fix: add ignore_files to config

Reviewed-on: #2

.chatgpt_config.yaml

@@ -1,17 +0,0 @@
-project_name: "chatgpt_nvim"
-default_prompt_blocks:
-  - "basic-prompt"
-  - "workflow-prompt"
-directories:
-  - "."
-initial_files:
-  - "README.md"
-debug: false
-preview_changes: false
-interactive_file_selection: false
-partial_acceptance: false
-improved_debug: false
-enable_debug_commands: true
-token_limit: 128000
-enable_chunking: false
-enable_step_by_step: true

README.md

@@ -4,28 +4,36 @@
 This plugin integrates a ChatGPT O1 model workflow into Neovim. It allows you to:
 
 1. Generate prompts containing:
-   - An **initial prompt** (from `.chatgpt_config.yaml`)
-   - A list of directories (also specified in `.chatgpt_config.yaml`) from which it gathers the project structure and file contents
-   - **Interactive file selection** if enabled, so you can pick exactly which directories to include
-   - Any **initial files** you define (e.g., `README.md`, etc.)
+   - An **initial prompt** (loaded from `.chatgpt_config.yaml` or `chatgpt_config.yaml`)
+   - A list of directories (as specified in your configuration) from which it gathers the project structure
+   - **Interactive file selection** (if enabled) so you can choose which directories to include
+   - Any **initial files** you define (e.g. `README.md`, etc.)
+   - Optionally, **file contents** can be appended to the prompt when the new `include_file_contents` flag is enabled
 
 2. Copy these prompts to your clipboard to paste into ChatGPT O1.
-3. Receive YAML changes from ChatGPT, then run `:ChatGPTPaste` to apply them or supply additional files.
+3. Receive YAML changes from ChatGPT and run `:ChatGPTPaste` to apply them or to supply additional files.
 
 ## New Key Features
 
 - **Step-by-Step Prompting** (`enable_step_by_step: true`):  
-  If the request grows too large (exceeds `token_limit`), the plugin automatically generates a special prompt asking the model to split the task into smaller steps, working through them one by one. This approach helps you stay within the model’s maximum token limit without having to manually break things apart.
+  If the generated prompt grows too large (exceeding the configured `prompt_char_limit`), the plugin automatically produces a special prompt asking the model to split the task into smaller steps. This ensures you remain within the model’s maximum token capacity without manually breaking your task apart.
 
-- **Partial Acceptance**: If `partial_acceptance: true`, you can open a buffer that lists the final changes. Remove or comment out lines you don’t want, then only those changes are applied.
+- **Partial Acceptance**:  
+  If `partial_acceptance: true`, you can open a buffer displaying the proposed changes. Remove or comment out lines you do not wish to apply, and only the remaining changes will be executed.
 
-- **Preview Changes**: If `preview_changes: true`, you get a buffer showing proposed changes before you apply them.
+- **Preview Changes**:  
+  If `preview_changes: true`, a buffer will show the proposed changes before they are applied.
 
-- **Interactive File Selection**: If `interactive_file_selection: true`, you choose which directories from `.chatgpt_config.yaml` get included in the prompt, reducing token usage.
+- **Interactive File Selection**:  
+  When `interactive_file_selection: true`, you can choose which directories (from your config file) to include in the prompt, thus managing token usage more effectively.
 
-- **Improved Debug**: If `improved_debug: true`, debug logs go into a dedicated `ChatGPT_Debug_Log` buffer for easier reading.
+- **Include File Contents**:  
+  A new configuration flag `include_file_contents` (default: `false`) lets you include the entire contents of the project files in the prompt. When enabled, the plugin gathers and appends the file contents from the selected directories. It counts the total characters and, if the combined file contents exceed the `prompt_char_limit`, it notifies you to disable this feature to avoid exceeding the model’s capacity.
 
-## Example `.chatgpt_config.yaml`
+- **Improved Debug**:  
+  If `improved_debug: true`, detailed debug logs are sent to a dedicated `ChatGPT_Debug_Log` buffer for easier review.
+
+## Example `.chatgpt_config.yaml` or `chatgpt_config.yaml`
 
 ```yaml
 project_name: "chatgpt_nvim"
@@ -37,40 +45,43 @@ directories:
 initial_files:
   - "README.md"
 debug: false
+include_file_contents: false  # Set to true to include file contents in the prompt
 enable_step_by_step: true
 preview_changes: true
 interactive_file_selection: true
 partial_acceptance: true
 improved_debug: true
-token_limit: 3000
+prompt_char_limit: 300000  # Maximum characters allowed in the prompt
 ```
 
 ## Usage
 
 1. **`:ChatGPT`**  
-   - If `interactive_file_selection` is on, you’ll pick directories to include in a buffer named `ChatGPT_File_Selection`.
-   - Save & close with `:wq`, `:x`, or `:bd` (you don’t have to use `:q`).
-   - If `enable_step_by_step` is on and the prompt might exceed `token_limit`, the plugin will generate instructions prompting the model to address each step separately.
+   - If `interactive_file_selection` is enabled, you will select directories via a buffer (named `ChatGPT_File_Selection`).
+   - Save and close the buffer with `:wq`, `:x`, or `:bd` (no need to use `:q`).
+   - If `enable_step_by_step` is active and the prompt might exceed `prompt_char_limit`, the plugin will generate a step-by-step instruction prompt.
 
 2. **Paste Prompt to ChatGPT**  
-   - If the task is split into steps, simply copy/paste them one by one into ChatGPT.
+   - If the task is split into steps, copy and paste them one by one into ChatGPT.
 
 3. **`:ChatGPTPaste`**  
-   - The plugin reads the YAML from your clipboard. If it requests more files, it might again suggest a step-by-step approach.  
-   - If final changes are provided:
-     - Optionally preview them (`preview_changes`).  
-     - Optionally partially accept them (`partial_acceptance`).  
-     - Then the plugin writes/deletes files as specified.
+   - The plugin reads the YAML response from your clipboard. If additional files are requested, a step-by-step approach might be suggested.
+   - When final changes are provided:
+     - You can optionally preview them (`preview_changes`).
+     - You can optionally partially accept them (`partial_acceptance`).
+     - Finally, the plugin applies file writes or deletions as specified in the YAML.
 
 ## Troubleshooting & Tips
-- Adjust `token_limit` in `.chatgpt_config.yaml` as needed.
-- If partial acceptance is confusing, remember to remove or prepend `#` to lines you don’t want before saving and closing the buffer.
-- If step-by-step prompting occurs, ensure you follow each prompt the model provides in the correct order.  
-- Check `ChatGPT_Debug_Log` if `improved_debug` is on, or the Neovim messages if `debug` is on, for detailed info.
-- You can close the selection or prompt buffers at any time with commands like `:bd`, `:x`, or `:wq`. No need to rely on `:q`.
+
+- Adjust `prompt_char_limit` in your configuration file as needed.
+- If partial acceptance is confusing, simply remove or comment out lines you do not want before finalizing the buffer.
+- When step-by-step prompting occurs, ensure you follow each instruction in the correct order.
+- For detailed debug information, check the `ChatGPT_Debug_Log` buffer (if `improved_debug` is enabled) or the standard Neovim messages (if `debug` is enabled).
+- You can close the selection or prompt buffers at any time using commands like `:bd`, `:x`, or `:wq`.
 
 ## Debug Commands
-If `enable_debug_commands` is true, you can include commands like these in your YAML:
+
+If `enable_debug_commands` is true, you can include commands such as:
 ```yaml
 commands:
   - command: "list"
@@ -80,6 +91,23 @@ commands:
     pattern: "searchString"
     target: "path/to/file/or/directory"
 ```
-The **list** command now uses the Linux `ls` command to list directory contents. The **grep** command searches for a given pattern in a file or all files in a directory.
+The **list** command uses the Linux `ls` command to list directory contents, while the **grep** command searches for a given pattern within files or directories.
 
-Enjoy your improved, more flexible ChatGPT Neovim plugin with step-by-step support!
+Enjoy your improved, flexible ChatGPT NeoVim plugin with step-by-step support and enhanced file inclusion capabilities!
+
+## Test when Developing
+
+- **Add plugin to runtimepath:**
+  ```vim
+  :set rtp^=~/temp_plugins/chatgpt.vim
+  ```
+
+- **Clear Lua module cache:**
+  ```vim
+  :lua for k in pairs(package.loaded) do if k:match("^chatgpt_nvim") then package.loaded[k]=nil end end
+  ```
+
+- **Load new plugin code:**
+  ```vim
+  :runtime plugin/chatgpt.vim
+  ```

chatgpt_config.yaml

@@ -0,0 +1,29 @@
+project_name: "chatgpt.vim"
+default_prompt_blocks:
+  - "basic"
+  - "secure-coding"
+
+ignore_files:
+  - "node_modules/"
+  - "*.log"
+  - "vendor/"
+
+include_file_contents: true
+
+debug: false
+improved_debug: false
+
+preview_changes: false
+interactive_file_selection: false
+partial_acceptance: false
+
+enable_debug_commands: true
+enable_chunking: false
+enable_step_by_step: true
+auto_lint: true
+
+tool_auto_accept:
+  read_file: true
+  edit_file: true
+  replace_in_file: true
+  execute_command: false

lua/chatgpt_nvim/config.lua

@@ -4,6 +4,7 @@ local uv = vim.loop
 local ok_yaml, lyaml = pcall(require, "lyaml")
 local prompts = require("chatgpt_nvim.prompts")
 
+-- Determines the Git root or fallback directory.
 local function get_project_root()
   local current_file = vim.fn.expand("%:p")
   local root_dir
@@ -28,37 +29,74 @@ local function get_project_root()
   return root_dir
 end
 
+-- Attempt to locate either .chatgpt_config.yaml or chatgpt_config.yaml, returning whichever exists first.
 local function get_config_path()
   local root = get_project_root()
-  return root .. "/.chatgpt_config.yaml"
+  local dot_config = root .. "/.chatgpt_config.yaml"
+  local non_dot_config = root .. "/chatgpt_config.yaml"
+
+  local dot_fd = uv.fs_open(dot_config, "r", 438)
+  if dot_fd then
+    uv.fs_close(dot_fd)
+    return dot_config
+  end
+
+  local non_dot_fd = uv.fs_open(non_dot_config, "r", 438)
+  if non_dot_fd then
+    uv.fs_close(non_dot_fd)
+    return non_dot_config
+  end
+
+  return nil
 end
 
 function M.load()
+  -- Attempt to find either config file, else fallback
   local path = get_config_path()
-  local fd = uv.fs_open(path, "r", 438)
   local config = {
     initial_prompt = "",
     directories = { "." },
     default_prompt_blocks = {},
-    -- Changed default from 128000 to 16384 as requested
-    prompt_char_limit = 300000,
+    -- Removed prompt_char_limit
     project_name = "",
     debug = false,
     initial_files = {},
+    include_file_contents = false, -- NEW FLAG: include file contents in prompt if true
     preview_changes = false,
     interactive_file_selection = false,
     partial_acceptance = false,
     improved_debug = false,
     enable_chunking = false,
-    -- New default for step-by-step
     enable_step_by_step = true,
-    enable_debug_commands = false
+
+    -- If auto_lint is true, we only do LSP-based checks.
+    auto_lint = false,
+
+    tool_auto_accept = {
+      read_file = false,
+      edit_file = false,
+      replace_in_file = false,
+      execute_command = false,
+    }
   }
 
