Toggle off display of prompts sent on each request

[ruicsh]

hey,

I have a bunch of prompts that get automatically added to any chat session based on the source buffer filetype and action chosen by keymaps. For example, it can be:

> /react
> /aggrid
> /fix

each of these are custom prompts I load from files on my nvim's config. Each prompt is around 3 or 4 paragraphs of text. With this new version of CopilotChat, the contents of these prompts get displayed on the chat buffer. I understand the idea: it's transparent what is really being sent down with the request. But, it gets repetitive to always see the same blob of text (my loaded sticky prompts) on the chat buffer and having to scroll down to see the actual info you're looking for - the LLMs response. For me, the repeated full text prompts are kind of useless info - I just want to make sure they are added to the request, seeing the sticky prompts headers is enough.

Would it be possible to toggle this display off? Either by a static config option or a parameter into chat.open or chat.ask?

Cheers

[deathbeam]

Well its not only for display and clarity, the chat buffer is now also synced with the actual stuff that is being sent to the LLM, so just hiding it isnt as simple as it sounds. Its definitely something I want to look at in future tho and maybe dynamically resolving stuff for each message, only downside of this is performance obviously, and just expanding the full message every time is fastest and least error prone

[ruicsh]

I found an ok solution: close the current fold when the response is received.

options = {
  callback = function()
    -- Close the current fold (the user's prompt)
    vim.cmd("normal! zc")
    -- Scroll to the bottom of the chat window
    vim.api.nvim_win_set_cursor(0, { vim.api.nvim_buf_line_count(0), 0 })
  end
}

A complete solution would be if I could close the fold when the request starts or when I submit the prompt. Can I do that?

I tried this, to have all folds closed by default, but it didn't work:

vim.api.nvim_create_autocmd("BufEnter", {
  pattern = "copilot-*",
  callback = function()
    vim.opt_local.foldlevel = 0
    vim.opt_local.foldlevelstart = 0
  end,
})

[deathbeam]

I found an ok solution: close the current fold when the response is received.

options = {
callback = function()
-- Close the current fold (the user's prompt)
vim.cmd("normal! zc")
-- Scroll to the bottom of the chat window
vim.api.nvim_win_set_cursor(0, { vim.api.nvim_buf_line_count(0), 0 })
end
}

A complete solution would be if I could close the fold when the request starts or when I submit the prompt. Can I do that?

I tried this, to have all folds closed by default, but it didn't work:

vim.api.nvim_create_autocmd("BufEnter", {
pattern = "copilot-*",
callback = function()
vim.opt_local.foldlevel = 0
vim.opt_local.foldlevelstart = 0
end,
})

that is actually really good idea, i like it a lot. implemented something for it

[deathbeam]

Now there is auto_fold config (disabled by default) but it automatically folds non-assistant messages. That includes tool responses so its a bit more complete solution and can let you focus on the assistant responses only.

[ruicsh]

glad you like it, but is it working for you? For me, I see it closing folds at random almost. In this screenshot, the first fold is open (the user's first prompt hi with all my auto system-prompts), then the next 2 are closed (1 assistant - the response, and 1 user - waiting for a new prompt). This is how it lands on a simple question/response chat.

my config in case it helps, but it's just the new auto_fold

[deathbeam]

for me its working but saw it misbehave few times as well since i started using it, will take a look tomorrow i guess

EDIT: Also can you try without the fold adjustments maybe in your config