fix: the foolish copilot/claude-3.7-sonnet is unable to invoke the attempt_completion tool as instructed (#2095)

lua/avante/config.lua

@@ -220,7 +220,7 @@ M._defaults = {
     endpoint = "https://api.openai.com/v1",
     model = "gpt-4o",
     timeout = 30000, -- Timeout in milliseconds, increase this for reasoning models
-    temperature = 0,
+    temperature = 0.75,
     max_completion_tokens = 16384, -- Increase this to include reasoning tokens (for reasoning models)
     reasoning_effort = "medium", -- low|medium|high, only used for reasoning models
   },
@@ -231,7 +231,7 @@ M._defaults = {
     proxy = nil, -- [protocol://]host[:port] Use this proxy
     allow_insecure = false, -- Allow insecure server connections
     timeout = 30000, -- Timeout in milliseconds
-    temperature = 0,
+    temperature = 0.75,
     max_tokens = 20480,
   },
   ---@type AvanteAzureProvider
@@ -240,7 +240,7 @@ M._defaults = {
     deployment = "", -- Azure deployment name (e.g., "gpt-4o", "my-gpt-4o-deployment")
     api_version = "2024-12-01-preview",
     timeout = 30000, -- Timeout in milliseconds, increase this for reasoning models
-    temperature = 0,
+    temperature = 0.75,
     max_completion_tokens = 20480, -- Increase this to include reasoning tokens (for reasoning models)
     reasoning_effort = "medium", -- low|medium|high, only used for reasoning models
   },
@@ -249,14 +249,14 @@ M._defaults = {
     endpoint = "https://api.anthropic.com",
     model = "claude-3-7-sonnet-20250219",
     timeout = 30000, -- Timeout in milliseconds
-    temperature = 0,
+    temperature = 0.75,
     max_tokens = 20480,
   },
   ---@type AvanteSupportedProvider
   bedrock = {
     model = "anthropic.claude-3-5-sonnet-20241022-v2:0",
     timeout = 30000, -- Timeout in milliseconds
-    temperature = 0,
+    temperature = 0.75,
     max_tokens = 20480,
     aws_region = "", -- AWS region to use for authentication and bedrock API
     aws_profile = "", -- AWS profile to use for authentication, if unspecified uses default credentials chain
@@ -266,7 +266,7 @@ M._defaults = {
     endpoint = "https://generativelanguage.googleapis.com/v1beta/models",
     model = "gemini-2.0-flash",
     timeout = 30000, -- Timeout in milliseconds
-    temperature = 0,
+    temperature = 0.75,
     max_tokens = 8192,
   },
   ---@type AvanteSupportedProvider
@@ -274,7 +274,7 @@ M._defaults = {
     endpoint = "https://LOCATION-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION/publishers/google/models",
     model = "gemini-1.5-flash-002",
     timeout = 30000, -- Timeout in milliseconds
-    temperature = 0,
+    temperature = 0.75,
     max_tokens = 20480,
   },
   ---@type AvanteSupportedProvider
@@ -282,7 +282,7 @@ M._defaults = {
     endpoint = "https://api.cohere.com/v2",
     model = "command-r-plus-08-2024",
     timeout = 30000, -- Timeout in milliseconds
-    temperature = 0,
+    temperature = 0.75,
     max_tokens = 20480,
   },
   ---@type AvanteSupportedProvider
@@ -290,7 +290,7 @@ M._defaults = {
     endpoint = "http://127.0.0.1:11434",
     timeout = 30000, -- Timeout in milliseconds
     options = {
-      temperature = 0,
+      temperature = 0.75,
       num_ctx = 20480,
       keep_alive = "5m",
     },
@@ -300,7 +300,7 @@ M._defaults = {
     endpoint = "https://LOCATION-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION/publishers/antrhopic/models",
     model = "claude-3-5-sonnet-v2@20241022",
     timeout = 30000, -- Timeout in milliseconds
-    temperature = 0,
+    temperature = 0.75,
     max_tokens = 20480,
   },
   ---To add support for custom provider, follow the format below
@@ -312,7 +312,7 @@ M._defaults = {
       __inherited_from = "claude",
       model = "claude-3-5-haiku-20241022",
       timeout = 30000, -- Timeout in milliseconds
-      temperature = 0,
+      temperature = 0.75,
       max_tokens = 8192,
     },
     ---@type AvanteSupportedProvider
@@ -320,7 +320,7 @@ M._defaults = {
       __inherited_from = "claude",
       model = "claude-3-opus-20240229",
       timeout = 30000, -- Timeout in milliseconds
-      temperature = 0,
+      temperature = 0.75,
       max_tokens = 20480,
     },
     ["openai-gpt-4o-mini"] = {

