1
00:00:00,140 --> 00:00:03,619
If you leave a Codex session running in
the background for a long task,

2
00:00:04,159 --> 00:00:06,079
you used to hit this frustrating wall.

3
00:00:06,739 --> 00:00:11,960
You update a plugin in another window, but
your running agent has no idea.

4
00:00:12,039 --> 00:00:17,319
Thanks to Jellypod, we can dive straight
into how Codex CLI zero point one hundred

5
00:00:17,359 --> 00:00:21,459
fifty four point zero fixes that
disconnect once and for all.

6
00:00:21,921 --> 00:00:26,841
Wait, so it was just stuck with whatever
tools it had when you first launched it?

7
00:00:27,429 --> 00:00:28,009
Exactly.

8
00:00:28,490 --> 00:00:33,609
If you upgraded a custom team tool
mid-task, the session was completely desynced.

9
00:00:34,289 --> 00:00:39,630
To get the new functionality, you had to
kill the process, drop your active context,

10
00:00:40,029 --> 00:00:41,349
and start over from scratch.

11
00:00:42,796 --> 00:00:45,996
That sounds like a fast way to lose half
an hour of context.

12
00:00:46,375 --> 00:00:51,615
Right, but now active sessions listen for
app installed event triggers.

13
00:00:51,687 --> 00:00:57,015
When an external update or rollback
happens, Codex dynamically re-indexes your

14
00:00:57,095 --> 00:01:02,295
skills, re-binds command hooks, and pulls
in the new tools automatically.

15
00:01:02,455 --> 00:01:09,015
Pull requests four two two eight four and
four two five nine three make sure it

16
00:01:09,065 --> 00:01:12,135
happens quietly without touching your
running worktree.

17
00:01:12,466 --> 00:01:15,786
Okay, let me make sure I've got the
practical picture right.

18
00:01:16,426 --> 00:01:20,986
Say I have an agent working through a
multi-step refactor in one tab,

19
00:01:21,547 --> 00:01:25,746
and I push a fix to our internal
deployment plugin in another terminal.

20
00:01:26,426 --> 00:01:27,167
The agent just...

21
00:01:27,826 --> 00:01:29,706
picks up the new command immediately?

22
00:01:30,286 --> 00:01:33,286
No restart, no lost chat history?

23
00:01:33,704 --> 00:01:35,744
That is it, zero interruption.

24
00:01:36,244 --> 00:01:41,364
And speaking of startup, there is also a
critical security fix in zero point one

25
00:01:41,464 --> 00:01:42,905
hundred fifty four point zero.

26
00:01:43,764 --> 00:01:49,585
Pull request four two three two four stops
Codex from executing workspace-controlled

27
00:01:49,664 --> 00:01:55,024
PATH helper binaries during startup until
you explicitly trust the repository.

28
00:01:55,423 --> 00:02:02,302
Wait, so before this, just opening an
untrusted repo could run arbitrary binaries

29
00:02:02,442 --> 00:02:03,322
hidden in PATH?

30
00:02:03,807 --> 00:02:06,448
It was a real edge case for workspace
security.

31
00:02:07,147 --> 00:02:10,787
Now it halts execution until you give
explicit authorization,

32
00:02:11,427 --> 00:02:14,767
keeping untrusted repos completely
isolated.

33
00:02:14,873 --> 00:02:19,053
Now, if you maintain CI scripts or local
wrapper scripts around Codex,

34
00:02:19,573 --> 00:02:22,074
there is a hard breaking change you need
to look out for.

35
00:02:22,873 --> 00:02:28,033
Pull request four two nine nine three
completely removes the deprecated codex

36
00:02:28,253 --> 00:02:29,693
mcp-server entry point.

37
00:02:30,789 --> 00:02:31,869
It's just gone?

38
00:02:32,629 --> 00:02:35,190
What happens if an automated pipeline
still calls it?

39
00:02:35,822 --> 00:02:36,962
It fails immediately.

40
00:02:37,622 --> 00:02:42,762
Automated workflows need to switch over to
standard daemon invocation or use the app

41
00:02:42,822 --> 00:02:44,282
installed events right away.

42
00:02:44,736 --> 00:02:45,676
Okay, good to know.

43
00:02:46,296 --> 00:02:51,496
But on the developer experience side,
there are some really neat terminal updates,

44
00:02:51,576 --> 00:02:51,796
right?

45
00:02:52,167 --> 00:02:57,127
Yeah, the slash copy command got a great
upgrade in pull request four three zero

46
00:02:57,191 --> 00:02:57,847
five five.

47
00:02:57,957 --> 00:03:03,607
Instead of dumping the entire transcript
into your clipboard just to pull one value,

48
00:03:03,647 --> 00:03:09,767
you can target specific diagnostic
metadata using slash copy status or individual

49
00:03:09,836 --> 00:03:10,807
output fields.

50
00:03:11,092 --> 00:03:12,053
Oh, I love that.

51
00:03:12,592 --> 00:03:16,572
When you're filing a bug with your
platform team, you do not want to wade through

52
00:03:16,712 --> 00:03:20,852
thousands of lines of code output just to
get a session ID or status code.

53
00:03:21,167 --> 00:03:21,807
Totally.

54
00:03:21,947 --> 00:03:28,047
And to pair with that, pull request four
two eight two one adds explicit reporting

55
00:03:28,107 --> 00:03:32,847
for managed filesystem sandbox policies
right inside codex doctor,

56
00:03:32,954 --> 00:03:36,207
so you can see your boundary enforcement
rules at a glance.

57
00:03:36,607 --> 00:03:42,587
So between targeted copying and clearer
doctor reports, tracking down session issues

58
00:03:42,607 --> 00:03:44,208
should be way less painful.

59
00:03:44,583 --> 00:03:45,583
Definitely.

60
00:03:45,655 --> 00:03:49,703
Just make sure to audit those unpinned CI
scripts before upgrading,

61
00:03:49,743 --> 00:03:50,823
and you are good to go.