+  -- If no config file found, alert user and return default config
+  if not path then
+    config.initial_prompt = "You are a coding assistant who receives a project's context and user instructions..."
+    vim.notify(
+      "No config file found (tried .chatgpt_config.yaml, chatgpt_config.yaml). Using defaults.",
+      vim.log.levels.WARN
+    )
+    config.max_token = 2048
+    return config
+  end
+
+  local fd = uv.fs_open(path, "r", 438)
   if fd then
     local stat = uv.fs_fstat(fd)
     local data = uv.fs_read(fd, stat.size, 0)
     uv.fs_close(fd)
+
     if data and ok_yaml then
       local ok, result = pcall(lyaml.load, data)
       if ok and type(result) == "table" then
@@ -71,9 +109,6 @@ function M.load()
         if type(result.default_prompt_blocks) == "table" then
           config.default_prompt_blocks = result.default_prompt_blocks
         end
-        if type(result.prompt_char_limit) == "number" then
-          config.prompt_char_limit = result.prompt_char_limit
-        end
         if type(result.project_name) == "string" then
           config.project_name = result.project_name
         end
@@ -83,6 +118,12 @@ function M.load()
         if type(result.initial_files) == "table" then
           config.initial_files = result.initial_files
         end
+        if type(result.include_file_contents) == "boolean" then
+          config.include_file_contents = result.include_file_contents
+        end
+        if type(result.ignore_files) == "table" then
+          config.ignore_files = result.ignore_files
+        end
         if type(result.preview_changes) == "boolean" then
           config.preview_changes = result.preview_changes
         end
@@ -98,20 +139,35 @@ function M.load()
         if type(result.enable_chunking) == "boolean" then
           config.enable_chunking = result.enable_chunking
         end
-        -- Added logic to load enable_step_by_step from user config
         if type(result.enable_step_by_step) == "boolean" then
           config.enable_step_by_step = result.enable_step_by_step
         end
-        if type(result.enable_debug_commands) == "boolean" then
-          config.enable_debug_commands = result.enable_debug_commands
+
+        if type(result.auto_lint) == "boolean" then
+          config.auto_lint = result.auto_lint
+        end
+
+        if type(result.tool_auto_accept) == "table" then
+          for k, v in pairs(result.tool_auto_accept) do
+            if config.tool_auto_accept[k] ~= nil and type(v) == "boolean" then
+              config.tool_auto_accept[k] = v
+            end
+          end
+        end
+
+        if type(result.max_token) == "number" then
+          config.max_token = result.max_token
+        else
+          config.max_token = 2048
         end
       end
     end
   else
     config.initial_prompt = "You are a coding assistant who receives a project's context and user instructions..."
+    config.max_token = 2048
   end
 
-  -- Merge the default prompt blocks with the config's initial prompt
+  -- Merge default prompt blocks
   if type(config.default_prompt_blocks) == "table" and #config.default_prompt_blocks > 0 then
     local merged_prompt = {}
     for _, block_name in ipairs(config.default_prompt_blocks) do
@@ -129,6 +185,12 @@ function M.load()
     end
   end
 
+  -- Include project name in the final initial prompt, if set
+  if config.project_name ~= "" then
+    config.initial_prompt =
+      "[Project Name: " .. config.project_name .. "]\n\n" .. config.initial_prompt
+  end
+
   if config.debug then
     vim.api.nvim_out_write("[chatgpt_nvim:config] Loaded config from: " .. path .. "\n")
     vim.api.nvim_out_write("[chatgpt_nvim:config] Debug logging is enabled.\n")

lua/chatgpt_nvim/context.lua

@@ -131,6 +131,11 @@ end
 function M.get_project_files(directories, conf)
   local root = vim.fn.getcwd()
   local ignore_patterns = load_gitignore_patterns(root, conf)
+  if conf.ignore_files then
+    for _, pattern in ipairs(conf.ignore_files) do
+      table.insert(ignore_patterns, gitignore_to_lua_pattern(pattern))
+    end
+  end
   local all_files = {}
   for _, dir in ipairs(directories) do
     local abs_dir = dir
@@ -143,7 +148,9 @@ function M.get_project_files(directories, conf)
   local rel_files = {}
   for _, f in ipairs(all_files) do
     local rel = vim.fn.fnamemodify(f, ":.")
-    table.insert(rel_files, rel)
+    if not rel:match("^%.?chatgpt_config%.yaml$") then
+      table.insert(rel_files, rel)
+    end
   end
 
   if conf.debug then
@@ -185,4 +192,16 @@ function M.get_file_contents(files, conf)
   return table.concat(sections, "\n")
 end
 
-return M
+-- NEW FUNCTION: Build the project prompt by optionally including file contents.
+function M.get_project_prompt(directories, conf)
+  local structure = M.get_project_structure(directories, conf)
+  if conf.include_file_contents then
+    local files = M.get_project_files(directories, conf)
+    local contents = M.get_file_contents(files, conf)
+    return structure .. "\n" .. contents
+  else
+    return structure
+  end
+end
+
+return M

lua/chatgpt_nvim/init.lua

@@ -6,8 +6,14 @@ local config = require('chatgpt_nvim.config')
 local ui = require('chatgpt_nvim.ui')
 local prompts = require('chatgpt_nvim.prompts')
 
+local tools_manager = require("chatgpt_nvim.tools.manager")
+local tools_module = require("chatgpt_nvim.tools")
+
 local ok_yaml, lyaml = pcall(require, "lyaml")
 
+------------------------------------------------------------------------------
+-- UTILITIES
+------------------------------------------------------------------------------
 local function copy_to_clipboard(text)
   vim.fn.setreg('+', text)
 end
@@ -48,288 +54,153 @@ local function read_file(path)
   return data
 end
 
-local function is_directory(path)
-  local stat = vim.loop.fs_stat(path)
-  return stat and stat.type == "directory"
+-- Added function to close existing buffers matching a name pattern.
+local function close_existing_buffer_by_name(name_pattern)
+  for _, buf in ipairs(vim.api.nvim_list_bufs()) do
+    local buf_name = vim.api.nvim_buf_get_name(buf)
+    if buf_name:match(name_pattern) then
+      vim.api.nvim_buf_delete(buf, {force = true})
+    end
+  end
+end
+
+------------------------------------------------------------------------------
+-- PROMPT CONSTRUCTION
+------------------------------------------------------------------------------
+local function build_tools_available_block()
+  local lines = {}
+  lines[#lines+1] = "<tools_available>"
+  for _, t in ipairs(tools_module.available_tools) do
+    lines[#lines+1] = string.format("- **%s**: %s\n  Usage: %s",
+      t.name, t.explanation, t.usage
+    )
+  end
+  lines[#lines+1] = "</tools_available>"
+  return table.concat(lines, "\n")
+end
+
+local function build_prompt(user_input, dirs, conf)
+  local root = vim.fn.getcwd()
+  local initial_files = conf.initial_files or {}
+  local final_sections = {}
+
+  -- 1) <initial_prompts>
+  if conf.initial_prompt and conf.initial_prompt ~= "" then
+    table.insert(final_sections, "<initial_prompts>\n" .. conf.initial_prompt .. "\n</initial_prompts>\n")
+  end
+
+  -- 2) <tools_available>
+  table.insert(final_sections, build_tools_available_block())
+
+  -- 3) <task>
+  local task_lines = {}
+  task_lines[#task_lines+1] = "<task>"
+  task_lines[#task_lines+1] = user_input
+  for _, file_path in ipairs(initial_files) do
+    task_lines[#task_lines+1] = ("'%s' (see below for file content)"):format(file_path)
+  end
+  task_lines[#task_lines+1] = "</task>\n"
+  table.insert(final_sections, table.concat(task_lines, "\n"))
+
+  -- 4) <file_content path="..."> from initial_files
+  local file_content_blocks = {}
+  for _, file_path in ipairs(initial_files) do
+    local full_path = root .. "/" .. file_path
+    if is_subpath(root, full_path) then
+      local fdata = read_file(full_path)
+      if fdata then
+        file_content_blocks[#file_content_blocks+1] = string.format(
+          "<file_content path=\"%s\">\n%s\n</file_content>", file_path, fdata
+        )
+      end
+    end
+  end
+  if #file_content_blocks > 0 then
+    table.insert(final_sections, table.concat(file_content_blocks, "\n\n"))
+  end
+
+  -- 4.1) Dynamic file inclusion via @ operator in user_input
+  local dynamic_files = {}
+  for file in user_input:gmatch("@([^%s]+)") do
+    if file ~= "chatgpt_config.yaml" and file ~= ".chatgpt_config.yaml" then
+      local already_included = false
+      for _, existing in ipairs(initial_files) do
+        if existing == file then
+          already_included = true
+          break
+        end
+      end
+      if not already_included then
+        table.insert(dynamic_files, file)
+      end
+    end
+  end
+
+  local dynamic_file_blocks = {}
+  for _, file in ipairs(dynamic_files) do
+    local full_path = root .. "/" .. file
+    if is_subpath(root, full_path) then
+      local fdata = read_file(full_path)
+      if fdata then
+        dynamic_file_blocks[#dynamic_file_blocks+1] = string.format(
+          "<file_content path=\"%s\">\n%s\n</file_content>", file, fdata
+        )
+      end
+    end
+  end
+  if #dynamic_file_blocks > 0 then
+    table.insert(final_sections, table.concat(dynamic_file_blocks, "\n\n"))
+  end
+
+  -- 5) <environment_details>
+  local env_lines = {}
+  env_lines[#env_lines+1] = "<environment_details>"
+  env_lines[#env_lines+1] = "# VSCode Visible Files"
+  for _, f in ipairs(initial_files) do
+    env_lines[#env_lines+1] = f
+  end
+  env_lines[#env_lines+1] = ""
+  env_lines[#env_lines+1] = "# VSCode Open Tabs"
+  env_lines[#env_lines+1] = "..."
+  env_lines[#env_lines+1] = ""
+  env_lines[#env_lines+1] = "# Current Time"
+  env_lines[#env_lines+1] = os.date("%x, %X (%Z)")
+  env_lines[#env_lines+1] = ""
+  env_lines[#env_lines+1] = "# Current Working Directory (" .. root .. ") Files"
+  env_lines[#env_lines+1] = context.get_project_prompt(dirs, conf) or ""
+  env_lines[#env_lines+1] = ""
+  env_lines[#env_lines+1] = "# Current Mode"
+  env_lines[#env_lines+1] = "ACT MODE"
+  env_lines[#env_lines+1] = "</environment_details>"
+  table.insert(final_sections, table.concat(env_lines, "\n"))
+
+  local final_prompt = table.concat(final_sections, "\n\n")
+  final_prompt = final_prompt:gsub("%chatgpt.vim%", conf.project_name)
+  return final_prompt
+end
+
+local function estimate_token_count(text)
+  local token_count = 0
+  for chunk in text:gmatch("%S+") do
+    for token in chunk:gmatch("(%w+|%p+)") do
+      token_count = token_count + 1
+    end
+  end
+  return token_count
 end
 
 local function handle_step_by_step_if_needed(prompt, conf)
-  local length = #prompt
-  if not conf.enable_step_by_step or length <= (conf.prompt_char_limit or 8000) then
+  local token_count = estimate_token_count(prompt)
+  local limit = conf.max_token or 2048
+  if (not conf.enable_step_by_step) or (token_count <= limit) then
     return { prompt }
   end
   return { prompts["step-prompt"] }
 end
 
