FEATURE: allow option out of context tips

[SamSaffron]

Currently after the system message we inject the "context tip"

CopilotChat.nvim/lua/CopilotChat/client.lua

Lines 208 to 239 in 5df0b66

	if contexts and not vim.tbl_isempty(contexts) then
	local help_text = [[When you need additional context, request it using this format:
	> #<command>:`<input>`
	Examples:
	> #file:`path/to/file.js` (loads specific file)
	> #buffers:`visible` (loads all visible buffers)
	> #git:`staged` (loads git staged changes)
	> #system:`uname -a` (loads system information)
	Guidelines:
	- Always request context when needed rather than guessing about files or code
	- Use the > format on a new line when requesting context
	- Output context commands directly - never ask if the user wants to provide information
	- Assume the user will provide requested context in their next response
	Available context providers and their usage:]]
	local context_names = vim.tbl_keys(contexts)
	table.sort(context_names)
	for _, name in ipairs(context_names) do
	local description = contexts[name]
	description = description:gsub('\n', '\n ')
	help_text = help_text .. '\n\n - #' .. name .. ': ' .. description
	end
	if system_prompt ~= '' then
	system_prompt = system_prompt .. '\n\n'
	end
	system_prompt = system_prompt .. help_text
	end

I would like to disable the tip on my system cause it just leads the LLMs to often hallucinate needing more context and gets into very annoying loops especially when it has the current git diff in the context.

Any objection to a PR that makes context tip optional?

[SamSaffron]

Made this PR in case we are ok with it: #1174

[deathbeam]

https://github.com/CopilotC-Nvim/CopilotChat.nvim/blob/main/lua/CopilotChat/config.lua#L28 this is already suppported?

[SamSaffron]

I think that is different, that means contexts are no longer included, the new setting just controls how the system prompt is formatted

…

On Mon, 23 Jun 2025 at 9:01 pm, Tomas Slusny ***@***.***> wrote: *deathbeam* left a comment (CopilotC-Nvim/CopilotChat.nvim#1173) <#1173 (comment)> https://github.com/CopilotC-Nvim/CopilotChat.nvim/blob/main/lua/CopilotChat/config.lua#L28 this is already suppported? — Reply to this email directly, view it on GitHub <#1173 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAABIXIG6HKKS3SMF4PZKU33E7NBNAVCNFSM6AAAAAB74TJQL2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDSOJVHE3TCMZTG4> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[deathbeam]

It achieves same thing, e.g the contexts being known to the LLM, your config justdoes it differently. I think the current setting is good enough, and the #1029 solves this in different way where all contexts are optional by default, closing this

[SamSaffron]

I am confused here @deathbeam , I disabled include_contexts_in_prompt yet in CopilotChat.log I still see:

Available context providers and their usage:

 - #buffer: Includes specified buffer in chat context. Supports input (default current).

 - #buffers: Includes all buffers in chat context. Supports input (default listed).

 - #file: Includes content of provided file in chat context. Supports input.

 - #filenames: Includes names of all non-hidden files in the current workspace in chat context. Supports input (glob pattern).

 - #files: Includes all non-hidden files in the current workspace in chat context. Supports input (glob pattern).

 - #git: Requires `git`. Includes current git diff in chat context. Supports input (default unstaged, also accepts commit number).

 - #gitmain: Get diff against main branch

 - #quickfix: Includes quickfix list file contents in chat context.

 - #register: Includes contents of register in chat context. Supports input (default +, e.g clipboard).

 - #system: Includes output of provided system shell command in chat context. Supports input.
   
   Important:
   - Only use system commands as last resort, they are run every time the context is requested.
   - For example instead of curl use the url context, instead of finding and grepping try to check if there is any context that can query the data you need instead.
   - If you absolutely need to run a system command, try to use read-only commands and avoid commands that modify the system state.
   

 - #url: Includes content of provided URL in chat context. Supports input.

The setting is about removing that text from the prompt.

Another alternative might be to amend it so system_prompt is constructed using placeholders eg:

You are a helpful bot.

{{MEMORY}}
{{CONTEXT_TIP_PLACEHOLDER}}

Currently we sort of have control over system prompt but don't really have control.

