anomalyco
question: bash tool timed out after 60 seconds?
Description
hey so seems like when llm runs some long running cli process such as docker builds via the bash tool . it seems to timeout the process after 60s ?or such so . even tho the process had flowing output on the tui . timeout is a parameter for llm to choose or its just default hardcoded ?
OpenCode version
No response
Steps to reproduce
No response
Screenshot and/or share link
No response
Operating System
No response
Terminal
No response
This issue might be a duplicate of existing issues. Please check:
- #1721: bash tool often times out - same issue with bash commands timing out, mentions default timeout behavior
- #4197: Set timout for long running commands - user encountering 120000ms timeout with test suite
- #3950: [FEATURE]: Configurable command timeout in config.json - feature request for configurable timeout for long-running operations
- #2985: Bash tool does not show time out to model or user - related issue about timeout handling and model notification
Feel free to ignore if none of these address your specific case.
![github-actions[bot] avatar](https://avatars.githubusercontent.com/in/15368?v=4 "Published by a pub.dev verified publisher"){kind=small}
it defaults to 60 seconds but the llm can choose a value, we cap it at 10min rn
could make it reset its timer if the terminal output is flowing ? or yeah 10 min also works but a dynamic logic is better ig
the main idea here is to prevent someone from starting a task and letting the llm work and then wondering why it never stops, like a dev server, some servers have background jobs that could still be logging things.
But i see what u want. Other people have asked for it to be configurable and stuff
I suggest that after a 60-second timeout, the task should be run in the background, and the LLM should be returned the logs already output along with a PID. The LLM should be instructed that it can use the job_output(pid, max_wait_time) tool to retrieve and wait for more output, or use job_kill(pid) to terminate the process.
If the process has not finished by the time job_output(pid, max_wait_time) exceeds the wait time, the LLM should be returned the output up to that point, and then prompted that it can wait for a longer duration.
This way, the LLM can keep development servers running continuously in the background, check outputs, or execute long-running tasks like pytest without interruption.
This idea comes from the crush project, and I think it's a great design. I submitted a PR, but they seem too busy to handle it... https://github.com/charmbracelet/crush/pull/1496
the default timeout is 2min now ig we could make it configured, keep in mind tho the agent can set it to any time it wants or you can instruct it to
the default timeout is 2min now ig we could make it configured, keep in mind tho the agent can set it to any time it wants or you can instruct it to
pls make it configurable, I cannot keep telling the agent or the agent keeps forgetting
added OPENCODE_EXPERIMENTAL_BASH_DEFAULT_TIMEOUT_MS env var, it will be in next release
you forgot to say that OPENCODE_EXPERIMENTAL_BASH_DEFAULT_TIMEOUT_MS only accepts a 32bit integer and 999999999999 ms is not allowed and it will default to 1 ms instead