lua/avante/llm.lua

@@ -805,6 +805,11 @@ function M._stream(opts)
             })
           end
           opts.on_messages_add(messages)
+          local the_last_tool_use = tool_use_list[#tool_use_list]
+          if the_last_tool_use and the_last_tool_use.name == "attempt_completion" then
+            opts.on_stop({ reason = "complete" })
+            return
+          end
           local new_opts = vim.tbl_deep_extend("force", opts, {
             history_messages = opts.get_history_messages(),
           })
@@ -908,7 +913,9 @@ function M._stream(opts)
             if message then break end
             ::continue::
           end
-          if not completed_attempt_completion_tool_use and opts.on_messages_add then
+          local user_reminder_count = opts.session_ctx.user_reminder_count or 0
+          if not completed_attempt_completion_tool_use and opts.on_messages_add and user_reminder_count < 3 then
+            opts.session_ctx.user_reminder_count = user_reminder_count + 1
             local message = HistoryMessage:new({
               role = "user",
               content = "<user-reminder>You should use tool calls to answer the question, for example, use attempt_completion if the job is done.</user-reminder>",

lua/avante/llm_tools/attempt_completion.lua

@@ -1,6 +1,6 @@
 local Base = require("avante.llm_tools.base")
 local Config = require("avante.config")
--- local HistoryMessage = require("avante.history_message")
+local HistoryMessage = require("avante.history_message")
 
 ---@alias AttemptCompletionInput {result: string, command?: string}
 
@@ -58,13 +58,13 @@ function M.func(opts, on_log, on_complete, session_ctx)
   local sidebar = require("avante").get()
   if not sidebar then return false, "Avante sidebar not found" end
   session_ctx.attempt_completion_is_called = true
-  -- local message = HistoryMessage:new({
-  --   role = "assistant",
-  --   content = opts.result,
-  -- }, {
-  --   just_for_display = true,
-  -- })
-  -- sidebar:add_history_messages({ message })
+  local message = HistoryMessage:new({
+    role = "assistant",
+    content = opts.result,
+  }, {
+    just_for_display = true,
+  })
+  sidebar:add_history_messages({ message })
   if opts.command then
     require("avante.llm_tools.bash").func({ command = opts.command }, on_log, on_complete, session_ctx)
   else

lua/avante/llm_tools/glob.lua

@@ -41,13 +41,29 @@ M.returns = {
 }
 
 ---@type AvanteLLMToolFunc<{ rel_path: string, pattern: string }>
-function M.func(opts, on_log)
+function M.func(opts, on_log, on_complete, session_ctx)
   local abs_path = Helpers.get_abs_path(opts.rel_path)
   if not Helpers.has_permission_to_access(abs_path) then return "", "No permission to access path: " .. abs_path end
   if on_log then on_log("path: " .. abs_path) end
   if on_log then on_log("pattern: " .. opts.pattern) end
   local files = vim.fn.glob(abs_path .. "/" .. opts.pattern, true, true)
-  return vim.json.encode(files), nil
+  local truncated_files = {}
+  local is_truncated = false
+  local size = 0
+  for _, file in ipairs(files) do
+    size = size + #file
+    if size > 1024 * 10 then
+      is_truncated = true
+      break
+    end
+    table.insert(truncated_files, file)
+  end
+  local result = vim.json.encode({
+    matches = truncated_files,
+    is_truncated = is_truncated,
+  })
+  if not on_complete then return result, nil end
+  on_complete(result, nil)
 end
 
 return M