I run a Claude Code agent 24/7 on a Mac Mini. After a few months my morning routine was gone and I was reviewing agent output at midnight. Built this to teach it boundaries.
The interesting part ended up being the error registry. Agents fail silently way more than you'd expect. Same error repeats 50 times burning tokens before you notice.
Zero dependencies, Python stdlib only. Would love feedback on what's missing.
I run a Claude Code agent 24/7 on a Mac Mini. After a few months my morning routine was gone and I was reviewing agent output at midnight. Built this to teach it boundaries.
The interesting part ended up being the error registry. Agents fail silently way more than you'd expect. Same error repeats 50 times burning tokens before you notice.
Zero dependencies, Python stdlib only. Would love feedback on what's missing.