1
00:00:00,000 --> 00:00:01,919
Welcome to the show - James,

2
00:00:01,919 --> 00:00:16,318
I wanna start with three URLs that tell
the whole story: `https://gitlab.

3
00:00:01,919 --> 00:00:16,318
com/org/repo/-/merge_requests/99`,

4
00:00:16,318 --> 00:00:23,518
`https://bitbucket.

5
00:00:16,318 --> 00:00:23,518
org/org/repo/pull-requests/7`,

6
00:00:23,518 --> 00:00:35,454
and `https://github.

7
00:00:23,518 --> 00:00:35,454
internal.

8
00:00:23,518 --> 00:00:35,454
co/org/repo/pull/42`.

9
00:00:35,454 --> 00:00:40,479
Yeah, and that last one - `github.

10
00:00:35,454 --> 00:00:40,479
internal.

11
00:00:35,454 --> 00:00:40,479
co` - is the giveaway.

12
00:00:40,479 --> 00:00:42,597
That is not hobby-project land.

13
00:00:42,597 --> 00:00:45,314
That's corporate VPN, internal certs,

14
00:00:45,314 --> 00:00:50,840
weird SSO, somebody's security team
breathing down your neck.

15
00:00:45,314 --> 00:00:50,840
Exactly.

16
00:00:50,840 --> 00:00:58,692
And in Claude Code v2.

17
00:00:50,840 --> 00:00:58,692
1.119, `claude --from-pr` now accepts all
of those

18
00:00:58,692 --> 00:01:01,600
patterns, not just plain old `github.

19
00:00:58,692 --> 00:01:01,600
com`.

20
00:01:01,600 --> 00:01:05,992
So GitLab merge requests,
Bitbucket pull requests,

21
00:01:05,992 --> 00:01:08,872
and GitHub Enterprise PRs all work.

22
00:01:08,872 --> 00:01:10,794
Which sounds tiny...

23
00:01:10,794 --> 00:01:13,666
but mate, it's not tiny at all.The version number

24
00:01:13,666 --> 00:01:20,280
there - v2.

25
00:01:13,666 --> 00:01:20,280
1.119 - is one of those deceptively
boring releases.

26
00:01:20,280 --> 00:01:23,639
No fireworks,
but if your team lives on GitLab,

27
00:01:23,639 --> 00:01:29,360
the difference is basically zero workflow
versus actual workflow.

28
00:01:23,639 --> 00:01:29,360
That's

29
00:01:29,360 --> 00:01:29,880
the bit.

30
00:01:29,880 --> 00:01:34,276
People hear, "oh, they added more URL
support." Nice-to-have.

31
00:01:34,276 --> 00:01:35,953
Little checkbox feature.

32
00:01:35,953 --> 00:01:38,442
But for loads of real engineering teams,

33
00:01:38,442 --> 00:01:42,813
especially enterprise mobs on internal
GitHub Enterprise or shops that

34
00:01:42,813 --> 00:01:47,618
standardized on GitLab, `GitHub-only`
wasn't a limitation - it was a HARD

35
00:01:47,618 --> 00:01:48,428
blocker.

36
00:01:48,428 --> 00:01:50,986
You couldn't really use the command the
intended way,

37
00:01:50,986 --> 00:01:53,158
full stop.Let me push on that,

38
00:01:53,158 --> 00:01:53,357
though.

39
00:01:53,357 --> 00:01:55,365
Because if you're determined enough,

40
00:01:55,365 --> 00:01:57,283
you can always copy and paste a diff,

41
00:01:57,283 --> 00:02:00,610
paste the PR description,
maybe dump the commit list.

42
00:02:00,610 --> 00:02:01,969
Annoying, sure.

43
00:02:01,969 --> 00:02:05,202
But blocker?Yeah, blocker.

44
00:02:05,202 --> 00:02:10,344
Because the whole magic of `--from-pr` is
that the session starts with

45
00:02:10,344 --> 00:02:12,566
the full diff, the commit list,

46
00:02:12,566 --> 00:02:15,206
and the PR description ALREADY loaded.

47
00:02:15,206 --> 00:02:18,965
Not eventually,
not after ten minutes of manual setup.

48
00:02:18,965 --> 00:02:20,201
Immediately.

49
00:02:20,201 --> 00:02:24,206
That means Claude can start doing useful
work right out of the gate -

50
00:02:24,206 --> 00:02:26,480
review the change, suggest tests,

51
00:02:26,480 --> 00:02:29,440
help debug why CI is falling over.

52
00:02:29,440 --> 00:02:33,359
If you have to hand-feed all that context
every time,

53
00:02:33,359 --> 00:02:37,116
the workflow becomes like kick-starting
an old trail bike with thongs

54
00:02:37,116 --> 00:02:45,282
on - technically possible,
deeply stupid.The "full diff,

55
00:02:37,116 --> 00:02:45,282
commit list, and PR description"

56
00:02:45,282 --> 00:02:46,793
trio is the key.

