fix: Move Quick Refactor agent to toml file

sources/agents/agents.qrc

@@ -7,6 +7,7 @@
         <file>claude_completion.toml</file>
         <file>claude_compression.toml</file>
         <file>claude_quick_refactor.toml</file>
+        <file>claude_quick_refactor_fast.toml</file>
         <file>openai_base_chat.toml</file>
         <file>openai_chat_completions.toml</file>
         <file>openai_responses_base.toml</file>
@@ -19,6 +20,8 @@
         <file>ollama_base_fim.toml</file>
         <file>ollama_fim.toml</file>
         <file>ollama_codellama_qml_fim.toml</file>
+        <file>ollama_quick_refactor.toml</file>
+        <file>ollama_compression.toml</file>
     </qresource>
     <qresource prefix="/roles">
         <file alias="qt-cpp-developer.md">roles/qt-cpp-developer.md</file>
@@ -29,5 +32,6 @@
     <qresource prefix="/tasks">
         <file alias="chat-compressor.md">tasks/chat-compressor.md</file>
         <file alias="code-completion.md">tasks/code-completion.md</file>
+        <file alias="quick-refactor.md">tasks/quick-refactor.md</file>
     </qresource>
 </RCC>

sources/agents/claude_quick_refactor.toml

@@ -2,12 +2,17 @@ schema_version = 1
 
 extends     = "Claude Base Chat"
 name        = "Claude Quick Refactor"
-description = "Anthropic Claude — deterministic inline refactor. QuickRefactorHandler supplies the refactor system prompt; output is cleaned and inserted into the editor. No thinking; tools off."
+description = "Anthropic Claude Sonnet — agentic inline refactor with extended thinking and tools (gathers context before editing). Static output rules from :/tasks/quick-refactor.md; QuickRefactorHandler injects the live editor context (file, code, cursor/selection)."
 
-model        = "claude-sonnet-4-6"
-enable_tools = false
-tags         = ["refactor", "claude", "cloud"]
+model           = "claude-sonnet-4-6"
+enable_tools    = true
+enable_thinking = true
+tags            = ["refactor", "claude", "cloud"]
+
+system_prompt = """{{ read_file(":/tasks/quick-refactor.md") }}"""
 
 [body]
-max_tokens  = 16000
-temperature = 0.2
+max_tokens    = 16000
+temperature   = 1
+thinking      = { type = "adaptive", display = "summarized" }
+output_config = { effort = "high" }

sources/agents/claude_quick_refactor_fast.toml

@@ -0,0 +1,16 @@
+schema_version = 1
+
+extends     = "Claude Base Chat"
+name        = "Claude Quick Refactor Fast"
+description = "Anthropic Claude Haiku — fast single-shot inline refactor. Static output rules from :/tasks/quick-refactor.md; QuickRefactorHandler injects the live editor context (file, code, cursor/selection). No thinking; tools off."
+
+model           = "claude-haiku-4-5"
+enable_tools    = false
+enable_thinking = false
+tags            = ["refactor", "claude", "cloud", "fast"]
+
+system_prompt = """{{ read_file(":/tasks/quick-refactor.md") }}"""
+
+[body]
+max_tokens  = 8000
+temperature = 0.2

sources/agents/ollama_chat_completion.toml

@@ -28,4 +28,4 @@ messages = """
 num_predict = 512
 temperature = 0.2
 keep_alive  = "5m"
-stop        = ["</code_context>"]
+stop        = ["</code_context>"]

sources/agents/ollama_codellama_qml_fim.toml

@@ -16,8 +16,18 @@ file_patterns = ["*.qml"]
 prompt = """{{ tojson("<SUF>" + ctx.suffix + "<PRE>" + ctx.prefix + "<MID>") }}"""
 
 [body.options]
-temperature    = 0
-top_p          = 1
+temperature = 0
+top_p = 1
 repeat_penalty = 1.05
-num_predict    = 500
-stop           = ["<SUF>", "<PRE>", "</PRE>", "</SUF>", "< EOT >", "\\end", "<MID>", "</MID>", "##"]
+num_predict = 500
+stop = [
+  "<SUF>",
+  "<PRE>",
+  "</PRE>",
+  "</SUF>",
+  "< EOT >",
+  "\\end",
+  "<MID>",
+  "</MID>",
+  "##",
+]

sources/agents/ollama_compression.toml

@@ -0,0 +1,17 @@
+schema_version = 1
+
+extends     = "Ollama Base Chat"
+name        = "Ollama Chat Compression"
+description = "Local Ollama conversation summarization (qwen2.5-coder). Carries the summary system prompt; no tools."
+
+model        = "qwen2.5-coder:7b"
+enable_tools = false
+tags         = ["compression", "ollama", "local"]
+
+system_prompt = """{{ read_file(":/tasks/chat-compressor.md") }}"""
+
+[body.options]
+num_predict = 2048
+temperature = 0.3
+num_ctx     = 8192
+keep_alive  = "5m"

sources/agents/ollama_fim.toml

@@ -8,6 +8,6 @@ model = "qwen2.5-coder:7b"
 tags  = ["completion", "ollama", "local", "fim"]
 
 [body.options]
-num_predict = 512
+num_predict = 256
 temperature = 0.2
 keep_alive  = "5m"

sources/agents/ollama_quick_refactor.toml

@@ -0,0 +1,16 @@
+schema_version = 1
+
+extends     = "Ollama Base Chat"
+name        = "Ollama Quick Refactor"
+description = "Local Ollama deterministic inline refactor (qwen2.5-coder). Static output rules from :/tasks/quick-refactor.md; QuickRefactorHandler injects the live editor context. No tools."
+
+model        = "qwen2.5-coder:7b"
+enable_tools = false
+tags         = ["refactor", "ollama", "local"]
+
+system_prompt = """{{ read_file(":/tasks/quick-refactor.md") }}"""
+
+[body.options]
+num_predict = 2048
+temperature = 0.2
+keep_alive  = "5m"

sources/agents/tasks/quick-refactor.md

@@ -0,0 +1,19 @@
+# Output Requirements
+
+## Formatting Rules:
+- Output ONLY the code itself, without ANY explanations or descriptions
+- Do NOT include markdown code blocks (no ```, no language tags)
+- Do NOT add comments explaining what you changed
+- Do NOT repeat existing code, be precise with context
+- Do NOT send in answer <cursor> or </cursor> and other tags
+- The output must be ready to insert directly into the editor as-is
+
+## Indentation and Whitespace:
+- Use the same indentation style (spaces or tabs) as the surrounding code
+- Maintain consistent indentation for nested blocks
+- Do NOT remove or reduce the base indentation level
+
+## Code Style:
+- Match the coding style of the surrounding code (naming, spacing, braces, etc.)
+- Preserve the original code structure when possible
+- Only change what is necessary to fulfill the user's request