3 ms·

Obfuscated ts/js code is not machine code to begin with, so not sure what’s the big deal. Also, not sure why anthropic doesn’t just make their cli open source

Obfuscated ts/js code is not machine code to begin with, so not sure what’s the big deal. Also, not sure why anthropic doesn’t just make their cli open source - it’s not like it’s something special (Claude is, this cli thingy isn’t)

petcat 6mo ago
> not sure why anthropic doesn’t just make their cli open source They don't want everyone to see how poorly it's implemented and that the whole thing is a big fragile mess riddled with bugs. That's my experience anyway. For instance, just recently their little CLI -> browser oauth login flow was generating malformed URLs and URLs pointing to a localhost port instead of their real website.
- restlake 6mo ago
  caught that too a few weeks ago, couldn’t log in for a few hours either. I did a double take at the localhost when it loaded up in my browser haha
- nasretdinov 6mo ago
  I don't think you really need to look at the source code to understand that it's probably been, let's say, written with a heavy help from Claude itself
- tempest_ 6mo ago
  Look at the gemini-cli. Pretty sure it will look like that
- bpodgursky 6mo ago
  I really don't think they care that much, but it's a tight race and gives them a slight edge over other labs building harnesses, since they are in the lead.
- ramraj07 6mo ago
  Browse through codex and think if anyone cares about the quality of the code before Open sourcing it.
jditu 6mo ago
[flagged]
- beacon294 6mo ago
  This is wrong, Claude is the worst of the best: codex, pi, opencode, droid/factory, there are many more others can vouch for.
  - jditu 6mo ago
    [flagged]
lukan 6mo ago
" - it’s not like it’s something special (Claude is, this cli thingy isn’t)" How do you know? Have you checked the source? Do you know how exactly context is created, memory files, skills? Subagents created with tasks? I don't, but am checking right now. Then I will judge.
- sheeshkebab 6mo ago
  bc I build stuff like this myself - it doesn’t take anything to build a wrapper client around a good llm, including using another llm.
  - lukan 6mo ago
    So .. the stuff you build yourself, you point it to claude then it runs just as productive as Claude CLI? Did you try?
    - sheeshkebab 6mo ago
      Yes
manquer 6mo ago
Naming conventions can reveal a lot about how teams internally are thinking about roadmap and product decisions. That cannot be reversed when obfuscated.

petcat 6mo ago

> not sure why anthropic doesn’t just make their cli open source They don't want everyone to see how poorly it's implemented and that the whole thing is a big fragile mess riddled with bugs. That's my experience anyway. For instance, just recently their little CLI -> browser oauth login flow was generating malformed URLs and URLs pointing to a localhost port instead of their real website.

restlake 6mo ago
caught that too a few weeks ago, couldn’t log in for a few hours either. I did a double take at the localhost when it loaded up in my browser haha
nasretdinov 6mo ago
I don't think you really need to look at the source code to understand that it's probably been, let's say, written with a heavy help from Claude itself
tempest_ 6mo ago
Look at the gemini-cli. Pretty sure it will look like that
bpodgursky 6mo ago
I really don't think they care that much, but it's a tight race and gives them a slight edge over other labs building harnesses, since they are in the lead.
ramraj07 6mo ago
Browse through codex and think if anyone cares about the quality of the code before Open sourcing it.

jditu 6mo ago

[flagged]

beacon294 6mo ago
This is wrong, Claude is the worst of the best: codex, pi, opencode, droid/factory, there are many more others can vouch for.
- jditu 6mo ago
  [flagged]

lukan 6mo ago

" - it’s not like it’s something special (Claude is, this cli thingy isn’t)" How do you know? Have you checked the source? Do you know how exactly context is created, memory files, skills? Subagents created with tasks? I don't, but am checking right now. Then I will judge.

sheeshkebab 6mo ago
bc I build stuff like this myself - it doesn’t take anything to build a wrapper client around a good llm, including using another llm.
- lukan 6mo ago
  So .. the stuff you build yourself, you point it to claude then it runs just as productive as Claude CLI? Did you try?
  - sheeshkebab 6mo ago
    Yes

manquer 6mo ago

Naming conventions can reveal a lot about how teams internally are thinking about roadmap and product decisions. That cannot be reversed when obfuscated.