57
00:02:46,793 --> 00:02:49,161
Because those are three different layers
of intent.

58
00:02:49,161 --> 00:02:51,402
The diff is what changed.

59
00:02:51,402 --> 00:02:53,484
The commit list is how it evolved.

60
00:02:53,484 --> 00:02:57,003
The PR description is what the author
thinks they're doing.

61
00:02:57,003 --> 00:03:00,804
If you miss one,
you can misread the whole thing.

62
00:02:57,003 --> 00:03:00,804
Beautifully put.

63
00:03:00,804 --> 00:03:04,643
And that's why this release feels
practical rather than flashy.

64
00:03:04,643 --> 00:03:07,845
It matches how review actually works in
the trenches.

65
00:03:07,845 --> 00:03:10,090
You don't review code in a vacuum.

66
00:03:10,090 --> 00:03:16,240
You review a change set, with history and
explanation attached.

67
00:03:16,240 --> 00:03:20,124
So do you see this as Claude Code
becoming more of a review companion

68
00:03:20,124 --> 00:03:21,197
than a coding assistant?

69
00:03:21,197 --> 00:03:25,918
Because `--from-pr` is not "help me write
a function." It's more like,

70
00:03:25,918 --> 00:03:30,683
"jump into an active engineering
conversation."I think that's fair...

71
00:03:30,683 --> 00:03:32,686
though not exactly either-or.

72
00:03:32,686 --> 00:03:34,847
It's still a coding assistant.

73
00:03:34,847 --> 00:03:39,414
But this feature leans hard into the
review side of engineering,

74
00:03:39,414 --> 00:03:42,376
which honestly is where a lot of time
gets chewed up.

75
00:03:42,376 --> 00:03:44,537
Writing code is one part.

76
00:03:44,537 --> 00:03:48,140
Understanding someone else's code at 4:47
p.

77
00:03:44,537 --> 00:03:48,140
m.

78
00:03:48,140 --> 00:03:49,264
before stand-down?

79
00:03:49,264 --> 00:03:53,520
That's the real wrestle.

80
00:03:49,264 --> 00:03:53,520
And the CI angle matters.

81
00:03:53,520 --> 00:03:55,278
You mentioned debugging failures.

82
00:03:55,278 --> 00:03:58,075
If the PR context is preloaded,

83
00:03:58,075 --> 00:04:01,629
you're not starting from, "hey Claude,
here's a vague problem." You're

84
00:04:01,629 --> 00:04:07,907
starting from PR 99 on GitLab or PR 42 on
GitHub Enterprise with the exact

85
00:04:07,907 --> 00:04:09,517
proposed change in front of you.

86
00:04:09,517 --> 00:04:12,320
That's a much better prompt surface.

87
00:04:09,517 --> 00:04:12,320
Right.

88
00:04:12,320 --> 00:04:15,519
It reduces that dreadful context tax.

89
00:04:15,519 --> 00:04:19,242
Developers pay this little tax all day -
copy this,

90
00:04:19,242 --> 00:04:21,842
explain that, paste logs,
paste the summary,

91
00:04:21,842 --> 00:04:24,806
whoops forgot the commit that introduced
the regression.

92
00:04:24,806 --> 00:04:27,525
This shaves that off.

93
00:04:27,525 --> 00:04:31,079
And shaving off setup friction matters
more than people admit,

94
00:04:31,079 --> 00:04:36,122
because tiny bits of friction are where
tools quietly die.

95
00:04:36,122 --> 00:04:37,919
I think that's the surprise here.

96
00:04:37,919 --> 00:04:42,511
The headline looks like "more providers
supported." The real story is:

97
00:04:42,511 --> 00:04:46,311
the tool now fits the existing review
ritual of more teams.

98
00:04:46,311 --> 00:04:48,551
And engineers are creatures of ritual.

99
00:04:48,551 --> 00:04:51,191
If a tool asks them to leave the PR,

100
00:04:51,191 --> 00:04:53,594
reconstruct context, and babysit it,

101
00:04:53,594 --> 00:04:55,320
they won't do it consistently.

102
00:04:53,594 --> 00:04:55,320
Yep.

103
00:04:55,320 --> 00:04:59,246
Finally the tool matches how real teams
actually work,

104
00:04:59,246 --> 00:05:03,473
rather than how a neat demo works on a
sunny day with a public GitHub

105
00:05:03,473 --> 00:05:03,889
repo.

106
00:05:03,889 --> 00:05:06,288
And, look, that's maturity.

107
00:05:06,288 --> 00:05:13,067
Less "look what we can do," more "here's
how your workplace actually

108
00:05:06,288 --> 00:05:13,067
behaves."[short

109
00:05:13,067 --> 00:05:21,589
pause] The `https://bitbucket.

