@ArchiveExplorer - Anthropic Engineers' Claude Config

Open original

@ArchiveExplorer on X The author says a friend at Anthropic gave them the .claude/ folder that Anthropic's own engineers use to build Claude, and they published it in a public GitHub repo with a one-line install script. The claim is that dropping this setup into a project stops Claude from writing trivial tests, declaring work done on its own smoke tests, and asking for confirmation on every step.

  • The folder reportedly contains 33 skills, a verifier subagent, and a 7-file harness, described as the internal version rather than the cleaned-up one in the public docs.
  • The author says the setup changed Claude's behavior: fewer tests that test nothing, no self-certified "done" based on its own smoke tests, and fewer confirmation requests.
  • The config was posted as a curl-to-bash install from a personal GitHub repo, and the post says the material was shared against the friend's request not to repost it.

The post is a tweet-length claim that someone obtained a private `.claude/` configuration folder from an Anthropic engineer, described as the setup internal coders use to build Claude. It lists the contents: 33 skills, a verifier subagent, and a seven-file harness. The author says that after dropping it into their own project, Claude stopped writing tests that verify nothing, stopped declaring features finished based on smoke tests it had written itself, and stopped asking for confirmation on every `ls` command. The point being sold is behavioral: the same model, configured this way, reportedly behaves more rigorously. The substance worth a second look is the install step and the sourcing. The author has published the whole thing in a public GitHub repo and offers a one-line `curl | bash` installer that runs a script from that repo directly on your machine. Running a remote script with shell access is a meaningful trust decision, so anyone tempted should read `install.sh` first and check what it writes and where. The claim of insider origin is unverified and the post itself says the material was shared against the source's wishes, which is a reason to be cautious about both the authenticity and the legal footing of redistributing it. The closing line about bookmarking "before Anthropic's lawyers notice" is a pressure tactic, not evidence. Separately, the behavioral improvements described are plausible in general terms, since explicit instructions about testing standards, verification subagents, and when to stop asking permission do change how an agent works. Whether these particular 33 skills and seven files produce those results is something you could test on a sample project in an afternoon, rather than taking the anecdote on faith. For a read-it-later queue, I'd save it as a lead to verify, not as a setup to install.

Ask about this article