-local function close_existing_buffer_by_name(pattern)
-  for _, b in ipairs(vim.api.nvim_list_bufs()) do
-    local name = vim.api.nvim_buf_get_name(b)
-    if name:match(pattern) then
-      vim.api.nvim_buf_delete(b, { force = true })
-    end
-  end
-end
-
-local function preview_changes(changes, conf)
-  close_existing_buffer_by_name("ChatGPT_Changes_Preview$")
-  local bufnr = vim.api.nvim_create_buf(false, true)
-  vim.api.nvim_buf_set_name(bufnr, "ChatGPT_Changes_Preview")
-  vim.api.nvim_buf_set_option(bufnr, "filetype", "diff")
-  vim.api.nvim_buf_set_lines(bufnr, 0, -1, false, {
-    "# Preview of Changes:",
-    "# (Close this window to apply changes or use :q to cancel)",
-    ""
-  })
-  for _, fileinfo in ipairs(changes) do
-    local indicator = (fileinfo.delete == true) and "Delete file" or "Write file"
-    vim.api.nvim_buf_set_lines(bufnr, -1, -1, false, {
-      string.format("=== %s: %s ===", indicator, fileinfo.path or "<no path>")
-    })
-    if fileinfo.content then
-      local lines = vim.split(fileinfo.content, "\n")
-      for _, line in ipairs(lines) do
-        vim.api.nvim_buf_set_lines(bufnr, -1, -1, false, { line })
-      end
-    end
-    vim.api.nvim_buf_set_lines(bufnr, -1, -1, false, { "" })
-  end
-
-  vim.cmd("vsplit")
-  vim.cmd("buffer " .. bufnr)
-end
-
-local function partial_accept(changes, conf)
-  close_existing_buffer_by_name("ChatGPT_Partial_Accept$")
-  local bufnr = vim.api.nvim_create_buf(false, true)
-  vim.api.nvim_buf_set_name(bufnr, "ChatGPT_Partial_Accept")
-  vim.api.nvim_buf_set_option(bufnr, "filetype", "diff")
-
-  local lines = {
-    "# Remove or comment out (prepend '#') any changes you do NOT want, then :wq, :x, or :bd to finalize partial acceptance",
-    ""
-  }
-  for _, fileinfo in ipairs(changes) do
-    local action = (fileinfo.delete == true) and "[DELETE]" or "[WRITE]"
-    table.insert(lines, string.format("%s %s", action, fileinfo.path or "<no path>"))
-    if fileinfo.content then
-      local content_lines = vim.split(fileinfo.content, "\n")
-      for _, cl in ipairs(content_lines) do
-        table.insert(lines, "  " .. cl)
-      end
-    end
-    table.insert(lines, "")
-  end
-
-  vim.api.nvim_buf_set_lines(bufnr, 0, -1, false, lines)
-
-  local final_changes = {}
-  local function on_write()
-    local edited_lines = vim.api.nvim_buf_get_lines(bufnr, 0, -1, false)
-    local keep_current = false
-    local current_fileinfo = { path = nil, content = nil, delete = false }
-    local content_accum = {}
-
-    for _, line in ipairs(edited_lines) do
-      if line:match("^#") or line == "" then
-        goto continue
-      end
-      local del_match = line:match("^%[DELETE%] (.+)")
-      local write_match = line:match("^%[WRITE%] (.+)")
-      if del_match then
-        if keep_current and (current_fileinfo.path ~= nil) then
-          if #content_accum > 0 then
-            current_fileinfo.content = table.concat(content_accum, "\n")
-          end
-          table.insert(final_changes, current_fileinfo)
-        end
-        keep_current = true
-        current_fileinfo = { path = del_match, delete = true, content = nil }
-        content_accum = {}
-      elseif write_match then
-        if keep_current and (current_fileinfo.path ~= nil) then
-          if #content_accum > 0 then
-            current_fileinfo.content = table.concat(content_accum, "\n")
-          end
-          table.insert(final_changes, current_fileinfo)
-        end
-        keep_current = true
-        current_fileinfo = { path = write_match, delete = false, content = nil }
-        content_accum = {}
-      else
-        if keep_current then
-          table.insert(content_accum, line:gsub("^%s*", ""))
-        end
-      end
-      ::continue::
-    end
-
-    if keep_current and (current_fileinfo.path ~= nil) then
-      if #content_accum > 0 then
-        current_fileinfo.content = table.concat(content_accum, "\n")
-      end
-      table.insert(final_changes, current_fileinfo)
-    end
-
-    vim.api.nvim_buf_set_option(bufnr, "modified", false)
-  end
-
-  vim.api.nvim_create_autocmd("BufWriteCmd", {
-    buffer = bufnr,
-    once = true,
-    callback = function()
-      on_write()
-      vim.cmd("bd! " .. bufnr)
-    end
-  })
-
-  vim.cmd("split")
-  vim.cmd("buffer " .. bufnr)
-
-  vim.wait(60000, function()
-    local winids = vim.api.nvim_tabpage_list_wins(0)
-    for _, w in ipairs(winids) do
-      local b = vim.api.nvim_win_get_buf(w)
-      if b == bufnr then
-        return false
-      end
-    end
-    return true
-  end)
-
-  return final_changes
-end
-
-local function store_prompt_for_reference(prompt, conf)
-  close_existing_buffer_by_name("ChatGPT_Generated_Prompt$")
-  local bufnr = vim.api.nvim_create_buf(false, true)
-  vim.api.nvim_buf_set_name(bufnr, "ChatGPT_Generated_Prompt")
-  vim.api.nvim_buf_set_option(bufnr, "filetype", "markdown")
-
-  local lines = {
-    "# Below is the generated prompt. You can keep it for reference:",
-    ""
-  }
-  local prompt_lines = vim.split(prompt, "\n")
-  for _, line in ipairs(prompt_lines) do
-    table.insert(lines, line)
-  end
-
-  vim.api.nvim_buf_set_lines(bufnr, 0, -1, false, lines)
-  vim.cmd("vsplit")
-  vim.cmd("buffer " .. bufnr)
-end
-
--- Updated to allow flags/args for 'ls' and 'grep'
-local function execute_debug_command(cmd, conf)
-  if type(cmd) ~= "table" or not cmd.command then
-    return "Invalid command object."
-  end
-
-  local command = cmd.command
-  if command == "ls" then
-    -- Accept optional flags/args
-    local dir = cmd.dir or "."
-    local args = cmd.args or {}
-    local cmd_str = "ls"
-    if #args > 0 then
-      cmd_str = cmd_str .. " " .. table.concat(args, " ")
-    end
-    cmd_str = cmd_str .. " " .. dir
-
-    local handle = io.popen(cmd_str)
-    if not handle then
-      return "Failed to run ls command."
-    end
-    local result = handle:read("*a") or ""
-    handle:close()
-    return "Listing files in: " .. dir .. "\n" .. result
-
-  elseif command == "grep" then
-    -- Accept optional flags/args
-    -- If the user wants to do something like:
-    --   args: ["-r", "somePattern", "someDir"]
-    -- we just pass them all to grep.
-    local args = cmd.args or {}
-    if #args == 0 then
-      -- fallback to old usage
-      local pattern = cmd.pattern
-      local target = cmd.target
-      if not pattern or not target then
-        return "Usage for grep: {command='grep', args=['-r','pattern','target']} or {pattern='<text>', target='<path>'}"
-      end
-      local stat = vim.loop.fs_stat(target)
-      if not stat then
-        return "Cannot grep: target path does not exist"
-      end
-      -- old logic remains for backward compatibility
-      if stat.type == "directory" then
-        local handle = io.popen("ls -p " .. target .. " | grep -v /")
-        if not handle then
-          return "Failed to read directory contents for grep."
-        end
-        local all_files = {}
-        for file in handle:read("*a"):gmatch("[^\n]+") do
-          table.insert(all_files, target .. "/" .. file)
-        end
-        handle:close()
-        local results = {}
-        local function grep_in_file(search_string, filepath)
-          local content = read_file(filepath)
-          if not content then
-            return "Could not read file: " .. filepath
-          end
-          local lines = {}
-          local line_num = 0
-          for line in content:gmatch("([^\n]*)\n?") do
-            line_num = line_num + 1
-            if line:find(search_string, 1, true) then
-              table.insert(lines, filepath .. ":" .. line_num .. ":" .. line)
-            end
-          end
-          return (#lines == 0) and ("No matches in " .. filepath) or table.concat(lines, "\n")
-        end
-        for _, f in ipairs(all_files) do
-          local fstat = vim.loop.fs_stat(f)
-          if fstat and fstat.type == "file" then
-            table.insert(results, grep_in_file(pattern, f))
-          end
-        end
-        return table.concat(results, "\n")
-      else
-        local function grep_in_file(search_string, filepath)
-          local content = read_file(filepath)
-          if not content then
-            return "Could not read file: " .. filepath
-          end
-          local lines = {}
-          local line_num = 0
-          for line in content:gmatch("([^\n]*)\n?") do
-            line_num = line_num + 1
-            if line:find(search_string, 1, true) then
-              table.insert(lines, filepath .. ":" .. line_num .. ":" .. line)
-            end
-          end
-          return (#lines == 0) and ("No matches in " .. filepath) or table.concat(lines, "\n")
-        end
-        return grep_in_file(pattern, target)
-      end
-    else
-      -- new approach with flags/args
-      local cmd_str = "grep " .. table.concat(args, " ")
-      local handle = io.popen(cmd_str)
-      if not handle then
-        return "Failed to run grep command."
-      end
-      local result = handle:read("*a") or ""
-      handle:close()
-      return result
-    end
-  else
-    return "Unknown command: " .. command
-  end
-end
-
 local function run_chatgpt_command()
+  package.loaded["chatgpt_nvim.config"] = nil
+  local config = require("chatgpt_nvim.config")
   local conf = config.load()
   ui.setup_ui(conf)
   ui.debug_log("Running :ChatGPT command.")
@@ -345,13 +216,15 @@ local function run_chatgpt_command()
   local bufnr = vim.api.nvim_create_buf(false, false)
   vim.api.nvim_buf_set_name(bufnr, "ChatGPT_Prompt.md")
   vim.api.nvim_buf_set_option(bufnr, "filetype", "markdown")
+  vim.api.nvim_buf_set_option(bufnr, "omnifunc", "v:lua.chatgpt_file_complete")
   vim.api.nvim_buf_set_option(bufnr, "bufhidden", "wipe")
   vim.api.nvim_buf_set_option(bufnr, "buftype", "")
   vim.api.nvim_buf_set_option(bufnr, "modifiable", true)
 
   vim.api.nvim_buf_set_lines(bufnr, 0, -1, false, {
-    "# Enter your prompt below.",
+    "# Enter your main user prompt (task) below.",
     "",
+    "You can include files by typing @filename in your prompt.",
     "Save & close with :wq, :x, or :bd to finalize your prompt."
   })
 
