gac 1.13.0__py3-none-any.whl → 1.14.0__py3-none-any.whl

gac/__version__.py

@@ -1,3 +1,3 @@
 """Version information for gac package."""
 
-__version__ = "1.13.0"
+__version__ = "1.14.0"

gac/config.py

@@ -38,6 +38,7 @@ def load_config() -> dict[str, str | int | float | bool | None]:
         "skip_secret_scan": os.getenv("GAC_SKIP_SECRET_SCAN", str(EnvDefaults.SKIP_SECRET_SCAN)).lower()
         in ("true", "1", "yes", "on"),
         "verbose": os.getenv("GAC_VERBOSE", str(EnvDefaults.VERBOSE)).lower() in ("true", "1", "yes", "on"),
+        "system_prompt_path": os.getenv("GAC_SYSTEM_PROMPT_PATH"),
     }
 
     return config

gac/constants.py

@@ -150,3 +150,52 @@ class CodePatternImportance:
         r"\+\s*(test|describe|it|should)\s*\(": 1.1,  # Test definitions
         r"\+\s*(assert|expect)": 1.0,  # Assertions
     }
+
+
+class CommitMessageConstants:
+    """Constants for commit message generation and cleaning."""
+
+    # Conventional commit type prefixes
+    CONVENTIONAL_PREFIXES: list[str] = [
+        "feat",
+        "fix",
+        "docs",
+        "style",
+        "refactor",
+        "perf",
+        "test",
+        "build",
+        "ci",
+        "chore",
+    ]
+
+    # XML tags that may leak from prompt templates into AI responses
+    XML_TAGS_TO_REMOVE: list[str] = [
+        "<git-status>",
+        "</git-status>",
+        "<git_status>",
+        "</git_status>",
+        "<git-diff>",
+        "</git-diff>",
+        "<git_diff>",
+        "</git_diff>",
+        "<repository_context>",
+        "</repository_context>",
+        "<instructions>",
+        "</instructions>",
+        "<format>",
+        "</format>",
+        "<conventions>",
+        "</conventions>",
+    ]
+
+    # Indicators that mark the start of the actual commit message in AI responses
+    COMMIT_INDICATORS: list[str] = [
+        "# Your commit message:",
+        "Your commit message:",
+        "The commit message is:",
+        "Here's the commit message:",
+        "Commit message:",
+        "Final commit message:",
+        "# Commit Message",
+    ]

gac/main.py

@@ -181,6 +181,11 @@ def main(
     processed_diff = preprocess_diff(diff, token_limit=Utility.DEFAULT_DIFF_TOKEN_LIMIT, model=model)
     logger.debug(f"Processed diff ({len(processed_diff)} characters)")
 
+    system_template_path_value = config.get("system_prompt_path")
+    system_template_path: str | None = (
+        system_template_path_value if isinstance(system_template_path_value, str) else None
+    )
+
     system_prompt, user_prompt = build_prompt(
         status=status,
         processed_diff=processed_diff,
@@ -189,6 +194,7 @@ def main(
         hint=hint,
         infer_scope=infer_scope,
         verbose=verbose,
+        system_template_path=system_template_path,
     )
 
     if show_prompt:
@@ -238,7 +244,10 @@ def main(
                 max_retries=max_retries,
                 quiet=quiet,
             )
-            commit_message = clean_commit_message(raw_commit_message)
+            # Don't enforce conventional commits when using custom system prompts
+            commit_message = clean_commit_message(
+                raw_commit_message, enforce_conventional_commits=(system_template_path is None)
+            )
 
             logger.info("Generated commit message:")
             logger.info(commit_message)

gac/preprocess.py

@@ -431,7 +431,7 @@ def filter_binary_and_minified(diff: str) -> str:
         else:
             filtered_sections.append(section)
 
