You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
wait_for_text's delays (90 seconds) is can lock agentic flows - blocking #59
wait_for_text with an unlucky query (things may have come to pass, whether it's not a match and didn't work, the wrong query didn't yield a match, or it worked) - it'll be stuck 90 seconds with the timeout.
perhaps rearchitect it and see if it's possible to have the terminal stream and let the agent read it and make up their own mind.
Wait for text can be flawed because in practice, it may miss oppurtunity to read all the way up a stack (e.g. is a server already running? old server already running? does it need to be ctrl-c'd?)
wait for text is what you could use if you have 100% certainty of the pane's current state, or else you could be stuck waiting for something that has already come to pass, or cannot pass (e.g. you need to exit and start a server again).
Superseded by #61 for the observation-first architecture and #60 for the concrete capture_since implementation path. Keeping the original symptom here for provenance: blocking wait_for_text can lock an agent out when the expected pattern is wrong or incomplete, but the actionable work now lives in those two issues.
Activity
tony commentedon May 22, 2026
Wait for text can be flawed because in practice, it may miss oppurtunity to read all the way up a stack (e.g. is a server already running? old server already running? does it need to be
ctrl-c'd?)wait for text is what you could use if you have 100% certainty of the pane's current state, or else you could be stuck waiting for something that has already come to pass, or cannot pass (e.g. you need to exit and start a server again).
also 90 seconds is a long time.
tony commentedon May 24, 2026
Superseded by #61 for the observation-first architecture and #60 for the concrete
capture_sinceimplementation path. Keeping the original symptom here for provenance: blockingwait_for_textcan lock an agent out when the expected pattern is wrong or incomplete, but the actionable work now lives in those two issues.