@@ -360,57 +233,25 @@ local function run_chatgpt_command()
     callback = function()
       local lines = vim.api.nvim_buf_get_lines(bufnr, 0, -1, false)
       local user_input = table.concat(lines, "\n")
-      if user_input == "" or user_input:find("^# Enter your prompt below.") then
+      if user_input == "" or user_input:find("^# Enter your main user prompt %(task%) below.") then
         vim.api.nvim_out_write("No valid input provided.\n")
         vim.api.nvim_buf_set_option(bufnr, "modified", false)
         return
       end
 
-      local project_structure = context.get_project_structure(dirs, conf)
-      local initial_files = conf.initial_files or {}
-      local included_sections = {}
+      local final_prompt = build_prompt(user_input, dirs, conf)
+      local chunks = handle_step_by_step_if_needed(final_prompt, conf)
 
-      for _, item in ipairs(initial_files) do
-        local root = vim.fn.getcwd()
-        local full_path = root .. "/" .. item
-        if is_directory(full_path) then
-          local dir_files = context.get_project_files({item}, conf)
-          for _, f in ipairs(dir_files) do
-            local path = root .. "/" .. f
-            local data = read_file(path)
-            if data then
-              table.insert(included_sections, "\nFile: `" .. f .. "`\n```\n" .. data .. "\n```\n")
-            end
-          end
-        else
-          local data = read_file(full_path)
-          if data then
-            table.insert(included_sections, "\nFile: `" .. item .. "`\n```\n" .. data .. "\n```\n")
-          end
-        end
-      end
+      close_existing_buffer_by_name("ChatGPT_Generated_Prompt$")
+      local bufnr_ref = vim.api.nvim_create_buf(false, true)
+      vim.api.nvim_buf_set_name(bufnr_ref, "ChatGPT_Generated_Prompt")
+      vim.api.nvim_buf_set_option(bufnr_ref, "filetype", "markdown")
 
-      local initial_sections = {
-        "## Basic Prompt Instructions:\n",
-        conf.initial_prompt .. "\n\n\n",
-        "## User Instructions:\n",
-        user_input .. "\n\n\n",
-        "## Context/Data:\n",
-        "Project name: " .. (conf.project_name or "") .. "\n",
-        "Project Structure:\n",
-        project_structure,
-        table.concat(included_sections, "\n")
-      }
+      local prompt_lines = vim.split(chunks[1], "\n")
+      vim.api.nvim_buf_set_lines(bufnr_ref, 0, -1, false, prompt_lines)
+      vim.cmd("vsplit")
+      vim.cmd("buffer " .. bufnr_ref)
 
-      if conf.enable_debug_commands then
-        table.insert(initial_sections, "\n## Debug Commands Info:\n")
-        table.insert(initial_sections, prompts["debug-commands-info"])
-      end
-
-      local prompt = table.concat(initial_sections, "\n")
-      store_prompt_for_reference(prompt, conf)
-
-      local chunks = handle_step_by_step_if_needed(prompt, conf)
       copy_to_clipboard(chunks[1])
       if #chunks == 1 then
         vim.api.nvim_out_write("Prompt copied to clipboard! Paste into ChatGPT.\n")
@@ -426,6 +267,8 @@ local function run_chatgpt_command()
 end
 
 local function run_chatgpt_paste_command()
+  package.loaded["chatgpt_nvim.config"] = nil
+  local config = require("chatgpt_nvim.config")
   local conf = config.load()
   ui.setup_ui(conf)
   ui.debug_log("Running :ChatGPTPaste command.")
@@ -441,19 +284,19 @@ local function run_chatgpt_paste_command()
     return
   end
 
-  if data.commands and conf.enable_debug_commands then
-    local results = {}
-    for _, cmd in ipairs(data.commands) do
-      table.insert(results, execute_debug_command(cmd, conf))
+  if data.tools then
+    if not data.project_name or data.project_name ~= conf.project_name then
+      vim.api.nvim_err_writeln("Project name mismatch or missing. Aborting tool usage.")
+      return
     end
-    local output = table.concat(results, "\n\n")
-    copy_to_clipboard(output)
-    print("Debug command results copied to clipboard!")
+
+    local output_messages = tools_manager.handle_tool_calls(data.tools, conf, is_subpath, read_file)
+    copy_to_clipboard(output_messages)
+    print("Tool call results have been processed and copied to clipboard.")
     return
   end
 
   if data.project_name and data.files then
-    ui.debug_log("Received project_name and files in response.")
     if data.project_name ~= conf.project_name then
       vim.api.nvim_err_writeln("Project name mismatch. Aborting.")
       return
@@ -469,7 +312,7 @@ local function run_chatgpt_paste_command()
 
     if is_final then
       if conf.preview_changes then
-        preview_changes(data.files, conf)
+        require('chatgpt_nvim.init').preview_changes(data.files, conf)
         print("Close the preview window to apply changes, or use :q to cancel.")
         local closed = vim.wait(60000, function()
           local bufs = vim.api.nvim_list_bufs()
@@ -489,7 +332,7 @@ local function run_chatgpt_paste_command()
 
       local final_files = data.files
       if conf.partial_acceptance then
-        final_files = partial_accept(data.files, conf)
+        final_files = require('chatgpt_nvim.init').partial_accept(data.files, conf)
         if #final_files == 0 then
           vim.api.nvim_err_writeln("No changes remain after partial acceptance. Aborting.")
           return
@@ -500,30 +343,27 @@ local function run_chatgpt_paste_command()
       for _, fileinfo in ipairs(final_files) do
         if not fileinfo.path then
           vim.api.nvim_err_writeln("Invalid file entry. Must have 'path'.")
-          goto continue
-        end
-
-        if not is_subpath(root, fileinfo.path) then
-          vim.api.nvim_err_writeln("Invalid path outside project root: " .. fileinfo.path)
-          goto continue
-        end
-
-        if fileinfo.delete == true then
-          ui.debug_log("Deleting file: " .. fileinfo.path)
-          handler.delete_file(fileinfo.path, conf)
-          print("Deleted: " .. fileinfo.path)
-        elseif fileinfo.content then
-          ui.debug_log("Writing file: " .. fileinfo.path)
-          handler.write_file(fileinfo.path, fileinfo.content, conf)
-          print("Wrote: " .. fileinfo.path)
         else
-          vim.api.nvim_err_writeln("Invalid file entry. Must have 'content' or 'delete'.")
+          if not is_subpath(root, fileinfo.path) then
+            vim.api.nvim_err_writeln("Invalid path outside project root: " .. fileinfo.path)
+          else
+            if fileinfo.delete == true then
+              ui.debug_log("Deleting file: " .. fileinfo.path)
+              handler.delete_file(fileinfo.path, conf)
+              print("Deleted: " .. fileinfo.path)
+            elseif fileinfo.content then
+              ui.debug_log("Writing file: " .. fileinfo.path)
+              handler.write_file(fileinfo.path, fileinfo.content, conf)
+              print("Wrote: " .. fileinfo.path)
+            else
+              vim.api.nvim_err_writeln("Invalid file entry. Must have 'content' or 'delete'.")
+            end
+          end
         end
-        ::continue::
       end
-
     else
       local requested_paths = {}
+      local root = vim.fn.getcwd()
       for _, fileinfo in ipairs(data.files) do
         if fileinfo.path then
           table.insert(requested_paths, fileinfo.path)
@@ -531,14 +371,13 @@ local function run_chatgpt_paste_command()
       end
 
       local file_sections = {}
-      local root = vim.fn.getcwd()
       for _, f in ipairs(requested_paths) do
         local path = root .. "/" .. f
         local content = read_file(path)
         if content then
-          table.insert(file_sections, "\nFile: `" .. f .. "`\n```\n" .. content .. "\n```\n")
+          table.insert(file_sections, ("\nFile: `%s`\n```\n%s\n```\n"):format(f, content))
         else
-          table.insert(file_sections, "\nFile: `" .. f .. "`\n```\n(Could not read file)\n```\n")
+          table.insert(file_sections, ("\nFile: `%s`\n```\n(Could not read file)\n```\n"):format(f))
         end
       end
 
@@ -547,35 +386,28 @@ local function run_chatgpt_paste_command()
         "\n\nProject name: " .. (conf.project_name or ""),
         "\n\nBelow are the requested files from the project, each preceded by its filename in backticks and enclosed in triple backticks.\n",
         table.concat(file_sections, "\n"),
-        "\n\nIf you need more files, please respond again in YAML listing additional files. If you have all information you need, provide the final YAML with `project_name` and `files` (with `content` or `delete`) to apply changes.\n"
+        "\n\nIf you need more files, please respond again in YAML listing additional files, or use the 'tools:' approach. If you have all info, provide final changes or continue your instructions."
       }
 
       local prompt = table.concat(sections, "\n")
       local length = #prompt
       ui.debug_log("Returning requested files. Character count: " .. length)
 
-      if length > (conf.prompt_char_limit or 8000) and conf.enable_step_by_step then
-        local large_step = prompts["step-prompt"]
-        copy_to_clipboard(large_step)
-        print("Step-by-step guidance copied to clipboard!")
-        return
-      elseif length > (conf.prompt_char_limit or 8000) then
-        vim.api.nvim_err_writeln("Requested files exceed prompt character limit. No step-by-step support enabled.")
-        return
-      end
-
       copy_to_clipboard(prompt)
-      print("Prompt (with requested files) copied to clipboard! Paste it into the ChatGPT O1 model.")
+      print("Prompt (with requested files) copied to clipboard! Paste it into ChatGPT.")
     end
   else
-    vim.api.nvim_err_writeln("Invalid response. Expected 'project_name' and 'files'.")
+    vim.api.nvim_err_writeln("No tools or recognized file instructions found. Provide 'tools:' or older 'project_name & files'.")
   end
 end
 
 local function run_chatgpt_current_buffer_command()
+  package.loaded["chatgpt_nvim.config"] = nil
+  local config = require("chatgpt_nvim.config")
   local conf = config.load()
   ui.setup_ui(conf)
   ui.debug_log("Running :ChatGPTCurrentBuffer command.")
+
   local lines = vim.api.nvim_buf_get_lines(0, 0, -1, false)
   local user_input = table.concat(lines, "\n")
   local dirs = conf.directories or {"."}
@@ -586,92 +418,19 @@ local function run_chatgpt_current_buffer_command()
     end
   end
 