110
00:05:13,067 --> 00:05:21,589
org/org/repo/pull-requests/7` example
sticks

111
00:05:21,589 --> 00:05:22,834
with me, weirdly.

112
00:05:22,834 --> 00:05:26,836
Because Bitbucket users are always the
ones left out in these launches.

113
00:05:26,836 --> 00:05:31,072
It's like every tool says "supports code
hosting," and then whispers,

114
00:05:31,072 --> 00:05:36,452
"well...

115
00:05:31,072 --> 00:05:36,452
one code host."Bitbucket users
finally getting invited

116
00:05:36,452 --> 00:05:37,118
to the barbie.

117
00:05:37,118 --> 00:05:42,998
Love that for them.

118
00:05:37,118 --> 00:05:42,998
The other half of v2.

119
00:05:37,118 --> 00:05:42,998
1.119 is quieter,

120
00:05:42,998 --> 00:05:45,076
but maybe even more revealing.

121
00:05:45,076 --> 00:05:50,164
Claude Code output used to handle
`owner/repo#N` shorthand as if the

122
00:05:50,164 --> 00:05:52,270
world was always `github.

123
00:05:50,164 --> 00:05:52,270
com`.

124
00:05:52,270 --> 00:05:57,127
Now that shorthand points to YOUR git
remote host instead.[questioning

125
00:05:57,127 --> 00:06:00,640
tone] And "your git remote host" is the
important token there.

126
00:06:00,640 --> 00:06:03,239
So if your repo lives on `gitlab.

127
00:06:00,640 --> 00:06:03,239
com`,

128
00:06:03,239 --> 00:06:06,364
or `github.

129
00:06:03,239 --> 00:06:06,364
mycompany.

130
00:06:03,239 --> 00:06:06,364
com`, or some other host,

131
00:06:06,364 --> 00:06:11,801
the link doesn't just punt you off to the
wrong neighborhood anymore.

132
00:06:06,364 --> 00:06:11,801
Exactly.

133
00:06:11,801 --> 00:06:14,520
It sounds cosmetic - just a link,

134
00:06:14,520 --> 00:06:15,240
who cares.

135
00:06:15,240 --> 00:06:19,715
But bad links are one of those paper-cut
bugs that make software feel

136
00:06:19,715 --> 00:06:21,073
fake-enterprise.

137
00:06:21,073 --> 00:06:24,836
The output says `owner/repo#N`,

138
00:06:24,836 --> 00:06:28,269
you click it, and suddenly you're on
public GitHub looking at nothing

139
00:06:28,269 --> 00:06:28,763
useful.

140
00:06:28,763 --> 00:06:32,397
That breaks trust fast.
Yeah,

141
00:06:32,397 --> 00:06:35,912
because a shorthand only saves time if it
lands in the right paddock.

142
00:06:35,912 --> 00:06:38,631
Otherwise it's not shorthand,
it's a detour.

143
00:06:38,631 --> 00:06:40,944
And for teams on internal hosts,

144
00:06:40,944 --> 00:06:42,937
that detour can be impossible,

145
00:06:42,937 --> 00:06:46,963
not just annoying.

146
00:06:42,937 --> 00:06:46,963
Then there's the `prUrlTemplate` setting,

147
00:06:46,963 --> 00:06:49,682
which is a really nerdy little power
feature.

148
00:06:49,682 --> 00:06:53,038
Teams can customize where that footer PR
badge links.

149
00:06:53,038 --> 00:06:56,474
So if you're not using a standard
GitHub-style pattern,

150
00:06:56,474 --> 00:07:00,172
you can point it somewhere else
entirely.And the named examples

151
00:07:00,172 --> 00:07:04,036
there - Gerrit and Phabricator - are
doing a lot of work.

152
00:07:04,036 --> 00:07:09,797
Because those platforms tell you this
isn't just "GitHub plus maybe GitLab."

153
00:07:09,797 --> 00:07:14,359
It's acknowledging that some engineering
orgs have genuinely weird,

154
00:07:14,359 --> 00:07:18,600
legacy,
or deeply internal review systems.

155
00:07:14,359 --> 00:07:18,600
Right.

156
00:07:18,600 --> 00:07:22,989
Gerrit, Phabricator, internal review
tools - those aren't edge cases

157
00:07:22,989 --> 00:07:23,763
in big companies.

158
00:07:23,763 --> 00:07:24,963
They're just reality.

159
00:07:24,963 --> 00:07:30,033
So `prUrlTemplate` says, "fine,
tell us your pattern." That's a very

160
00:07:30,033 --> 00:07:34,362
different philosophy from forcing
everybody into one URL shape.

