1
00:00:00,219 --> 00:00:05,239
So, uh, I was looking through the release
notes for Codex zero point one forty seven

2
00:00:05,440 --> 00:00:09,599
point zero, and, well, they completely
killed it.

3
00:00:09,659 --> 00:00:15,000
They removed the deprecated codex exec
dash dash full auto flag entirely.

4
00:00:16,101 --> 00:00:19,001
Wait, they actually deleted full auto?

5
00:00:19,621 --> 00:00:23,521
Oh man, that explain, that explains why
our nightly test suite was,

6
00:00:23,721 --> 00:00:26,441
uh, completely exploding at three in the
morning!

7
00:00:27,430 --> 00:00:29,730
Yeah, it is gone gone.

8
00:00:30,289 --> 00:00:36,109
The release explicitly says to remove the
deprecated codex exec dash dash full auto

9
00:00:36,170 --> 00:00:40,549
flag, and use dash dash sandbox workspace
write instead.

10
00:00:41,683 --> 00:00:46,262
I, I spent two hours this morning trying
to figure out why the CI runner suddenly

11
00:00:46,342 --> 00:00:49,043
started failing with an invalid flag
error.

12
00:00:49,803 --> 00:00:54,583
I mean, we relied on full auto for all our
automated scripts because it just,

13
00:00:54,602 --> 00:00:57,142
you know, bypassed all the prompts without
stopping.

14
00:00:57,677 --> 00:00:59,817
Right, but that was always the danger,
wasn't it?

15
00:01:00,458 --> 00:01:03,697
Full auto mashed together two very
different things.

16
00:01:03,757 --> 00:01:08,857
It mixed execution permissions with
approve, approval rights.

17
00:01:08,958 --> 00:01:14,517
So if an agent went rogue, it had full
root access to your entire filesystem and

18
00:01:14,577 --> 00:01:14,937
network.

19
00:01:15,298 --> 00:01:22,157
Okay, so by splitting them up, how does
dash dash sandbox workspace write actually

20
00:01:22,277 --> 00:01:22,757
fix that?

21
00:01:23,202 --> 00:01:25,422
It creates a strict security boundary.

22
00:01:26,063 --> 00:01:30,583
Workspace write locks the agent down so it
can only write files inside your local

23
00:01:30,682 --> 00:01:31,642
workspace directory.

24
00:01:32,182 --> 00:01:37,802
It can't touch system files, and it blocks
unauthorized network calls out to random

25
00:01:37,882 --> 00:01:38,702
external hosts.

26
00:01:39,209 --> 00:01:40,469
Ah, okay.

27
00:01:41,069 --> 00:01:44,609
So workspace write restricts where it can
touch the disk.

28
00:01:45,269 --> 00:01:47,970
But what about the non interactive
approval part?

29
00:01:48,590 --> 00:01:53,489
If we don't have full auto, does the
script just hang waiting for a human to click

30
00:01:53,549 --> 00:01:53,889
yes?

31
00:01:54,369 --> 00:01:56,029
That is where the new flag comes in.

32
00:01:56,509 --> 00:02:01,129
You pair sandbox workspace write with dash
dash approve for me.

33
00:02:01,869 --> 00:02:06,449
Approve for me handles the automated
approvals so headless scripts keep running,

34
00:02:07,029 --> 00:02:10,510
but without giving subagents blanket host
level access.

35
00:02:10,942 --> 00:02:14,303
Wow, that is actually so much cleaner.

36
00:02:14,983 --> 00:02:19,682
So instead of one dangerous master key
that lets a script delete your home

37
00:02:19,762 --> 00:02:25,382
directory, you decouple the filesystem
sandbox from the auto approval mechanism.

38
00:02:25,989 --> 00:02:26,769
Exactly.

39
00:02:27,169 --> 00:02:31,829
You get headless automation for your
pipelines, but if an agent tries to escape the

40
00:02:31,929 --> 00:02:35,449
workspace folder, the sandbox slams the
door shut.

41
00:02:35,899 --> 00:02:40,900
You know, once I got past the breakages in
zero point one forty seven point zero,

42
00:02:41,619 --> 00:02:46,340
there was actually a huge quality of life
feature that blew me away.

