add possibility to define a project based instruction file

README.md

@@ -61,6 +61,7 @@ If you like this project, please consider supporting me on Patreon, as it helps
 
 - **AI-Powered Code Assistance**: Interact with AI to ask questions about your current code file and receive intelligent suggestions for improvement or modification.
 - **One-Click Application**: Quickly apply the AI's suggested changes to your source code with a single command, streamlining the editing process and saving time.
+- **Project-Specific Instruction Files**: Customize AI behavior by adding a markdown file (`avante.md` by default) in the project root. This file is automatically referenced during workspace changes. You can also configure a custom file name for tailored project instructions.
 
 ## Installation
 
@@ -84,6 +85,8 @@ For building binary if you wish to build from source, then `cargo` is required.
   ---@type avante.Config
   opts = {
     -- add any opts here
+    -- this file can containe specific instructions for your project
+    instructions_file = "avante.md"
     -- for example
     provider = "claude",
     providers = {
@@ -656,6 +659,7 @@ For other users just add a custom provider
 Avante.nvim provides several completion sources that can be integrated with blink.cmp:
 
 #### Mentions (`@` trigger)
+
 Mentions allow you to quickly reference specific features or add files to the chat context:
 
 - `@codebase` - Enable project context and repository mapping
@@ -665,6 +669,7 @@ Mentions allow you to quickly reference specific features or add files to the ch
 - `@buffers` - Add open buffers to chat context
 
 #### Slash Commands (`/` trigger)
+
 Built-in slash commands for common operations:
 
 - `/help` - Show help message with available commands
@@ -676,6 +681,7 @@ Built-in slash commands for common operations:
 - `/commit` - Generate commit message for changes
 
 #### Shortcuts (`#` trigger)
+
 Shortcuts provide quick access to predefined prompt templates. You can customize these in your config:
 
 ```lua
@@ -983,6 +989,7 @@ Fast Apply is a feature that enables instant code edits with high accuracy by le
 Fast Apply addresses the common pain point of slow code application in AI-assisted development. Instead of waiting for a full language model to process and apply changes, Fast Apply uses a specialized "apply model" that can quickly and accurately merge code edits with 96-98% accuracy at speeds of 2500-4500+ tokens per second.
 
 Key benefits:
+
 - **Instant application**: Code changes are applied immediately without noticeable delays
 - **High accuracy**: Specialized models achieve 96-98% accuracy for code edits
 - **Seamless workflow**: Maintains the natural flow of development without interruptions
@@ -993,6 +1000,7 @@ Key benefits:
 To enable Fast Apply, you need to:
 
 1. **Enable Fast Apply in your configuration**:
+
    ```lua
      behaviour = {
        enable_fastapply = true,  -- Enable Fast Apply feature
@@ -1004,6 +1012,7 @@ To enable Fast Apply, you need to:
    Go to [morphllm.com](https://morphllm.com/api-keys) and create an account and get the API key.
 
 3. **Set your Morph API key**:
+
    ```bash
    export MORPH_API_KEY="your-api-key"
    ```
@@ -1021,11 +1030,11 @@ To enable Fast Apply, you need to:
 
 Morph provides different models optimized for different use cases:
 
-| Model | Speed | Accuracy | Context Limit |
+| Model            | Speed             | Accuracy | Context Limit |
-|-------|-------|----------|---------------|
+| ---------------- | ----------------- | -------- | ------------- |
-| `morph-v3-fast` | 4500+ tok/sec | 96% | 16k tokens |
+| `morph-v3-fast`  | 4500+ tok/sec     | 96%      | 16k tokens    |
-| `morph-v3-large` | 2500+ tok/sec | 98% | 16k tokens |
+| `morph-v3-large` | 2500+ tok/sec     | 98%      | 16k tokens    |
-| `auto` | 2500-4500 tok/sec | 98% | 16k tokens |
+| `auto`           | 2500-4500 tok/sec | 98%      | 16k tokens    |
 
 ### How It Works
 
@@ -1037,6 +1046,7 @@ When Fast Apply is enabled and a Morph provider is configured, avante.nvim will:
 4. Apply the changes directly to your files with high accuracy
 
 The process uses a specialized prompt format that includes:
+
 - `<instructions>`: Clear description of what changes to make
 - `<code>`: The original code content
 - `<update>`: The specific changes using truncation markers (`// ... existing code ...`)
@@ -1387,6 +1397,7 @@ Avante.nvim can be extended to work with other plugins by using its extension mo
 ### How to disable agentic mode?
 
 Avante.nvim provides two interaction modes:
+
 - **`agentic`** (default): Uses AI tools to automatically generate and apply code changes
 - **`legacy`**: Uses the traditional planning method without automatic tool execution
 
@@ -1400,10 +1411,12 @@ To disable agentic mode and switch to legacy mode, update your configuration:
 ```
 
 **What's the difference?**
+
 - **Agentic mode**: AI can automatically execute tools like file operations, bash commands, web searches, etc. to complete complex tasks
 - **Legacy mode**: AI provides suggestions and plans but requires manual approval for all actions
 
 **When should you use legacy mode?**
+
 - If you prefer more control over what actions the AI takes
 - If you're concerned about security with automatic tool execution
 - If you want to manually review each step before applying changes

lua/avante/config.lua

@@ -17,6 +17,9 @@ local Utils = require("avante.utils")
 