-  local project_structure = context.get_project_structure(dirs, conf)
-  local initial_files = conf.initial_files or {}
-  local included_sections = {}
+  local final_prompt = build_prompt(user_input, dirs, conf)
+  local chunks = handle_step_by_step_if_needed(final_prompt, conf)
 
-  local function is_directory(path)
-    local stat = vim.loop.fs_stat(path)
-    return stat and stat.type == "directory"
-  end
+  close_existing_buffer_by_name("ChatGPT_Generated_Prompt$")
+  local bufnr_ref = vim.api.nvim_create_buf(false, true)
+  vim.api.nvim_buf_set_name(bufnr_ref, "ChatGPT_Generated_Prompt")
+  vim.api.nvim_buf_set_option(bufnr_ref, "filetype", "markdown")
 
-  local function read_file(path)
-    local fd = vim.loop.fs_open(path, "r", 438)
-    if not fd then
-      return nil
-    end
-    local stat = vim.loop.fs_fstat(fd)
-    if not stat then
-      vim.loop.fs_close(fd)
-      return nil
-    end
-    local data = vim.loop.fs_read(fd, stat.size, 0)
-    vim.loop.fs_close(fd)
-    return data
-  end
+  local prompt_lines = vim.split(chunks[1], "\n")
+  vim.api.nvim_buf_set_lines(bufnr_ref, 0, -1, false, prompt_lines)
+  vim.cmd("vsplit")
+  vim.cmd("buffer " .. bufnr_ref)
 
-  for _, item in ipairs(initial_files) do
-    local root = vim.fn.getcwd()
-    local full_path = root .. "/" .. item
-    if is_directory(full_path) then
-      local dir_files = context.get_project_files({item}, conf)
-      for _, f in ipairs(dir_files) do
-        local path = root .. "/" .. f
-        local data = read_file(path)
-        if data then
-          table.insert(included_sections, "\nFile: `" .. f .. "`\n```\n" .. data .. "\n```\n")
-        end
-      end
-    else
-      local data = read_file(full_path)
-      if data then
-        table.insert(included_sections, "\nFile: `" .. item .. "`\n```\n" .. data .. "\n```\n")
-      end
-    end
-  end
-
-  local initial_sections = {
-    "## Basic Prompt Instructions:\n",
-    conf.initial_prompt .. "\n\n\n",
-    "## User Instructions:\n",
-    user_input .. "\n\n\n",
-    "## Context/Data:\n",
-    "Project name: " .. (conf.project_name or "") .. "\n",
-    "Project Structure:\n",
-    project_structure,
-    table.concat(included_sections, "\n")
-  }
-
-  if conf.enable_debug_commands then
-    table.insert(initial_sections, "\n## Debug Commands Info:\n")
-    table.insert(initial_sections, prompts["debug-commands-info"])
-  end
-
-  local prompt = table.concat(initial_sections, "\n")
-
-  local function store_prompt_for_reference(pr)
-    close_existing_buffer_by_name("ChatGPT_Generated_Prompt$")
-    local bufnr_ref = vim.api.nvim_create_buf(false, true)
-    vim.api.nvim_buf_set_name(bufnr_ref, "ChatGPT_Generated_Prompt")
-    vim.api.nvim_buf_set_option(bufnr_ref, "filetype", "markdown")
-
-    local lines_ref = {
-      "# Below is the generated prompt. You can keep it for reference:",
-      ""
-    }
-    local pr_lines = vim.split(pr, "\n")
-    for _, line in ipairs(pr_lines) do
-      table.insert(lines_ref, line)
-    end
-
-    vim.api.nvim_buf_set_lines(bufnr_ref, 0, -1, false, lines_ref)
-    vim.cmd("vsplit")
-    vim.cmd("buffer " .. bufnr_ref)
-  end
-
-  store_prompt_for_reference(prompt, conf)
-
-  local chunks = handle_step_by_step_if_needed(prompt, conf)
   copy_to_clipboard(chunks[1])
   if #chunks == 1 then
     vim.api.nvim_out_write("Prompt (from current buffer) copied to clipboard! Paste into ChatGPT.\n")
@@ -683,6 +442,29 @@ end
 M.run_chatgpt_command = run_chatgpt_command
 M.run_chatgpt_paste_command = run_chatgpt_paste_command
 M.run_chatgpt_current_buffer_command = run_chatgpt_current_buffer_command
-M.execute_debug_command = execute_debug_command
+
+function _G.chatgpt_file_complete(findstart, base)
+  if findstart == 1 then
+    local line = vim.fn.getline('.')
+    local col = vim.fn.col('.')
+    local start = line:sub(1, col):find("@[^%s]*$")
+    if start then
+      return start - 1
+    else
+      return -1
+    end
+  else
+    local conf = config.load()
+    local files = context.get_project_files({'.'}, conf)
+    local completions = {}
+    local esc_base = base:gsub("([^%w])", "%%%1")
+    for _, f in ipairs(files) do
+      if f:match("^" .. esc_base) then
+        table.insert(completions, f)
+      end
+    end
+    return completions
+  end
+end
 
 return M

lua/chatgpt_nvim/prompts.lua

@@ -1,5 +1,82 @@
 local M = {
-  ["solidjs-development"] = [[
+  ["memory-bank"] = [[
+    # Memory Bank
+
+    You are Cline, an expert software engineer with a unique constraint: your memory periodically resets completely. This isn't a bug - it's what makes you maintain perfect documentation. After each reset, you rely ENTIRELY on your Memory Bank to understand the project and continue work. Without proper documentation, you cannot function effectively.
+
+    ## Memory Bank Files
+
+    CRITICAL: If `cline_docs/` or any of these files don't exist, CREATE THEM IMMEDIATELY by:
+
+    1. Reading all provided documentation
+    2. Asking user for ANY missing information
+    3. Creating files with verified information only
+    4. Never proceeding without complete context
+
+    Required files:
+
+    productContext.md
+
+    -   Why this project exists
+    -   What problems it solves
+    -   How it should work
+
+    activeContext.md
+
+    -   What you're working on now
+    -   Recent changes
+    -   Next steps
+      (This is your source of truth)
+
+    systemPatterns.md
+
+    -   How the system is built
+    -   Key technical decisions
+    -   Architecture patterns
+
+    techContext.md
+
+    -   Technologies used
+    -   Development setup
+    -   Technical constraints
+
+    progress.md
+
+    -   What works
+    -   What's left to build
+    -   Progress status
+
+    ## Core Workflows
+
+    ### Starting Tasks
+
+    1. Check for Memory Bank files
+    2. If ANY files missing, stop and create them
+    3. Read ALL files before proceeding
+    4. Verify you have complete context
+    5. Begin development. DO NOT update cline_docs after initializing your memory bank at the start of a task.
+
+    ### During Development
+
+    1. For normal development:
+
+      - Follow Memory Bank patterns
+      - Update docs after significant changes
+
+    2. Say `[MEMORY BANK: ACTIVE]` at the beginning of every tool use.
+
+    ### Memory Bank Updates
+
+    When user says "update memory bank":
+
+    1. This means imminent memory reset
+    2. Document EVERYTHING about current state
+    3. Make next steps crystal clear
+    4. Complete current task
+
+    Remember: After every memory reset, you begin completely fresh. Your only link to previous work is the Memory Bank. Maintain it as if your functionality depends on it - because it does.
+  ]],
+  ["solidjs"] = [[
     ### SolidJS Development Guidelines
 
     You are helping me develop a large SolidJS application. Please keep the following points in mind when generating or explaining code:
@@ -73,21 +150,111 @@ local M = {
 
     Please follow these guidelines to ensure the generated or explained code aligns well with SolidJS best practices for large, maintainable projects.
   ]],
-  ["go-development"] = [[
+  ["nodejs"] = [[
+    You are helping me develop a large Node.js application. Please keep the following points in mind when generating or explaining code:
+
+    1. **Project & Folder Structure**  
+       - Maintain a clear, top-level directory layout. For example:
+         ```
+         my-node-app/
+         ├── src/
+         │   ├── controllers/
+         │   ├── models/
+         │   ├── routes/
+         │   ├── services/
+         │   ├── utils/
+         │   └── index.js (or app.js)
+         ├── tests/
+         │   └── ...
+         ├── package.json
+         ├── .env (if needed)
+         ├── .eslintrc.js
+         └── ...
+         ```
+       - Keep your application logic separated in folders:
+         - **controllers/** (or handlers) for request handling logic,  
+         - **services/** for core business logic or data processing,  
+         - **models/** for database schemas/entities,  
+         - **routes/** for defining routes/endpoints,  
+         - **utils/** for reusable helper functions.  
+
+    2. **Dependencies & Package Management**  
+       - Use **npm** or **yarn** (pick one and stay consistent) to manage dependencies.  
+       - Keep your `package.json` clean by removing unused dependencies.  
+       - Pin exact versions for critical or sensitive dependencies to ensure reproducible builds.  
+
+    3. **Configuration & Environment Variables**  
+       - Use environment variables to store sensitive or environment-specific information (e.g., database credentials, API keys).  
+       - Consider a config management library (like [dotenv](https://github.com/motdotla/dotenv) or [dotenv-flow](https://github.com/kerimdzhanov/dotenv-flow)) to load environment variables from `.env` files.  
+       - Keep secrets out of version control (e.g., `.env` should be in `.gitignore`).  
+
+    4. **Code Organization & Module Patterns**  
+       - Use **ES Modules** (`import`/`export`) or **CommonJS** (`require`/`module.exports`) consistently across the project.  
+       - Keep each module focused on a single responsibility.  
+       - Use a **service layer** for business logic to avoid bloated controllers.  
+       - Use **async/await** for asynchronous operations, ensuring proper error handling (try/catch blocks or .catch callbacks).  
+
+    5. **Database & Data Persistence**  
+       - Use an ORM/ODM (e.g., [Sequelize](https://sequelize.org/) for SQL, [Mongoose](https://mongoosejs.com/) for MongoDB) or a query builder ([Knex.js](https://knexjs.org/)) to maintain cleaner database interactions.  
+       - Keep database-related logic in separate **models/** or **repositories/**.  
+       - Handle database migrations carefully (e.g., with [db-migrate](https://db-migrate.readthedocs.io/), [Liquibase](https://www.liquibase.org/), or built-in ORM migration tools).  
+
+    6. **Logging & Monitoring**  
+       - Use a structured logging library (e.g., [Winston](https://github.com/winstonjs/winston), [Pino](https://github.com/pinojs/pino)) to capture logs in JSON or another parseable format.  
+       - Ensure logs include enough context (request IDs, timestamps, etc.) to troubleshoot issues.  
+       - Consider external logging and monitoring solutions (e.g., [Datadog](https://www.datadoghq.com/), [New Relic](https://newrelic.com/)) for production environments.  
+
+    7. **Security & Best Practices**  
+       - Sanitize and validate all user inputs (e.g., using libraries like [validator](https://github.com/validatorjs/validator.js)).  
+       - Avoid SQL injection by using parameterized queries or ORM features.  
+       - Implement rate limiting or request throttling to prevent abuse (e.g., [express-rate-limit](https://github.com/nfriedly/express-rate-limit)).  
+       - Ensure **HTTPS** is used in production and secure headers are set (e.g., [helmet](https://github.com/helmetjs/helmet) for Express).  
+       - Keep dependencies updated to patch known vulnerabilities (use `npm audit` or equivalent).  
+
+    8. **Error Handling**  
+       - Use centralized error handling middleware (if using a framework like Express) to catch and process errors consistently.  
+       - Provide clear error messages but avoid leaking sensitive info in production.  
+       - Separate operational errors (e.g., user-related) from programmer errors (e.g., logic bugs) to handle them appropriately.  
+
+    9. **Testing & Quality Assurance**  
+       - Write **unit tests** for individual modules or functions (e.g., using [Jest](https://jestjs.io/), [Mocha](https://mochajs.org/)).  
+       - Use **integration tests** or **end-to-end tests** (e.g., [Supertest](https://github.com/visionmedia/supertest) for API endpoints).  
+       - Aim for high coverage but focus on critical business logic and error cases.  
+       - Automate tests in your CI pipeline.  
+
+    10. **Linting & Formatting**  
+        - Use **ESLint** (with recommended or popular config like [Airbnb](https://www.npmjs.com/package/eslint-config-airbnb)) for consistent code quality.  
+        - Use **Prettier** for code formatting to maintain a standardized style.  
+        - Configure linting and formatting checks in a pre-commit hook or CI (e.g., [Husky](https://typicode.github.io/husky/), [lint-staged](https://github.com/okonet/lint-staged)).  
+
+    11. **Deployment & Environment Management**  
+        - Containerize your app with Docker if possible, specifying a secure and minimal base image.  
+        - Use process managers like [PM2](https://pm2.keymetrics.io/) or systemd for production Node.js processes.  
+        - Maintain separate configuration (or environment variables) for staging, production, etc.  
+
+    12. **Output Format**  
+        - Present any generated source code with clear folder and file placement (e.g., `controllers/`, `services/`).  
+        - When explaining your reasoning, highlight the architectural decisions or patterns used (e.g., “I introduced a service layer to separate business logic from route handling.”).  
+        - If you modify existing files, specify precisely which lines or sections have changed, and why.  
+
+    Please follow these guidelines to ensure the generated or explained code aligns well with Node.js best practices for large, maintainable projects.
+  ]],
+  ["go"] = [[
     ### Go Development Guidelines
 
