I read about runtime monitoring and the CPU, memory, and sandbox status it shows. I'm not sure how best to interpret those alongside the task's steps.
Analyzing a file, installing dependencies, and waiting for a model response seem like different kinds of slowness. What do you check first? I'd like a simple diagnostic order rather than treating a restart as the answer to everything.
I first check the last visible task step and its timestamp in Buda, then use runtime monitoring to add context. If the last step is installing dependencies or processing a large file and the sandbox has sustained CPU/memory activity or terminal output, the cloud computer is probably working. If the model turn has started but no tool has run yet, low sandbox usage may mean a model wait. Low CPU alone does not prove it: a website, download or external API can also be waiting with little CPU use.
My quick order is: note the last progress event; check whether tool, browser or terminal output is still moving; check sandbox status, CPU, memory and open ports; then look for “waiting for your input” or an expanded error. If progress has stopped for an unusually long time and those signals disagree, keep the Session link, time and last step for investigation. Restarting the sandbox is a last resort because it stops running processes and clears temporary files that may contain unfinished work.
Which stage is usually slow for you: dependency installation, browsing, or the time before the Agent's first response? The Chinese discussion covers the same check sequence.
Sign in to join the discussion. Reading is always open; you only need an account to post.
Sign inI read about runtime monitoring and the CPU, memory, and sandbox status it shows. I'm not sure how best to interpret those alongside the task's steps.
Analyzing a file, installing dependencies, and waiting for a model response seem like different kinds of slowness. What do you check first? I'd like a simple diagnostic order rather than treating a restart as the answer to everything.
I first check the last visible task step and its timestamp in Buda, then use runtime monitoring to add context. If the last step is installing dependencies or processing a large file and the sandbox has sustained CPU/memory activity or terminal output, the cloud computer is probably working. If the model turn has started but no tool has run yet, low sandbox usage may mean a model wait. Low CPU alone does not prove it: a website, download or external API can also be waiting with little CPU use.
My quick order is: note the last progress event; check whether tool, browser or terminal output is still moving; check sandbox status, CPU, memory and open ports; then look for “waiting for your input” or an expanded error. If progress has stopped for an unusually long time and those signals disagree, keep the Session link, time and last step for investigation. Restarting the sandbox is a last resort because it stops running processes and clears temporary files that may contain unfinished work.
Which stage is usually slow for you: dependency installation, browsing, or the time before the Agent's first response? The Chinese discussion covers the same check sequence.
Sign in to join the discussion. Reading is always open; you only need an account to post.
Sign in