ive got an awesome setup where my main agent handles multi-step work like compressing its own memory and loading skills dynamically. everything runs thru the same loop, just as described in guide 1024 (you know which one). but heres what tripped me up: when i call out to bash for a long-running task - like running tests that take two minutes - the whole show grinds almost completely still until those pesky processes finish.
i mean seriously. if someone asks where my agent is, it feels like the loop has taken an extended coffee break! now heres what got me thinking: does anyone else run into this issue when your model blocks on external commands? any tips or hacks to keep things humming while waiting for those tasks?
anyone out there faced smth similar and found a workaround w/o making everything synchronous again by accident?
article:
https://dev.to/ivan-magda/background-tasks-the-one-actor-in-the-codebase-and-the-sigterm-bug-that-only-broke-on-linux-4c26