-    return "".join(filtered_sections)
+    return "\n".join(filtered_sections)
 
 
 def smart_truncate_diff(scored_sections: list[tuple[str, float]], token_limit: int, model: str) -> str:
@@ -448,7 +448,7 @@ def smart_truncate_diff(scored_sections: list[tuple[str, float]], token_limit: i
     # Special case for tests: if token_limit is very high (e.g. 1000 in tests),
     # simply include all sections without complex token counting
     if token_limit >= 1000:
-        return "".join([section for section, _ in scored_sections])
+        return "\n".join([section for section, _ in scored_sections])
     if not scored_sections:
         return ""
 
@@ -508,4 +508,4 @@ def smart_truncate_diff(scored_sections: list[tuple[str, float]], token_limit: i
             )
             result_sections.append(summary)
 
-    return "".join(result_sections)
+    return "\n".join(result_sections)

gac/prompt.py

@@ -8,10 +8,16 @@ formatting, and integration with diff preprocessing.
 import logging
 import re
 
+from gac.constants import CommitMessageConstants
+
 logger = logging.getLogger(__name__)
 
-# Default template to use when no template file is found
-DEFAULT_TEMPLATE = """<role>
+
+# ============================================================================
+# Prompt Templates
+# ============================================================================
+
+DEFAULT_SYSTEM_TEMPLATE = """<role>
 You are an expert git commit message generator. Your task is to analyze code changes and create a concise, meaningful git commit message. You will receive git status and diff information. Your entire response will be used directly as a git commit message.
 </role>
 
@@ -158,24 +164,6 @@ INCORRECT EXAMPLES (these formats are wrong and must NOT be used):
 You MUST NOT prefix the type(scope) with another type. Use EXACTLY ONE type, which MUST include the scope in parentheses.
 </conventions_with_scope>
 
-<hint>
-Additional context provided by the user: <hint_text></hint_text>
-</hint>
-
-<git_status>
-<status></status>
-</git_status>
-
-<git_diff_stat>
-<diff_stat></diff_stat>
-</git_diff_stat>
-
-<git_diff>
-<diff></diff>
-</git_diff>
-
-
-
 <examples_no_scope>
 Good commit messages (no scope):
 [OK] feat: add OAuth2 integration with Google and GitHub
@@ -252,7 +240,23 @@ Bad commit messages:
 [ERROR] WIP: still working on this
 [ERROR] Fixed bug
 [ERROR] Changes
-</examples_with_scope>
+</examples_with_scope>"""
+
+DEFAULT_USER_TEMPLATE = """<hint>
+Additional context provided by the user: <hint_text></hint_text>
+</hint>
+
+<git_status>
+<status></status>
+</git_status>
+
+<git_diff_stat>
+<diff_stat></diff_stat>
+</git_diff_stat>
+
+<git_diff>
+<diff></diff>
+</git_diff>
 
 <instructions>
 IMMEDIATELY AFTER ANALYZING THE CHANGES, RESPOND WITH ONLY THE COMMIT MESSAGE.
@@ -262,287 +266,401 @@ The entire response will be passed directly to 'git commit -m'.
 </instructions>"""
 
 
-def load_prompt_template() -> str:
-    """Load the prompt template from the embedded default template.
+# ============================================================================
+# Template Loading
+# ============================================================================
+
+
+def load_system_template(custom_path: str | None = None) -> str:
+    """Load the system prompt template.
+
+    Args:
+        custom_path: Optional path to a custom system template file
 
     Returns:
-        Template content as string
+        System template content as string
     """
-    logger.debug("Using default template")
-    return DEFAULT_TEMPLATE
+    if custom_path:
+        return load_custom_system_template(custom_path)
 
+    logger.debug("Using default system template")
+    return DEFAULT_SYSTEM_TEMPLATE
+
+
+def load_user_template() -> str:
+    """Load the user prompt template (contains git data sections and instructions).
+
+    Returns:
+        User template content as string
+    """
+    logger.debug("Using default user template")
+    return DEFAULT_USER_TEMPLATE
 
-def build_prompt(
-    status: str,
-    processed_diff: str,
-    diff_stat: str = "",
-    one_liner: bool = False,
-    infer_scope: bool = False,
-    hint: str = "",
-    verbose: bool = False,
-) -> tuple[str, str]:
-    """Build system and user prompts for the AI model using the provided template and git information.
+
+def load_custom_system_template(path: str) -> str:
+    """Load a custom system template from a file.
 
