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
did the llm say something like "I am sorry I cannot edit files" and did this by chance happen after switching from plan mode? I have seen that happen and we definitely need to address
As in:
switch from plan -> build, ask it to do action, "I cannot do X because I am in plan mode ..."
@rekram1-node in build mode. I get statements like I modified the code or made changes but in actual file, no changes are made. I had installed using curl, the uninstalled and re-installed using bun. No improvement
I am having the exact same problem. I am on a Mac and using the CLI, installed with brew, with GPT5. I am in build mode and it says the changes and new files have been saved but nothing has changed in the project. I am starting opencode from the project directory and using !pwd to ensure the correct directory once opencode is started. To ensure the simplest scenario I tried doing /init in an existing project and it says it created the AGENTS.md but it isn't. Hope this is helpful, happy to provide more details if needed.
Symptom:
The agent will start a task, "working..." will constantly tick over but its either timed out and hasn't recovered or . I am not sure how opencode as yet logs its faults etc, but i'd wager there is some sort of timeout and unable to recover from that moment. thus you see "working..." constantly spinning but its a false positive.
Models
Sonnet | Opus | Sonic
Frequency:
After about 2-3 sessions it starts to arise. That or about after about 8-10 chat cycles (ie prompt/answer/prompt/answer) it eventually becomes a "stalling"
Like I said, I haven't to be blunt taken the time to investigate where or if opencode has any visible logs for this. I will try now but in case I don't this is all I have to give for now.
Same for me. macOS client. Stuck indefinitely on planning phase and unable to move further. Whenever asked to implement the plan it's was giving the same planning output over and over. Model is kimi k2 and to some degree Gemini 2,5 Pro / DeepSeek. Also DeepSeek weren't even able to see files in Project directory resulting in [use the bash tool to list files in the data folder.
I am having a similar issue. Model tries to edit files, I have to manually approve edits (persmission is set to ask). As it spits out multiple few-line edits to the same file, I approve one-by-one (enter), but seems like only the last edit is applied.
Last tried with openrouter grok-4-fast:free.
Closing this as stale - the original reports from Aug-Sep 2025 couldn't be reproduced, and we've since added fixes:
Per-file locking to prevent race conditions
Clearer "Edit applied successfully" messages
For those who were using GPT-5-chat-latest: That model doesn't support tool calling, which is required for file edits. Use a model that supports tools (Claude, GPT-4o, etc.).
If you're still experiencing edit issues on current versions (v1.1.x): Please open a new issue with:
Activity
github-actions commentedon Aug 17, 2025
This issue might be a duplicate of existing issues. Please check:
Feel free to ignore if none of these address your specific case.
rekram1-node commentedon Aug 17, 2025
can you elaborate? What model are you using?
Generally this is a model/prompting issue
Artawower commentedon Aug 17, 2025
Exactly the same behavior. This is not prompting and not a model. I use sonnet4, and claude code edits files, opencode does not.
rekram1-node commentedon Aug 17, 2025
can you provide any more detail at all? Are you in plan mode? or build mode?
fresh install w/ sonnet4 I have 0 issue so I just need a tad more info to help you debug
Artawower commentedon Aug 17, 2025
In my case, it was build mode. However, I am unable to reproduce this behavior again (new session)
rekram1-node commentedon Aug 17, 2025
did the llm say something like "I am sorry I cannot edit files" and did this by chance happen after switching from plan mode? I have seen that happen and we definitely need to address
As in:
switch from plan -> build, ask it to do action, "I cannot do X because I am in plan mode ..."
laflinwalker commentedon Aug 18, 2025
@rekram1-node in build mode. I get statements like I modified the code or made changes but in actual file, no changes are made. I had installed using curl, the uninstalled and re-installed using bun. No improvement
alexwinston commentedon Aug 18, 2025
I am having the exact same problem. I am on a Mac and using the CLI, installed with brew, with GPT5. I am in build mode and it says the changes and new files have been saved but nothing has changed in the project. I am starting opencode from the project directory and using !pwd to ensure the correct directory once opencode is started. To ensure the simplest scenario I tried doing /init in an existing project and it says it created the AGENTS.md but it isn't. Hope this is helpful, happy to provide more details if needed.
rekram1-node commentedon Aug 18, 2025
gpt-5-chat-latest does not support tools btw
scbarnes commentedon Aug 21, 2025
Symptom:
The agent will start a task, "working..." will constantly tick over but its either timed out and hasn't recovered or . I am not sure how opencode as yet logs its faults etc, but i'd wager there is some sort of timeout and unable to recover from that moment. thus you see "working..." constantly spinning but its a false positive.
Models
Sonnet | Opus | Sonic
Frequency:
After about 2-3 sessions it starts to arise. That or about after about 8-10 chat cycles (ie prompt/answer/prompt/answer) it eventually becomes a "stalling"
Like I said, I haven't to be blunt taken the time to investigate where or if opencode has any visible logs for this. I will try now but in case I don't this is all I have to give for now.
SassyMemer commentedon Aug 24, 2025
Same for me. macOS client. Stuck indefinitely on planning phase and unable to move further. Whenever asked to implement the plan it's was giving the same planning output over and over. Model is kimi k2 and to some degree Gemini 2,5 Pro / DeepSeek. Also DeepSeek weren't even able to see files in Project directory resulting in [use the bash tool to list files in the data folder.
rekram1-node commentedon Sep 24, 2025
Is anyone still having this issue?
laszloagardi commentedon Sep 29, 2025
I am having a similar issue. Model tries to edit files, I have to manually approve edits (persmission is set to
ask). As it spits out multiple few-line edits to the same file, I approve one-by-one (enter), but seems like only the last edit is applied.Last tried with openrouter grok-4-fast:free.
rekram1-node commentedon Sep 29, 2025
hmm @laszloagardi are u on latest?
laszloagardi commentedon Sep 29, 2025
yes, v0.13.4
rekram1-node commentedon Sep 29, 2025
got it I will try to replicate
R44VC0RP commentedon Jan 25, 2026
Closing this as stale - the original reports from Aug-Sep 2025 couldn't be reproduced, and we've since added fixes:
For those who were using GPT-5-chat-latest: That model doesn't support tool calling, which is required for file edits. Use a model that supports tools (Claude, GPT-4o, etc.).
If you're still experiencing edit issues on current versions (v1.1.x): Please open a new issue with:
opencode --version)Thanks everyone for the reports!