bug: @this and @buffer reads file instead of current buffer

[yihao03]

Did you check docs and existing issues?

• I have read the plugin docs
• I have read the opencode docs
• I have searched the existing plugin issues
• I have searched the existing opencode issues
• I am using the latest version of the plugin

:checkhealth opencode output

============================================================================== opencode: ✅ opencode.nvim ~ - OS: Linux 6.6.87.2-microsoft-standard-WSL2 (x86_64) - nvim version: 0.12.0-dev+gf847aa6208. - opencode.nvim git commit hash: 1088ee70dd997d785a1757d351c07407f0abfc9f. - vim.g.opencode_opts: {} opencode.nvim [binaries] ~ - ✅ OK opencode available with version 1.2.26. - ✅ OK curl available. - ✅ OK pgrep available. - ✅ OK lsof available. opencode.nvim [snacks] ~ - ✅ OK snacks.input is enabled: ask() will be enhanced. - ✅ OK snacks.picker is enabled: select() will be enhanced.

Describe the bug

When using @this and @buffer commands, one would expect that the exact content in the buffer is being read. However if it is a new change that has yet to be saved, opencode wouldn't be able to see the content and be confused.

Steps To Reproduce

1. create a new file and write something to it (do not save)
2. prompt opencode with a @this

Expected Behavior

OpenCode sees what's in the buffer and answer accordingly

[nickjvandyke]

Yeah very fair - I have thought to @buffer to @file to clear up this confusion. Thoughts on that? Would @files then be clear, that it sends the file corresponding to each open buffer? Since Neovim doesn't necessarily have a concept of "open files", just "open buffers".

Alternatively we could of course just send the buffer content, but I think the LLM gives a better response when we send just a location and let it gather surrounding context. At least, IME. Open to others!

[yihao03]

I thought sending the buffer content is the easiest way too. But then would it confuse the LLMs when they search the file and see something different?

Another workaround could be just save the file on query.

[nickjvandyke]

But then would it confuse the LLMs when they search the file and see something different?

Ah yeah good point, it might still read the file. And will definitely error when it attempts to edit it.

Another workaround could be just save the file on query.

Perhaps, but I think users might not expect that. Less intrusive to let them intentionally add a save to their prompt keymaps or something.