43
00:02:46,899 --> 00:02:49,659
The, uh, the MCP protocol update!

44
00:02:50,280 --> 00:02:53,740
Oh, the, the twenty twenty six zero seven
twenty eight update?

45
00:02:54,276 --> 00:02:54,917
Yes!

46
00:02:55,296 --> 00:02:59,036
They adopted the non blocking server
initialization from that spec.

47
00:02:59,576 --> 00:03:03,476
It used to be that if you loaded heavy
Model Context Protocol tools,

48
00:03:03,896 --> 00:03:08,796
your terminal input would completely
freeze until every single server booted up.

49
00:03:09,242 --> 00:03:11,022
Ugh, I remember that.

50
00:03:11,642 --> 00:03:16,302
You would launch a session and just sit
there for ten seconds staring at a frozen

51
00:03:16,362 --> 00:03:19,622
cursor while optional background tools
initialized.

52
00:03:20,384 --> 00:03:21,245
Exactly!

53
00:03:21,845 --> 00:03:24,164
Now it is completely non blocking.

54
00:03:24,545 --> 00:03:29,404
The terminal is responsive instantly, and
the MCP tools spin up in the background

55
00:03:29,444 --> 00:03:30,784
without locking your prompt.

56
00:03:31,344 --> 00:03:33,384
That is huge for agent velocity.

57
00:03:34,064 --> 00:03:38,204
And speaking of security and cleanliness,
did you see what they added for logging?

58
00:03:38,844 --> 00:03:43,844
They added automatic secret redacting
across command displays and replayed chat

59
00:03:43,904 --> 00:03:44,204
history.

60
00:03:45,169 --> 00:03:46,169
Wait, really?

61
00:03:46,669 --> 00:03:49,289
It catches bearer tokens automatically?

62
00:03:49,846 --> 00:03:56,107
Yeah, bearer tokens, raw API credentials,
private headers, it redacts them before

63
00:03:56,147 --> 00:03:59,167
they ever hit stdout logs or transcript
history.

64
00:03:59,946 --> 00:04:03,507
So if a subagent runs a curl command with
an authorization header,

65
00:04:04,066 --> 00:04:06,026
it won't leak into your terminal logs.

66
00:04:07,062 --> 00:04:08,682
That is such a relief.

67
00:04:09,222 --> 00:04:13,563
I can't tell you how many times I've had
to scrub raw bearer tokens out of shared

68
00:04:13,602 --> 00:04:16,562
terminal recordings or debug transcripts.

69
00:04:17,016 --> 00:04:17,636
Same here.

70
00:04:18,236 --> 00:04:22,656
Oh, and they also added persisted,
manually ordered thread sections.

71
00:04:23,256 --> 00:04:27,337
So when you are managing complex, multi
agent sessions over several days,

72
00:04:27,877 --> 00:04:32,496
you can organize your long running threads
without losing your exact context or

73
00:04:32,636 --> 00:04:33,677
transcript positioning.

74
00:04:34,027 --> 00:04:37,107
Oh, that is so nice for keeping context
clean!

75
00:04:37,708 --> 00:04:42,567
So if a team wants to roll out zero point
one forty seven point zero across all

76
00:04:42,607 --> 00:04:46,027
their developers today, what is the best
way to standardize it?

77
00:04:46,572 --> 00:04:47,013
Simple.

78
00:04:47,532 --> 00:04:51,412
Update your local dot codex slash config
dot toml file.

79
00:04:52,052 --> 00:04:57,812
Set your default flags in config dot toml
to include sandbox workspace write and

80
00:04:57,852 --> 00:05:03,632
approve for me across the team, so nobody
relies on old deprecated flags like full

81
00:05:03,692 --> 00:05:05,072
auto ever again.

82
00:05:05,524 --> 00:05:06,904
Done and done.

83
00:05:07,044 --> 00:05:09,545
I am updating our team config right now.

84
00:05:09,625 --> 00:05:10,324
Good chat, Ethan!

85
00:05:11,324 --> 00:05:12,485
Talk soon, Maya.