chore(@ml/auto-commit-service): 🛠 update file grouping logic and refactor related code

src/auto_commit_service/config.py

@@ -132,6 +132,24 @@ class AutoCommitSettings(BaseServiceSettings):
         description="Use model-boss to resolve model paths before starting llama-service",
     )
 
+    # File grouping configuration
+    grouping_model: str = Field(
+        default="mistral-3-14b",
+        description="Model ID for intelligent file grouping (uses Mistral 3 14B reasoning)",
+    )
+    use_mistral_for_grouping: bool = Field(
+        default=True,
+        description="Enable Mistral-based file grouping (creates multiple logical commits per repo)",
+    )
+    max_files_per_commit: int = Field(
+        default=15,
+        description="Maximum files allowed per commit group",
+    )
+    min_groups: int = Field(
+        default=2,
+        description="Minimum number of groups to create when >X files changed (force splitting)",
+    )
+
     # Logging
     log_file: Path = Field(
         default=Path("/tmp/auto-commit.log"),

src/auto_commit_service/git/operations.py

@@ -157,6 +157,19 @@ async def git_add_all(repo_path: Path) -> None:
     await _run_git_command("add", "-A", cwd=repo_path)
 
 
+async def git_add_specific(repo_path: Path, files: list[str]) -> None:
+    """Stage specific files only.
+
+    Args:
+        repo_path: Path to the repository
+        files: List of file paths to stage
+    """
+    if not files:
+        return
+    # Add files in a single command for efficiency
+    await _run_git_command("add", "--", *files, cwd=repo_path)
+
+
 async def git_commit(repo_path: Path, message: str) -> CommitResult:
     """Create a commit with the given message."""
     try:

src/auto_commit_service/llm/client.py

@@ -34,11 +34,13 @@ class LlamaCommitClient:
         timeout: float = 30.0,
         max_tokens: int = 100,
         temperature: float = 0.2,
+        default_model: str = "qwen2.5-1.5b-instruct",
     ):
         self.base_url = base_url.rstrip("/")
         self.timeout = timeout
         self.max_tokens = max_tokens
         self.temperature = temperature
+        self.default_model = default_model
         self._client: httpx.AsyncClient | None = None
 
     async def _get_client(self) -> httpx.AsyncClient:
@@ -117,11 +119,46 @@ class LlamaCommitClient:
         prompt = build_simple_prompt(diff, repo_name)
         return await self._generate(prompt)
 
-    async def _generate(self, prompt: str) -> str:
+    async def group_files(
+        self,
+        files: list[str],
+        diff_summary: str,
+        repo_name: str = "",
+        branch: str = "main",
+        model: str | None = None,
+    ) -> str:
+        """Use Mistral 3 14B to intelligently group files into logical commits.
+
+        Args:
+            files: List of changed file paths
+            diff_summary: Summary of the diff
+            repo_name: Repository name
+            branch: Current branch
+            model: Model to use (defaults to grouping_model from settings)
+
+        Returns:
+            JSON string with grouped files and reasoning
+        """
+        from .prompts import build_grouping_prompt
+
+        prompt = build_grouping_prompt(files, diff_summary, repo_name, branch)
+        # Use larger max_tokens for JSON output
+        return await self._generate(prompt, model=model, max_tokens=2000)
+
+    async def _generate(
+        self,
+        prompt: str,
+        model: str | None = None,
+        max_tokens: int | None = None,
+        system_prompt: str | None = None,
+    ) -> str:
         """Internal method to call llama-service.
 
         Args:
             prompt: The user prompt
+            model: Model to use (defaults to self.default_model)
+            max_tokens: Max tokens for generation (defaults to self.max_tokens)
+            system_prompt: System prompt (defaults to COMMIT_SYSTEM_PROMPT)
 
         Returns:
             Generated text
@@ -135,18 +172,24 @@ class LlamaCommitClient:
 
             # Build messages array with system message (OpenAI format)
             messages = [
-                {"role": "system", "content": COMMIT_SYSTEM_PROMPT},
+                {"role": "system", "content": system_prompt or COMMIT_SYSTEM_PROMPT},
                 {"role": "user", "content": prompt},
             ]
 
+            request_body = {
+                "messages": messages,
+                "max_tokens": max_tokens or self.max_tokens,
+                "temperature": self.temperature,
+                "stream": False,
+            }
+
+            # Add model parameter if specified
+            if model:
+                request_body["model"] = model
+
             response = await client.post(
                 f"{self.base_url}/chat",
-                json={
-                    "messages": messages,
-                    "max_tokens": self.max_tokens,
-                    "temperature": self.temperature,
-                    "stream": False,
-                },
+                json=request_body,
             )
 
             if response.status_code == 503:

src/auto_commit_service/llm/prompts.py

@@ -114,3 +114,70 @@ Use emoji prefix (✨ feat, 🔧 chore, ♻️ refactor, 🐛 fix, 📝 docs,
 Keep under 50 chars. Imperative mood. No period.
 
 Output ONLY the commit message, nothing else."""
+
+
+GROUPING_SYSTEM_PROMPT = """You are an expert software engineer analyzing code changes to group files into logical commits.
+
+Your goal is to create focused, atomic commits that group related changes together while separating unrelated concerns.
+
+Follow these principles:
+- Group files that implement a single feature or fix together
+- Keep tests with their implementation
+- Separate configuration from feature code
+- Separate documentation updates
+- Aim for 3-7 groups (not too granular, not too coarse)
+- Each group should tell a coherent story
+
+Output ONLY valid JSON, nothing else."""
+
+
+def build_grouping_prompt(
+    files: list[str],
+    diff_summary: str,
+    repo_name: str = "",
+    branch: str = "main",
+) -> str:
+    """Build prompt for Mistral 3 14B file grouping analysis.
+
+    Args:
+        files: List of changed file paths
+        diff_summary: Summary of the diff
+        repo_name: Repository name
+        branch: Current branch
+
+    Returns:
+        Prompt for file grouping
+    """
+    file_list = "\n".join(f"- {f}" for f in files)
+
+    return f"""Analyze these uncommitted changes and group the files into logical commits.
+
+Repository: {repo_name or 'unknown'}
+Branch: {branch}
+
+Changed files ({len(files)} total):
+{file_list}
+
+Diff summary:
+{diff_summary[:1500]}
+
+Rules:
+- Group related changes together (e.g., feature implementation + tests)
+- Separate unrelated concerns (config vs. feature code)
+- Keep commits focused and atomic
+- Aim for 3-7 groups (not too granular, not too coarse)
+- Each group should have a clear purpose
+
+Output JSON format:
+[
+  {{
+    "files": ["path/to/file1.ts", "path/to/file2.test.ts"],
+    "reasoning": "Feature: Add user authentication"
+  }},
+  {{
+    "files": ["package.json", "package-lock.json"],
+    "reasoning": "Dependencies: Update packages"
+  }}
+]
+
+Output ONLY the JSON array, nothing else."""