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
Agent unable to save due to malformed write payload #12412
Started a new project and have been trying to formulate a proper Plan (as people always suggest). I first made a PLAN.md from another agent. That is the only file in the project, other than .opencode having been created by opencode.
Upon attempting to save the plan (to .opencode/plan/****.md) it fails every time. Somehow the payload to the write tool is always malformed.
It seems to try to indicate the filePath twice, and the second instance is escaped improperly. <- causing the JSON parse error
The path it was intending to write to is incorrect. My working directory for the project is /home/amatecha/git/myproject (and that is where I launched opencode from), but the logged filePath is /home/amatecha/.opencode/plans/gateway-development-plan.md for some reason. It should have been /home/amatecha/git/myproject/.opencode/plans/gateway-development-plan.md
(note I replaced the plan text with "exampletext" above. The length of the string previously there was 16663 bytes, if that's relevant)
Installed opencode via shell script install method. I have the following feature flag in my .bashrc:
export OPENCODE_EXPERIMENTAL_PLAN_MODE=1
I'm using qwen3-coder-next via llama.cpp if that's relevant.
Plugins
none that I know of
OpenCode version
1.1.53 (happened in 1.1.51 as well)
Steps to reproduce
1.have Experimental Plan mode enabled
2. create a new dir
3. cd into dir
4. run opencode
5. go to Plan mode and come up with a quick plan
6. tell the agent to save the plan and confirm whatever filename it suggests
7. note the failure to save due to above-described error
I got it working with this reverse proxy which I wrote some time ago to connect a streaming client to llama-server when it wasn't able to stream when tool calling was used. https://github.com/crashr/llama-stream
changed the title [-]Plan mode (experimental) unable to save due to malformed write payload[/-][+]Agent unable to save due to malformed write payload[/+]on Feb 9, 2026
Activity
github-actions commentedon Feb 6, 2026
This issue might be a duplicate of existing issues. Please check:
Feel free to ignore if none of these address your specific case.
amatecha commentedon Feb 6, 2026
Issue also being reported on the model page at https://huggingface.co/unsloth/Qwen3-Coder-Next-GGUF/discussions/5 and https://huggingface.co/unsloth/Qwen3-Coder-Next-GGUF/discussions/2 , may be something wrong with the model itself? Even on the original Qwen model, https://huggingface.co/Qwen/Qwen3-Coder-Next/discussions/14 /shrug
crashr commentedon Feb 6, 2026
ggml-org/llama.cpp#19382 (comment)
crashr commentedon Feb 6, 2026
I got it working with this reverse proxy which I wrote some time ago to connect a streaming client to llama-server when it wasn't able to stream when tool calling was used.
https://github.com/crashr/llama-stream
[-]Plan mode (experimental) unable to save due to malformed write payload[/-][+]Agent unable to save due to malformed write payload[/+]github-actions commentedon Apr 11, 2026
To stay organized issues are automatically closed after 90 days of no activity. If the issue is still relevant please open a new one.