huge
quality-of-life update.

58
00:03:23,581 --> 00:03:27,688
It sounds like they're really focusing on
making this tool resilient for actual,

59
00:03:27,736 --> 00:03:29,528
real-world developer environments.

60
00:03:29,828 --> 00:03:30,468
Too right.

61
00:03:30,628 --> 00:03:32,948
It's all about keeping you in the flow
state, mate.

62
00:03:33,062 --> 00:03:35,589
Anyway, those are the big updates for this
week.

63
00:03:35,695 --> 00:03:36,788
Catch you later!

64
00:03:37,090 --> 00:03:38,210
Sounds good, talk soon!