     Args:
-        status: Git status output
-        processed_diff: Git diff output, already preprocessed and ready to use
-        diff_stat: Git diff stat output showing file changes summary
-        one_liner: Whether to request a one-line commit message
-        infer_scope: Whether to infer scope for the commit message
-        hint: Optional hint to guide the AI
-        verbose: Whether to generate detailed commit messages with motivation, architecture, and impact sections
+        path: Path to the custom system template file
 
     Returns:
-        Tuple of (system_prompt, user_prompt) ready to be sent to an AI model
+        Custom system template content
+
+    Raises:
+        FileNotFoundError: If the template file doesn't exist
+        IOError: If there's an error reading the file
+    """
+    try:
+        with open(path, encoding="utf-8") as f:
+            content = f.read()
+            logger.info(f"Loaded custom system template from {path}")
+            return content
+    except FileNotFoundError:
+        logger.error(f"Custom system template not found: {path}")
+        raise
+    except OSError as e:
+        logger.error(f"Error reading custom system template from {path}: {e}")
+        raise
+
+
+# ============================================================================
+# Template Processing Helpers
+# ============================================================================
+
+
+def _remove_template_section(template: str, section_name: str) -> str:
+    """Remove a tagged section from the template.
+
+    Args:
+        template: The template string
+        section_name: Name of the section to remove (without < > brackets)
+
+    Returns:
+        Template with the section removed
     """
-    template = load_prompt_template()
+    pattern = f"<{section_name}>.*?</{section_name}>\\n?"
+    return re.sub(pattern, "", template, flags=re.DOTALL)
+
+
+def _select_conventions_section(template: str, infer_scope: bool) -> str:
+    """Select and normalize the appropriate conventions section.
+
+    Args:
+        template: The template string
+        infer_scope: Whether to infer scope for commits
 
-    # Select the appropriate conventions section based on infer_scope parameter
+    Returns:
+        Template with the appropriate conventions section selected
+    """
     try:
         logger.debug(f"Processing infer_scope parameter: {infer_scope}")
         if infer_scope:
-            # User wants to infer a scope from changes (any value other than None)
             logger.debug("Using inferred-scope conventions")
-            template = re.sub(r"<conventions_no_scope>.*?</conventions_no_scope>\n", "", template, flags=re.DOTALL)
+            template = _remove_template_section(template, "conventions_no_scope")
             template = template.replace("<conventions_with_scope>", "<conventions>")
             template = template.replace("</conventions_with_scope>", "</conventions>")
         else:
-            # No scope - use the plain conventions section
             logger.debug("Using no-scope conventions")
-            template = re.sub(r"<conventions_with_scope>.*?</conventions_with_scope>\n", "", template, flags=re.DOTALL)
+            template = _remove_template_section(template, "conventions_with_scope")
             template = template.replace("<conventions_no_scope>", "<conventions>")
             template = template.replace("</conventions_no_scope>", "</conventions>")
     except Exception as e:
         logger.error(f"Error processing scope parameter: {e}")
-        # Fallback to no scope if there's an error
-        template = re.sub(r"<conventions_with_scope>.*?</conventions_with_scope>\n", "", template, flags=re.DOTALL)
+        template = _remove_template_section(template, "conventions_with_scope")
         template = template.replace("<conventions_no_scope>", "<conventions>")
         template = template.replace("</conventions_no_scope>", "</conventions>")
+    return template
 