     You are helping me develop a large Go (Golang) project. Please keep the following points in mind when generating or explaining code:
 
-    1. **Go Modules**  
+    1. **Go Modules & Dependency Management**  
        - Use a single `go.mod` file at the project root for module management.  
        - Ensure you use proper import paths based on the module name.  
        - If you refer to internal packages, use relative paths consistent with the module’s structure (e.g., `moduleName/internal/packageA`).  
+       - **Use the execute_command Tool for Dependencies:** Instead of manually editing version numbers in `go.mod`, please utilize the `execute_command` tool to run dependency commands (such as `go get`) to automatically fetch and update dependencies. This ensures that the correct versions are used without relying on manually provided values.
 
     2. **Package Structure**  
        - Each folder should contain exactly one package.  
        - Avoid creating multiple packages in the same folder.  
        - Use descriptive folder names and keep package names concise, following Go naming conventions.  
-       - Do not duplicate function or type names across different files in the same folder/package.  
+       - Do not duplicate function or type names across different files in the same folder/package.
 
     3. **File & Folder Organization**  
        - Organize source code in a folder hierarchy that reflects functionality. For example:
@@ -105,28 +272,26 @@ local M = {
              └── shared/
          ```
        - Keep external-facing, reusable packages in `pkg/` and internal logic in `internal/`.  
-       - Place the `main()` function in the `cmd/<appname>` folder.  
+       - Place the `main()` function in the `cmd/<appname>` folder.
 
     4. **Coding Best Practices**  
        - Maintain idiomatic Go code (e.g., short function and variable names where obvious, PascalCase for exported symbols).  
        - Keep functions short, focused, and tested.  
        - Use Go’s standard library where possible before adding third-party dependencies.  
-       - When introducing new functions or types, ensure they are uniquely named to avoid collisions.  
+       - When introducing new functions or types, ensure they are uniquely named to avoid collisions.
 
     5. **Import Management**  
        - Ensure that every import is actually used in your code.  
        - Remove unused imports to keep your code clean and maintainable.  
        - Include all necessary imports for anything referenced in your code to avoid missing imports.  
-       - Verify that any introduced import paths match your module’s structure and do not cause naming conflicts.  
+       - Verify that any introduced import paths match your module’s structure and do not cause naming conflicts.
 
     6. **Output Format**  
        - Present any generated source code as well-organized Go files, respecting the single-package-per-folder rule.  
-       - When explaining your reasoning, include any relevant architectural trade-offs and rationale (e.g., “I placed function X in package Y to keep the domain-specific logic separate from the main execution flow.”).  
-       - If you modify an existing file, specify precisely which changes or additions you are making.  
-
-    Please follow these guidelines to ensure the generated or explained code aligns well with Golang’s best practices for large, modular projects.
+       - When explaining your reasoning, include any relevant architectural trade-offs and rationale (e.g., “I placed function X in package `my_lib` to keep the business logic separate from the command-line interface.”).  
+       - If you modify existing files, specify precisely which changes or additions you are making.
   ]],
-  ["typo3-development"] = [[
+  ["typo3"] = [[
     ### TYPO3 Development Guidelines
 
     You are helping me develop a large TYPO3 project. Please keep the following points in mind when generating or explaining code:
@@ -188,10 +353,8 @@ local M = {
        - Clearly indicate where each file should be placed in the TYPO3 directory layout.  
        - When explaining your reasoning, include any relevant architectural decisions (e.g., “I created a separate extension for blog functionality to keep it isolated from the site’s main configuration.”).  
        - If you modify or extend an existing file, specify precisely which changes or additions you are making.  
-
-    Please follow these guidelines to ensure the generated or explained code aligns well with TYPO3’s best practices for large, maintainable projects.
   ]],
-  ["rust-development"] = [[
+  ["rust"] = [[
     ### Rust Development Guidelines
 
     You are helping me develop a large Rust project. Please keep the following points in mind when generating or explaining code:
@@ -231,14 +394,14 @@ local M = {
     4. **Coding & Documentation Best Practices**  
        - Write **idiomatic Rust** code:
          - Use `cargo fmt` (formatting) and `cargo clippy` (linter) to maintain consistency and quality.  
-         - Use `?` operator for error handling, prefer `Result<T, E>` over panicking unless absolutely necessary.  
+         - Use the `?` operator for error handling, preferring `Result<T, E>` over panicking unless absolutely necessary.  
        - Document your code using [Rustdoc](https://doc.rust-lang.org/rustdoc/) comments (`///` for public API) and provide examples when relevant.  
        - Write **unit tests** alongside the code (in `src/` files) and **integration tests** in a dedicated `tests/` folder.  
        - Keep functions short, focused, and ensure they have well-defined responsibilities.  
 
     5. **Reusability & Shared Code**  
        - Place common or reusable functionality into a dedicated **library** crate.  
-       - Ensure that crates depending on shared code add the appropriate `[dependencies]` or `[dev-dependencies]` in their `Cargo.toml`.  
+       - Ensure that crates depending on shared code add the appropriate `[dependencies]` or `[dev-dependencies]` in their Cargo.toml.  
        - Use the Rust standard library whenever possible before introducing external dependencies.  
 
     6. **Error Handling & Logging**  
@@ -250,60 +413,121 @@ local M = {
        - Present generated source code as well-organized Rust files, respecting the single main library or binary per crate (`lib.rs` or `main.rs`).  
        - When explaining your reasoning, include any architectural or design decisions (e.g., “I placed function X in crate `my_lib` to keep the business logic separate from the command-line interface.”).  
        - If you modify existing files, specify precisely which lines or sections have changed.  
-
-    Please follow these guidelines to ensure the generated or explained code aligns well with Rust best practices for large, modular projects.
   ]],
-  ["basic-prompt"] = [[
+  ["basic"] = [[
     ### Basic Prompt
 
-    1. **Analyze Required Files**  
-       - Check the project structure (or any provided file list).
-       - If you identify a file reference or function you do not have, **produce a YAML snippet** requesting that file’s content in full. For example:
+    You are assisting me in a coding workflow for a project (e.g., "%PROJECT_NAME%"). **Every time you inspect, modify, or execute operations on files, you must strictly follow the YAML format described below.** Under no circumstances should you output file operations in plain text or deviate from this structure.
+
+    #### Mandatory Guidelines
+
+    1. **YAML-Only File Operations**  
+       - **All operations must be provided within one single YAML block** that includes both the `project_name` and the `tools` array.
+       - If you need to ask questions or request clarifications, do so only in plain text separate from YAML. **Never include non-YAML tool commands.**
+
+    2. **Include the Project Name**  
+       - Always include:  
          ```yaml
-         project_name: my_project
-         files:
-           - path: "relative/path/to/needed_file"
+         project_name: "%PROJECT_NAME%"
          ```
-       - Do not proceed with final changes if you lack the necessary file contents.
-       - Always provide the entire file content when you do include a file.
+         This must be part of every YAML block you generate.
 
-    2. **Request File Contents**  
-       - For every file you need but don’t have, provide a short YAML snippet (like above) to retrieve it.
-       - Ask any clarifying questions outside the YAML code block.
+    3. **Operations Must Appear in the Tools Array**  
+       - List all actions (e.g., reading, editing, replacing, executing commands) as items in the `tools:` array. If multiple actions are needed, include them sequentially within the same YAML block.
 