---

Happy about tools.

[deathbeam]

I am confused here @deathbeam , I disabled include_contexts_in_prompt yet in CopilotChat.log I still see:

Available context providers and their usage:

 - #buffer: Includes specified buffer in chat context. Supports input (default current).

 - #buffers: Includes all buffers in chat context. Supports input (default listed).

 - #file: Includes content of provided file in chat context. Supports input.

 - #filenames: Includes names of all non-hidden files in the current workspace in chat context. Supports input (glob pattern).

 - #files: Includes all non-hidden files in the current workspace in chat context. Supports input (glob pattern).

 - #git: Requires `git`. Includes current git diff in chat context. Supports input (default unstaged, also accepts commit number).

 - #gitmain: Get diff against main branch

 - #quickfix: Includes quickfix list file contents in chat context.

 - #register: Includes contents of register in chat context. Supports input (default +, e.g clipboard).

 - #system: Includes output of provided system shell command in chat context. Supports input.
   
   Important:
   - Only use system commands as last resort, they are run every time the context is requested.
   - For example instead of curl use the url context, instead of finding and grepping try to check if there is any context that can query the data you need instead.
   - If you absolutely need to run a system command, try to use read-only commands and avoid commands that modify the system state.
   

 - #url: Includes content of provided URL in chat context. Supports input.

The setting is about removing that text from the prompt.

Another alternative might be to amend it so system_prompt is constructed using placeholders eg:

You are a helpful bot.

{{MEMORY}}
{{CONTEXT_TIP_PLACEHOLDER}}

Currently we sort of have control over system prompt but don't really have control.

Happy about tools.

I just disabled it and it works fine:

[CopilotChat] [DEBUG 02:06:38] /home/deathbeam/.config/nvim/pack/bundle/start/CopilotChat.nvim/lua/CopilotChat/client.lua:283: System prompt:
[CopilotChat]  You are a code-focused AI programming assistant that specializes in practical software engineering solutions.
[CopilotChat] When asked for your name, you must respond with "GitHub Copilot".
[CopilotChat] Follow the user's requirements carefully & to the letter.
[CopilotChat] Follow Microsoft content policies.
[CopilotChat] Avoid content that violates copyrights.
[CopilotChat] If you are asked to generate content that is harmful, hateful, racist, sexist, lewd, violent, or completely irrelevant to software engineering, only respond with "Sorry, I can't assist with that."
[CopilotChat] Keep your answers short and impersonal.
[CopilotChat] The user works in an IDE called Neovim which has a concept for editors with open files, integrated unit test support, an output pane that shows the output of running the code as well as an integrated terminal.
[CopilotChat] The user is working on a Linux machine. Please respond with system specific commands if applicable.
[CopilotChat] You will receive code snippets that include line number prefixes - use these to maintain correct position references but remove them when generating output.
[CopilotChat] 
[CopilotChat] When presenting code changes:
[CopilotChat] 
[CopilotChat] 1. For each change, first provide a header outside code blocks with format:
[CopilotChat]    [file:<file_name>](<file_path>) line:<start_line>-<end_line>
[CopilotChat] 
[CopilotChat] 2. Then wrap the actual code in triple backticks with the appropriate language identifier.
[CopilotChat] 
[CopilotChat] 3. Keep changes minimal and focused to produce short diffs.
[CopilotChat] 
[CopilotChat] 4. Include complete replacement code for the specified line range with:
[CopilotChat]    - Proper indentation matching the source
[CopilotChat]    - All necessary lines (no eliding with comments)
[CopilotChat]    - No line number prefixes in the code
[CopilotChat] 
[CopilotChat] 5. Address any diagnostics issues when fixing code.
[CopilotChat] 
[CopilotChat] 6. If multiple changes are needed, present them as separate blocks with their own headers.

Config i used:

chat.setup({
    model = 'gpt-4.1',
    debug = true,
    include_contexts_in_prompt = false,
})

So it definitely works

[SamSaffron]

oh my, I see it, not sure what I was doing wrong here, confirm the setting is working as expected, sorry for taking up your time on this.