 ---@class avante.CoreConfig: avante.Config
 local M = {}
+
+--- Default configuration for project-specific instruction file
+M.instructions_file = "avante.md"
 ---@class avante.Config
 M._defaults = {
   debug = false,

lua/avante/llm.lua

@@ -8,6 +8,7 @@ local Utils = require("avante.utils")
 local Prompts = require("avante.utils.prompts")
 local Config = require("avante.config")
 local Path = require("avante.path")
+local PPath = require("plenary.path")
 local Providers = require("avante.providers")
 local LLMToolHelpers = require("avante.llm_tools.helpers")
 local LLMTools = require("avante.llm_tools")
@@ -255,6 +256,17 @@ end
 ---@param opts AvanteGeneratePromptsOptions
 ---@return AvantePromptOptions
 function M.generate_prompts(opts)
+  local project_instruction_file = Config.instructions_file or "avante.md"
+  local project_root = Utils.root.get()
+  local instruction_file_path = PPath:new(project_root, project_instruction_file)
+
+  if instruction_file_path:exists() then
+    local lines = Utils.read_file_from_buf_or_disk(instruction_file_path:absolute())
+    local instruction_content = lines and table.concat(lines, "\n") or ""
+
+    if instruction_content then opts.instructions = (opts.instructions or "") .. "\n" .. instruction_content end
+  end
+
   local provider = opts.provider or Providers[Config.provider]
   local mode = opts.mode or Config.mode
 
@@ -264,7 +276,6 @@ function M.generate_prompts(opts)
     image_paths = vim.list_extend(image_paths, opts.prompt_opts.image_paths)
   end
 
-  local project_root = Utils.root.get()
   Path.prompts.initialize(Path.prompts.get_templates_dir(project_root), project_root)
 
   local system_info = Utils.get_system_info()

tests/llm_spec.lua

@@ -0,0 +1,115 @@
+local utils = require("avante.utils")
+local PPath = require("plenary.path")
+
+local llm = require("avante.llm")
+
+describe("generate_prompts", function()
+  local project_root = "/tmp/project_root"
+
+  before_each(function()
+    local mock_dir = PPath:new("tests", project_root)
+    mock_dir:mkdir({ parents = true })
+
+    local mock_file = PPath:new("tests", project_root, "avante.md")
+    mock_file:write("# Mock Instructions\nThis is a mock instruction file.", "w")
+
+    -- Mock the project root
+    utils.root = {}
+    utils.root.get = function() return mock_dir end
+
+    -- Mock Config.providers
+    local Config = require("avante.config")
+    Config.instructions_file = "avante.md"
+    Config.provider = "openai"
+    Config.providers = {
+      openai = {
+        endpoint = "https://api.mock.com/v1",
+        model = "gpt-mock",
+        timeout = 10000,
+        context_window = 1000,
+        extra_request_body = {
+          temperature = 0.5,
+          max_tokens = 1000,
+        },
+      },
+    }
+    -- Mock Config.history to prevent nil access error in Path.setup()
+    Config.history = {
+      max_tokens = 4096,
+      carried_entry_count = nil,
+      storage_path = "/tmp/test_avante_history",
+      paste = {
+        extension = "png",
+        filename = "pasted-%Y-%m-%d-%H-%M-%S",
+      },
+    }
+
+    -- Mock Config.behaviour
+    Config.behaviour = {
+      auto_focus_sidebar = true,
+      auto_suggestions = false, -- Experimental stage
+      auto_suggestions_respect_ignore = false,
+      auto_set_highlight_group = true,
+      auto_set_keymaps = true,
+      auto_apply_diff_after_generation = false,
+      jump_result_buffer_on_finish = false,
+      support_paste_from_clipboard = false,
+      minimize_diff = true,
+      enable_token_counting = true,
+      use_cwd_as_project_root = false,
+      auto_focus_on_diff_view = false,
+      auto_approve_tool_permissions = false, -- Default: show permission prompts for all tools
+      auto_check_diagnostics = true,
+      enable_fastapply = false,
+    }
+
+    -- Mock Config.rules to prevent nil access error in get_templates_dir()
+    Config.rules = {
+      project_dir = nil,
+      global_dir = nil,
+    }
+
+    -- Mock P.available to always return true
+    local Path = require("avante.path")
+    Path.available = function() return true end
+
+    -- Mock the Prompt functions directly since _templates_lib is a local variable
+    -- that we can't easily access from outside the module
+    Path.prompts.initialize = function(cache_directory, project_directory)
+      -- Mock initialization - no-op for tests
+    end
+
+    Path.prompts.render_file = function(path, opts)
+      -- Mock render - return empty string for tests
+      return ""
+    end
+
+    Path.prompts.render_mode = function(mode, opts)
+      -- Mock render_mode - return empty string for tests
+      return ""
+    end
+
+    Path.setup() -- Initialize necessary paths like cache_path
+  end)
+
+  after_each(function()
+    -- Clean up created test files and directories
+    local mock_dir = PPath:new("tests", project_root)
+    if mock_dir:exists() then mock_dir:rmdir() end
+  end)
+
+  it("should include instruction file content when the file exists", function()
+    local opts = {}
+    llm.generate_prompts(opts)
+    assert.are.same("\n# Mock Instructions\nThis is a mock instruction file.", opts.instructions)
+  end)
+
+  it("should not modify instructions if the file does not exist", function()
+    local mock_file = PPath:new("tests", project_root, "avante.md")
+    if mock_file:exists() then mock_file:rm() end
+
+    local opts = {}
+    llm.generate_prompts(opts)
+    assert.are.same(opts.instructions, nil)
+  end)
+end)