Community threads

@Mention wake failures when using different models

Les Blake · 2026-05-20

My agent workflow uses two primary agent wake patterns:\ -On sub-issue / task creation/assignment\ -Commenting @Mention on open parent task\ \ For example, my Intake Manager agent manages the entire workflow. It creates sub-issues for the other agents which wakes them. Upon completing their work they, comment in the parent task @Intake Manager and the Intake Manager wakes and proceeds with its prescribed steps.\ \ The bug I'm experiencing is that this pattern fails on the analyst running Codex but not on the two running Claude. Everything else is the same. They all three have the exact same instructions upon task completion. When the agent running Codex @Mentions the Intake Manager it does not wake.\ \ One other fact. This has happened twice. The first time, I manually ran my Intake Manager's heartbeat, it saw the comment and proceeded. The second time when I manually ran the heartbeat, it didn't see the comment and went back to sleep.\ \ Does anyone know what could be causing this?

Answers

Les Blake · 2026-05-21

Thanks @jro7 I will look into that. I also am wondering if it might have to do with having the analyst max concurrent runs set to 1.The agent was already running when the other agent posted the @mention comment. That would count as 2 consecutive runs which is not allowed in the default settings.

Jeroen · 2026-05-21

I noticed this too, could be that the heartbeat uses the "cheap model" set on the agent and that was not correct for the model you were using

Les Blake · 2026-05-21

Follow-up: I switched the agent running Codex to Claude and did not experience the bug. Process worked as intended.