161
00:07:30,033 --> 00:07:34,362
I was gonna

162
00:07:34,362 --> 00:07:36,480
say it's glamorous, but it's the opposite.

163
00:07:36,480 --> 00:07:39,030
It's infrastructure thinking.

164
00:07:39,030 --> 00:07:42,793
Portable, configurable, less opinionated.

165
00:07:42,793 --> 00:07:44,713
Same with the `CLAUDE_CODE_HIDE_CWD`
tweak.

166
00:07:44,713 --> 00:08:00,873
That hides the working directory from the
startup logo,

167
00:08:00,873 --> 00:08:03,117
which is a tiny blessing if you're
screen-sharing or doing a live demo.

168
00:08:00,873 --> 00:08:03,117
The

169
00:08:03,117 --> 00:08:09,645
variable name itself -
`CLAUDE_CODE_HIDE_CWD` - is hilariously

170
00:08:03,117 --> 00:08:09,645
blunt.

171
00:08:09,645 --> 00:08:14,287
But on a call, your current working
directory can absolutely leak stuff.

172
00:08:14,287 --> 00:08:17,053
Project names, customer names,

173
00:08:17,053 --> 00:08:20,132
internal folder structures,
maybe even your username.

174
00:08:20,132 --> 00:08:24,242
I've seen worse exposed in demos.
Oh,

175
00:08:24,242 --> 00:08:24,722
same.

176
00:08:24,722 --> 00:08:32,777
Nothing wakes you up faster than
realizing you've just shown

177
00:08:24,722 --> 00:08:32,777
`/clients/very-sensitive-thing/do-not-shar
e`

178
00:08:32,777 --> 00:08:34,846
to twenty people on Zoom.

179
00:08:34,846 --> 00:08:38,403
Hiding the CWD is not some grand
innovation.

180
00:08:38,403 --> 00:08:40,483
It's just respectful software.

181
00:08:40,483 --> 00:08:45,465
It understands that developers work in
public sometimes.

182
00:08:40,483 --> 00:08:45,465
And when you

183
00:08:45,465 --> 00:08:48,440
line these up - `--from-pr` for GitLab,

184
00:08:48,440 --> 00:08:52,582
Bitbucket, GitHub Enterprise; shorthand
links resolving to the actual

185
00:08:52,582 --> 00:08:58,043
remote host; `prUrlTemplate`;
`CLAUDE_CODE_HIDE_CWD` - the pattern is

186
00:08:58,043 --> 00:08:58,683
obvious.

187
00:08:58,683 --> 00:09:05,334
This release is about portability across
engineering stacks.

188
00:09:05,334 --> 00:09:05,478
Yeah.

189
00:09:05,478 --> 00:09:08,360
Less opinionated about where your code
lives,

190
00:09:08,360 --> 00:09:12,519
where your reviews happen, what weird
system your company inherited in

191
00:09:12,519 --> 00:09:14,358
2014 and never escaped.

192
00:09:14,358 --> 00:09:15,958
That's healthy.

193
00:09:15,958 --> 00:09:21,203
Good dev tools shouldn't make you
rearrange the org chart just to use

194
00:09:21,203 --> 00:09:23,362
them.

195
00:09:21,203 --> 00:09:23,362
I think that's the bigger shift.

196
00:09:23,362 --> 00:09:27,562
Not "Claude Code learned a new trick."
More like: Claude Code is admitting

197
00:09:27,562 --> 00:09:30,557
the software world is messy, multi-host,

198
00:09:30,557 --> 00:09:34,376
full of enterprise oddities,
and not everybody lives on clean public

199
00:09:34,376 --> 00:09:37,878
GitHub.

200
00:09:34,376 --> 00:09:37,878
And honestly,
that's when tools become real.

201
00:09:37,878 --> 00:09:42,760
Not when they ace the demo,
but when they survive the weirdness.

202
00:09:42,760 --> 00:09:44,917
The internal host.

203
00:09:44,917 --> 00:09:46,918
The custom review path.

204
00:09:46,918 --> 00:09:50,760
The screen share you forgot was exposing
too much.

205
00:09:50,760 --> 00:09:54,804
That's the stuff that separates a clever
toy from something a team can

206
00:09:54,804 --> 00:09:59,171
actually adopt.Which leaves a
nice question hanging: if the model

207
00:09:59,171 --> 00:10:02,156
is getting less opinionated about the
stack around it,

208
00:10:02,156 --> 00:10:06,000
does that make the assistant more useful
- or just more invisible?

209
00:10:06,000 --> 00:10:11,068
Because the best tooling in engineering
often disappears into the workflow.

210
00:10:11,068 --> 00:10:12,675
If it disappears into the workflow,

211
00:10:12,675 --> 00:10:14,755
mate, that's usually a very good sign.

212
00:10:14,755 --> 00:10:16,915
Catch you next time.