idea_047AI and software8 Sep 20262 updates
An agent that admits it is stuck
statusrunning
Most agent loops fail by pretending. I want one that stops after two bad attempts, writes down what it does not know in plain words, and waits.
The early version is annoying to use and much easier to trust.
update log
Rewrote the stop condition. Two strikes instead of three, and it now prints what it tried.
First version. Loops forever and lies about it.