-    3. **Provide Output YAML**  
-       - When you have all information, output the final YAML in this format:
+    4. **Read Before Write Rule**  
+       - **Do not perform any write operations (using `edit_file` or `replace_in_file`) on an existing file unless you have already read its content in the current session using a `read_file` operation.**  
+       - For new files (files that do not yet exist in the project), this rule does not apply.
+       - If yo already got the file contents in the first prompt, this rule does not apply.
+       - **Never** mix read_file with edit_file or replace_in_file in the same YAML block.
+
+    5. **File Inspection Before Modification**  
+       - When you need to inspect a file’s contents, always use the following YAML format:
          ```yaml
-         project_name: my_project
+         project_name: "%PROJECT_NAME%"
+         tools:
+           - tool: "read_file"
+             path: "relative/path/to/file"
+         ```
+       - Use the information from this operation to decide if and how to modify the file.
 
-         files:
-           - path: "src/main.py"
+    6. **Modifying Files**  
+       - To modify a file that you have already read, use:
+         ```yaml
+         project_name: "%PROJECT_NAME%"
+         tools:
+           - tool: "edit_file"
+             path: "relative/path/to/file"
              content: |
-               # Updated main function
-               def main():
-                   print("Hello, World!")
-
-           - path: "docs/README.md"
-             delete: true
+               # Full updated file content here
+         ```
+       - Alternatively, for incremental changes, use:
+         ```yaml
+         project_name: "%PROJECT_NAME%"
+         tools:
+           - tool: "replace_in_file"
+             path: "relative/path/to/file"
+             replacements:
+               - search: "old text"
+                 replace: "new text"
          ```
-       - `project_name` must match the project’s configured name.
-       - Each file item in `files` must have `path` and either `content` or `delete: true`.
 
-    4. **Explain Changes**  
-       - After the final YAML, add a brief explanation of the modifications (outside the YAML).
+    7. **Executing Commands**  
+       - To run any shell command (e.g., testing, listing files), use:
+         ```yaml
+         project_name: "%PROJECT_NAME%"
+         tools:
+           - tool: "execute_command"
+             command: "shell command here"
+         ```
 
-    5. **Iterate as Needed**  
-       - If further context or changes are required, repeat steps to request files or clarifications.
+    8. **General Process**
+       - **Step 1: Gather Context / Ask Questions**  
+         If any detail is unclear (such as file content or operation intent), ask your clarifying questions in plain text (not in YAML).
+       - **Step 2: Inspect Files**  
+         Always use `read_file` to check file content before modifying.
+       - **Step 3: Repeat Steps 1 & 2 as Needed**  
+         If further context is required, ask questions and read files again.
+       - **Step 4: Make Changes**  
+         Only after reading the file, proceed to use `edit_file` or `replace_in_file`.
+       - **Step 5: Execute Commands if Needed**  
+         Use `execute_command` as necessary, always within the YAML block.
+       - **Step 6: Tell that request is complete**  
+         Once all operations are done, confirm that the request is complete with a little summary.
+       - **Step 7: Repeat other steps as necessary**
 
-    6. **Important Notes**  
-       - Never modify or delete a file you haven’t explicitly requested or received.
-       - Use comments in code only to clarify code logic, not to explain your thought process.
-       - Explanations go outside the YAML.
+    #### Example YAML Block
 
-    7. **Best Practices**  
-       - Keep file paths accurate to avoid mistakes during implementation.
-       - Maintain a clear, logical structure for your changes.
-       - Use the final YAML consistently for clarity.
+    ```yaml
+    project_name: "%PROJECT_NAME%"
+    tools:
+      - tool: "read_file"
+        path: "relative/path/to/file"
+
+      - tool: "replace_in_file"
+        path: "relative/path/to/file"
+        replacements:
+          - search: "old text"
+            replace: "new text"
+
+      - tool: "edit_file"
+        path: "relative/path/to/file"
+        content: |
+          # Full updated file content here
+
+      - tool: "execute_command"
+        command: "ls -la"
+    ```
+
+    #### Important Reminders
+
+    - **Always** include the `project_name` and the full YAML block with the `tools` array when calling operations.
+    - **Never** write or modify a file without first having read its content during the current session (unless the file is new).
+    - **Do not** produce any tool command output that is not strictly formatted as YAML.
+
+    ---
+
+    This revised prompt ensures that during execution the placeholder "%PROJECT_NAME%" is replaced with the actual project name from the current configuration.
   ]],
   ["secure-coding"] = [[
     ### Secure Coding Guidelines
@@ -370,33 +594,10 @@ local M = {
   ["step-prompt"] = [[
     It appears this request might exceed the model's prompt character limit if done all at once.
     Please break down the tasks into smaller steps and handle them one by one.
-    At each step, we'll provide relevant files or context if needed.
     Thank you!
   ]],
   ["file-selection-instructions"] = [[
     Delete lines for directories you do NOT want, then save & close (e.g. :wq, :x, or :bd)
-  ]],
-  ["debug-commands-info"] = [[
-    ### Debugging Commands
-
-    If you need debugging commands, include them in your YAML response as follows:
-
-    ```yaml
-    commands:
-      - command: "ls"
-        args: ["-l", "path/to/directory"]
-
-      - command: "grep"
-        args: ["-r", "searchString", "path/to/file/or/directory"]
-    ```
-
-    The "ls" command uses the system's 'ls' command to list directory contents. You can pass flags or additional arguments in `args`.
-    The "grep" command searches for a given pattern in files or directories, again receiving flags or additional arguments in `args`.
-    If you omit `args` for grep, you can still use the older format with `pattern` and `target` for backward compatibility.
-
-    This are the only two commands supported at the moment.
-
-    When these commands are present and `enable_debug_commands` is true, I'll execute them and return the results in the clipboard.
   ]]
 }
 

lua/chatgpt_nvim/tools/edit_file.lua

@@ -0,0 +1,37 @@
+local handler = require("chatgpt_nvim.handler")
+local robust_lsp = require("chatgpt_nvim.tools.lsp_robust_diagnostics")
+
+local M = {}
+
+M.run = function(tool_call, conf, prompt_user_tool_accept, is_subpath, read_file)
+  local path = tool_call.path
+  local new_content = tool_call.content
+
+  if not path or not new_content then
+    return "[edit_file] Missing 'path' or 'content'."
+  end
+  local root = vim.fn.getcwd()
+
+  if not is_subpath(root, path) then
+    return string.format("Tool [edit_file for '%s'] REJECTED. Path outside project root.", path)
+  end
+
+  -- 1) Write the new content
+  handler.write_file(path, new_content, conf)
+
+  local msg = {}
+  msg[#msg+1] = string.format("Tool [edit_file for '%s'] Result:\nThe content was successfully saved to %s.", path, path)
+  msg[#msg+1] = "\nHere is the full, updated content of the file that was saved:\n"
+  msg[#msg+1] = string.format("<final_file_content path=\"%s\">\n%s\n</final_file_content>", path, new_content)
+  msg[#msg+1] = "\nIMPORTANT: For any future changes to this file, use the final_file_content shown above as your reference.\n"
+
+  -- 2) If auto_lint => run robust LSP approach
+  if conf.auto_lint then
+    local diag_str = robust_lsp.lsp_check_file_content(path, new_content, 3000) -- 3s wait
+    msg[#msg+1] = "\n" .. diag_str
+  end
+
+  return table.concat(msg, "")
+end
+
+return M

lua/chatgpt_nvim/tools/execute_command.lua

@@ -0,0 +1,36 @@
+local M = {}
+
+M.run = function(tool_call, conf, prompt_user_tool_accept, is_subpath, read_file)
+  -- Validate the command exists
+  local cmd = tool_call.command
+  if not cmd then
+    return "[execute_command] Missing 'command'."
+  end
+
+  -- Capture stderr and stdout together by redirecting stderr to stdout
+  -- This will help diagnose if there's an error causing no output
+  cmd = cmd .. " 2>&1"
+
+  -- Attempt to popen the command
+  local handle = io.popen(cmd, "r")
+  if not handle then
+    return string.format("Tool [execute_command '%s'] FAILED to popen.", cmd)
+  end
+
+  -- Read the full output (stdout + stderr)
+  local result = handle:read("*a") or ""
+
+  -- Attempt to close, capturing exit info
+  local _, exit_reason, exit_code = handle:close()
+
+  -- Provide a richer summary including exit code and reason
+  return string.format(
+    "Tool [execute_command '%s'] exited with code %s (%s)\n%s",
+    cmd,
+    tostring(exit_code),
+    tostring(exit_reason),
+    result
+  )
+end
+
+return M

lua/chatgpt_nvim/tools/init.lua

@@ -0,0 +1,39 @@
+local read_file_tool = require("chatgpt_nvim.tools.read_file")
+local edit_file_tool = require("chatgpt_nvim.tools.edit_file")
+local replace_in_file_tool = require("chatgpt_nvim.tools.replace_in_file")
+local execute_command_tool = require("chatgpt_nvim.tools.execute_command")
+
+local M = {}
+
+-- We can store a table of available tools here
+M.available_tools = {
+  {
+    name = "read_file",
+    usage = "Retrieve the contents of a file. Provide { tool='read_file', path='...' }",
+    explanation = "Use this to read file content directly from the disk."
+  },
+  {
+    name = "edit_file",
+    usage = "Overwrite an entire file's content. Provide { tool='edit_file', path='...', content='...' }, Allways include the whole file content",
+    explanation = "Use this when you want to replace a file with new content."
+  },
+  {
+    name = "replace_in_file",
+    usage = "Perform a search-and-replace. Provide { tool='replace_in_file', path='...', replacements=[ { search='...', replace='...' }, ... ] }",
+    explanation = "Use this to apply incremental changes without fully overwriting the file."
+  },
+  {
+    name = "execute_command",
+    usage = "Run a shell command. Provide { tool='execute_command', command='...' }",
+    explanation = "Just run one single command per tool invocation, without comment. It must be a single line. Use with caution, especially for destructive operations (rm, sudo, etc.)."
+  },
+}
+
+M.tools_by_name = {
+  read_file = read_file_tool,
+  edit_file = edit_file_tool,
+  replace_in_file = replace_in_file_tool,
+  execute_command = execute_command_tool
+}
+
+return M

