6 ms·

We are talking about a thing that's consumed all human knowledge whose superpowers are summarization and understanding. Why the fuck are we writing all these .m

by Swizec 27d ago

We are talking about a thing that's consumed all human knowledge whose superpowers are summarization and understanding. Why the fuck are we writing all these .md files!? I resent everything about this. But I write the .md files because others on the team are not as good at Just Talking To It. The md files are there because juniors don't know what to ask for.

groby_b 27d ago
Uh. The .md files are also there so I don't need to say every single time "In this house, we believe in uv & PEP723. Also, no fucking localized imports, no just-in-case try/except." (Or the equivalent incantations for other languages) They aren't there for others, they are there to make sure the code produced measures up to my standards. Sure, it helps that it also applies to requests made by others, but that's a bonus.
- Swizec 27d ago
  > time "In this house, we believe in uv & PEP723. Also, no fucking localized imports, no just-in-case try/except." My point is the all knowing all seeing big brain super bot should already know how to write good code. It feels dumb that we have to keep reminding it
  - AndrewDucker 27d ago
    If "What good code looks like" was something that everyone agreed on then there wouldn't be a problem.
- electric_mayhem 27d ago
  And you don’t find that Claude cheerfully ignores the content of your CLAUDE.md and does whatever the fuck it wants instead? I caught it ignoring the prompt’s details, so added “include the output of the Unix ‘date’ command with each response. And that worked great in that its absence became an indicator that Claude was off-prompt. Until it started including made-up strings that resembled, but had nothing to do with, the output of the Unix ‘date’ command.
  - groby_b 27d ago
    It's working reasonably well for me. Given your description of the behavior, my guess is context window compression bit you. Less likely if it's a root level CLAUDE.md and you used a recent version of Claude Code.
- isoprophlex 27d ago
  won't the oracle do something vaguely pep723-like once its read a bunch of existing, conformant files into its context? And/or wont the git hooks bitch at it when your scripts folder has files without inline metadata?
  - groby_b 27d ago
    Possibly. But telling it in a rules file eats significantly less context than if it needs to read a bunch of files every time just to find what "conformant" means. AGENTS.md is a poor substitute for agents actually learning, but it somewhat works. And sure, you can (and should) enforce constraints via githooks whenever you can (determinism is good), but if the agent only finds out at checkin time, it's created a whole bunch of code it needs to redo, and that doesn't run properly if you do the caveman thing and actually run your code before committing. Again, token & context waste.

by Swizec 27d ago

groby_b 27d ago

Uh. The .md files are also there so I don't need to say every single time "In this house, we believe in uv & PEP723. Also, no fucking localized imports, no just-in-case try/except." (Or the equivalent incantations for other languages) They aren't there for others, they are there to make sure the code produced measures up to my standards. Sure, it helps that it also applies to requests made by others, but that's a bonus.

Swizec 27d ago
> time "In this house, we believe in uv & PEP723. Also, no fucking localized imports, no just-in-case try/except." My point is the all knowing all seeing big brain super bot should already know how to write good code. It feels dumb that we have to keep reminding it
- AndrewDucker 27d ago
  If "What good code looks like" was something that everyone agreed on then there wouldn't be a problem.
electric_mayhem 27d ago
And you don’t find that Claude cheerfully ignores the content of your CLAUDE.md and does whatever the fuck it wants instead? I caught it ignoring the prompt’s details, so added “include the output of the Unix ‘date’ command with each response. And that worked great in that its absence became an indicator that Claude was off-prompt. Until it started including made-up strings that resembled, but had nothing to do with, the output of the Unix ‘date’ command.
- groby_b 27d ago
  It's working reasonably well for me. Given your description of the behavior, my guess is context window compression bit you. Less likely if it's a root level CLAUDE.md and you used a recent version of Claude Code.
isoprophlex 27d ago
won't the oracle do something vaguely pep723-like once its read a bunch of existing, conformant files into its context? And/or wont the git hooks bitch at it when your scripts folder has files without inline metadata?
- groby_b 27d ago
  Possibly. But telling it in a rules file eats significantly less context than if it needs to read a bunch of files every time just to find what "conformant" means. AGENTS.md is a poor substitute for agents actually learning, but it somewhat works. And sure, you can (and should) enforce constraints via githooks whenever you can (determinism is good), but if the agent only finds out at checkin time, it's created a whole bunch of code it needs to redo, and that doesn't run properly if you do the caveman thing and actually run your code before committing. Again, token & context waste.