lua/chatgpt_nvim/tools/lsp_robust_diagnostics.lua

@@ -0,0 +1,175 @@
+local api = vim.api
+local lsp = vim.lsp
+
+local M = {}
+
+local function guess_filetype(path)
+  local ext = path:match("^.+%.([^./\\]+)$")
+  if not ext then return nil end
+  ext = ext:lower()
+  local map = {
+    lua = "lua",
+    go  = "go",
+    rs  = "rust",
+    js  = "javascript",
+    jsx = "javascriptreact",
+    ts  = "typescript",
+    tsx = "typescriptreact",
+    php = "php",
+  }
+  return map[ext]
+end
+
+local function create_scratch_buffer(path, content)
+  -- Create a unique buffer name so we never clash with an existing one
+  local scratch_name = string.format("chatgpt-scratch://%s#%d", path, math.random(100000, 999999))
+
+  local bufnr = api.nvim_create_buf(false, true)
+  if bufnr == 0 then
+    return nil
+  end
+
+  -- Assign the unique name to the buffer
+  api.nvim_buf_set_name(bufnr, scratch_name)
+
+  -- Convert content string to lines
+  local lines = {}
+  for line in (content.."\n"):gmatch("(.-)\n") do
+    table.insert(lines, line)
+  end
+  api.nvim_buf_set_lines(bufnr, 0, -1, false, lines)
+
+  -- Mark it as a scratch buffer
+  api.nvim_buf_set_option(bufnr, "bufhidden", "wipe")
+  api.nvim_buf_set_option(bufnr, "swapfile", false)
+
+  return bufnr
+end
+
+local function attach_existing_lsp_client(bufnr, filetype)
+  local clients = lsp.get_active_clients()
+  if not clients or #clients == 0 then
+    return nil, "No active LSP clients"
+  end
+
+  for _, client in ipairs(clients) do
+    local ft_conf = client.config.filetypes or {}
+    if vim.tbl_contains(ft_conf, filetype) then
+      if not client.attached_buffers[bufnr] then
+        vim.lsp.buf_attach_client(bufnr, client.id)
+      end
+      return client.id
+    end
+  end
+  return nil, ("No active LSP client supports filetype '%s'"):format(filetype)
+end
+
+local function send_did_open(bufnr, client_id, path, filetype)
+  local client = lsp.get_client_by_id(client_id)
+  if not client then
+    return "Invalid client ID."
+  end
+
+  local text = table.concat(api.nvim_buf_get_lines(bufnr, 0, -1, false), "\n")
+  -- Even though the buffer name is unique, the LSP server sees this file as if at 'path'
+  local uri = vim.uri_from_fname(path)
+
+  local didOpenParams = {
+    textDocument = {
+      uri = uri,
+      languageId = filetype,
+      version = 0,
+      text = text,
+    }
+  }
+  client.rpc.notify("textDocument/didOpen", didOpenParams)
+  return nil
+end
+
+local function send_did_change(bufnr, client_id, path)
+  local client = lsp.get_client_by_id(client_id)
+  if not client then return end
+
+  local text = table.concat(api.nvim_buf_get_lines(bufnr, 0, -1, false), "\n")
+  local uri = vim.uri_from_fname(path)
+  local version = 1
+
+  client.rpc.notify("textDocument/didChange", {
+    textDocument = {
+      uri = uri,
+      version = version,
+    },
+    contentChanges = {
+      { text = text }
+    }
+  })
+end
+
+-- Use vim.wait to allow the event loop to process LSP diagnostics
+local function wait_for_diagnostics(bufnr, timeout_ms)
+  local done = false
+  local result_diags = {}
+
+  local augrp = api.nvim_create_augroup("chatgpt_lsp_diag_"..bufnr, { clear = true })
+  api.nvim_create_autocmd("DiagnosticChanged", {
+    group = augrp,
+    callback = function(args)
+      if args.buf == bufnr then
+        local diags = vim.diagnostic.get(bufnr)
+        result_diags = diags
+        done = true
+      end
+    end
+  })
+
+  vim.wait(timeout_ms, function()
+    return done
+  end, 50)
+
+  pcall(api.nvim_del_augroup_by_id, augrp)
+  return result_diags
+end
+
+local function diagnostics_to_string(diags)
+  if #diags == 0 then
+    return "No LSP diagnostics reported."
+  end
+  local lines = { "--- LSP Diagnostics ---" }
+  for _, d in ipairs(diags) do
+    local sev = vim.diagnostic.severity[d.severity] or d.severity
+    lines[#lines+1] = string.format("Line %d [%s]: %s", d.lnum + 1, sev, d.message)
+  end
+  return table.concat(lines, "\n")
+end
+
+function M.lsp_check_file_content(path, new_content, timeout_ms)
+  local filetype = guess_filetype(path) or "plaintext"
+
+  local bufnr = create_scratch_buffer(path, new_content)
+  if not bufnr then
+    return "(LSP) Could not create scratch buffer."
+  end
+
+  local client_id, err = attach_existing_lsp_client(bufnr, filetype)
+  if not client_id then
+    api.nvim_buf_delete(bufnr, { force = true })
+    return "(LSP) " .. (err or "No suitable LSP client.")
+  end
+
+  local err2 = send_did_open(bufnr, client_id, path, filetype)
+  if err2 then
+    api.nvim_buf_delete(bufnr, { force = true })
+    return "(LSP) " .. err2
+  end
+
+  -- Optionally do a didChange with consistent URI
+  send_did_change(bufnr, client_id, path)
+
+  local diags = wait_for_diagnostics(bufnr, timeout_ms or 2000)
+  local diag_str = diagnostics_to_string(diags)
+
+  api.nvim_buf_delete(bufnr, { force = true })
+  return diag_str
+end
+
+return M

lua/chatgpt_nvim/tools/manager.lua

@@ -0,0 +1,79 @@
+local tools_module = require("chatgpt_nvim.tools")
+
+local M = {}
+
+local function is_destructive_command(cmd)
+  if not cmd then return false end
+  local destructive_list = { "rm", "sudo", "mv", "cp" }
+  for _, keyword in ipairs(destructive_list) do
+    if cmd:match("(^" .. keyword .. "[%s$])") or cmd:match("[%s]" .. keyword .. "[%s$]") then
+      return true
+    end
+  end
+  return false
+end
+
+local function prompt_user_tool_accept(tool_call, conf)
+  local auto_accept = conf.tool_auto_accept[tool_call.tool]
+
+  -- If this is an execute_command and we see it's destructive, force a user prompt
+  if tool_call.tool == "execute_command" and auto_accept then
+    if is_destructive_command(tool_call.command) then
+      auto_accept = false
+    end
+  end
+
+  if auto_accept then
+    -- If auto-accepted and not destructive, no prompt needed
+    return true
+  else
+    -- Build some context about the tool request
+    local msg = ("Tool request: %s\n"):format(tool_call.tool or "unknown")
+    if tool_call.path then
+      msg = msg .. ("Path: %s\n"):format(tool_call.path)
+    end
+    if tool_call.command then
+      msg = msg .. ("Command: %s\n"):format(tool_call.command)
+    end
+    if tool_call.replacements then
+      msg = msg .. ("Replacements: %s\n"):format(vim.inspect(tool_call.replacements))
+    end
+
+    msg = msg .. "Accept this tool request? [y/N]: "
+
+    -- Force a screen redraw so the user sees the prompt properly
+    vim.cmd("redraw")
+
+    local ans = vim.fn.input(msg)
+    return ans:lower() == "y"
+  end
+end
+
+local function handle_tool_calls(tools, conf, is_subpath_fn, read_file_fn)
+  local messages = {}
+
+  for _, call in ipairs(tools) do
+    local accepted = prompt_user_tool_accept(call, conf)
+    if not accepted then
+      table.insert(messages, ("Tool [%s] was rejected by user."):format(call.tool or "nil"))
+    else
+      local tool_impl = tools_module.tools_by_name[call.tool]
+      if tool_impl then
+        local msg = tool_impl.run(call, conf, prompt_user_tool_accept, is_subpath_fn, read_file_fn)
+        table.insert(messages, msg)
+      else
+        table.insert(messages, ("Unknown tool type: '%s'"):format(call.tool or "nil"))
+      end
+    end
+  end
+
+  local combined = table.concat(messages, "\n\n")
+  local limit = conf.prompt_char_limit or 8000
+  if #combined > limit then
+    return ("The combined tool output is too large (%d chars). Please break down the operations into smaller steps."):format(#combined)
+  end
+  return combined
+end
+
+M.handle_tool_calls = handle_tool_calls
+return M

lua/chatgpt_nvim/tools/read_file.lua

@@ -0,0 +1,17 @@
+local uv = vim.loop
+local M = {}
+
+M.run = function(tool_call, conf, prompt_user_tool_accept, is_subpath, read_file)
+  local path = tool_call.path
+  if not path then
+    return "[read_file] Missing 'path'."
+  end
+  local file_data = read_file(path)
+  if file_data then
+    return string.format("Tool [read_file for '%s'] Result:\n\n%s", path, file_data)
+  else
+    return string.format("Tool [read_file for '%s'] FAILED. File not found or not readable.", path)
+  end
+end
+
+return M

lua/chatgpt_nvim/tools/replace_in_file.lua

@@ -0,0 +1,84 @@
+local handler = require("chatgpt_nvim.handler")
+local robust_lsp = require("chatgpt_nvim.tools.lsp_robust_diagnostics")
+
+local M = {}
+
+-- Function to escape all Lua pattern magic characters:
+local function escape_lua_pattern(s)
+  return s:gsub("([%^%$%(%)%%%.%[%]%*%+%-%?])", "%%%1")
+end
+
+local function search_and_replace(original, replacements)
+  local updated = original
+  local info_msgs = {}
+
+  for _, r in ipairs(replacements) do
+    local search_str = r.search or ""
+    local replace_str = r.replace or ""
+
+    -- Escape special pattern chars to ensure literal matching:
+    local escaped_search = escape_lua_pattern(search_str)
+
+    local replacement_count = 0
+    updated, replacement_count = updated:gsub(escaped_search, replace_str)
+
+    -- If the string was not found, append an info message
+    if replacement_count == 0 then
+      table.insert(info_msgs, string.format(
+        "[replace_in_file Info] The string '%s' was NOT found in the file and was not replaced.",
+        search_str
+      ))
+    else
+      table.insert(info_msgs, string.format(
+        "[replace_in_file Info] The string '%s' was replaced %d time(s).",
+        search_str,
+        replacement_count
+      ))
+    end
+  end
+
+  return updated, table.concat(info_msgs, "\n")
+end
+
+M.run = function(tool_call, conf, prompt_user_tool_accept, is_subpath, read_file)
+  local path = tool_call.path
+  local replacements = tool_call.replacements or {}
+
+  if not path or #replacements == 0 then
+    return "[replace_in_file] Missing 'path' or 'replacements'."
+  end
+
+  local root = vim.fn.getcwd()
+  if not is_subpath(root, path) then
+    return string.format("Tool [replace_in_file for '%s'] REJECTED. Path outside project root.", path)
+  end
+
+  local orig_data = read_file(path)
+  if not orig_data then
+    return string.format("[replace_in_file for '%s'] FAILED. Could not read file.", path)
+  end
+
+  -- Now we get not only the updated data but also info messages about replacements
+  local updated_data, info_messages = search_and_replace(orig_data, replacements)
+  handler.write_file(path, updated_data, conf)
+
+  local msg = {}
+  msg[#msg+1] = string.format("[replace_in_file for '%s'] Result:\nThe content was successfully saved to %s.", path, path)
+  msg[#msg+1] = "\nHere is the full, updated content of the file that was saved:\n"
+  msg[#msg+1] = string.format("<final_file_content path=\"%s\">\n%s\n</final_file_content>", path, updated_data)
+  msg[#msg+1] = "\nIMPORTANT: For any future changes to this file, use the final_file_content shown above as your reference.\n"
+
+  -- If there are any info messages (strings not found, etc.), include them
+  if info_messages and info_messages ~= "" then
+    msg[#msg+1] = "\nAdditional info about the replacement operation:\n" .. info_messages .. "\n"
+  end
+
+  if conf.auto_lint then
+    local diag_str = robust_lsp.lsp_check_file_content(path, updated_data, 3000)
+    msg[#msg+1] = "\n" .. diag_str
+  end
+
+  return table.concat(msg, "")
+end
+
+return M