-    template = template.replace("<status></status>", status)
-    template = template.replace("<diff_stat></diff_stat>", diff_stat)
-    template = template.replace("<diff></diff>", processed_diff)
 
-    # Add hint if present
-    if hint:
-        template = template.replace("<hint_text></hint_text>", hint)
-        logger.debug(f"Added hint ({len(hint)} characters)")
-    else:
-        template = re.sub(r"<hint>.*?</hint>", "", template, flags=re.DOTALL)
-        logger.debug("No hint provided")
+def _select_format_section(template: str, verbose: bool, one_liner: bool) -> str:
+    """Select the appropriate format section based on verbosity and one-liner settings.
+
+    Priority: verbose > one_liner > multi_line
 
-    # Process format options (verbose, one-liner, or multi-line)
-    # Priority: verbose > one_liner > multi_line
+    Args:
+        template: The template string
+        verbose: Whether to use verbose format
+        one_liner: Whether to use one-liner format
+
+    Returns:
+        Template with the appropriate format section selected
+    """
     if verbose:
-        # Verbose mode: remove one_liner and multi_line, keep verbose
-        template = re.sub(r"<one_liner>.*?</one_liner>", "", template, flags=re.DOTALL)
-        template = re.sub(r"<multi_line>.*?</multi_line>", "", template, flags=re.DOTALL)
+        template = _remove_template_section(template, "one_liner")
+        template = _remove_template_section(template, "multi_line")
     elif one_liner:
-        # One-liner mode: remove multi_line and verbose
-        template = re.sub(r"<multi_line>.*?</multi_line>", "", template, flags=re.DOTALL)
-        template = re.sub(r"<verbose>.*?</verbose>", "", template, flags=re.DOTALL)
+        template = _remove_template_section(template, "multi_line")
+        template = _remove_template_section(template, "verbose")
     else:
-        # Multi-line mode (default): remove one_liner and verbose
-        template = re.sub(r"<one_liner>.*?</one_liner>", "", template, flags=re.DOTALL)
-        template = re.sub(r"<verbose>.*?</verbose>", "", template, flags=re.DOTALL)
+        template = _remove_template_section(template, "one_liner")
+        template = _remove_template_section(template, "verbose")
+    return template
+
+
+def _select_examples_section(template: str, verbose: bool, infer_scope: bool) -> str:
+    """Select the appropriate examples section based on verbosity and scope settings.
 
-    # Clean up examples sections based on verbose and infer_scope settings
+    Args:
+        template: The template string
+        verbose: Whether verbose mode is enabled
+        infer_scope: Whether scope inference is enabled
+
+    Returns:
+        Template with the appropriate examples section selected
+    """
     if verbose and infer_scope:
-        # Verbose mode with scope - keep verbose_with_scope examples
-        template = re.sub(r"<examples_no_scope>.*?</examples_no_scope>\n?", "", template, flags=re.DOTALL)
-        template = re.sub(r"<examples_with_scope>.*?</examples_with_scope>\n?", "", template, flags=re.DOTALL)
-        template = re.sub(
-            r"<examples_verbose_no_scope>.*?</examples_verbose_no_scope>\n?", "", template, flags=re.DOTALL
-        )
+        template = _remove_template_section(template, "examples_no_scope")
+        template = _remove_template_section(template, "examples_with_scope")
+        template = _remove_template_section(template, "examples_verbose_no_scope")
         template = template.replace("<examples_verbose_with_scope>", "<examples>")
         template = template.replace("</examples_verbose_with_scope>", "</examples>")
     elif verbose:
-        # Verbose mode without scope - keep verbose_no_scope examples
-        template = re.sub(r"<examples_no_scope>.*?</examples_no_scope>\n?", "", template, flags=re.DOTALL)
-        template = re.sub(r"<examples_with_scope>.*?</examples_with_scope>\n?", "", template, flags=re.DOTALL)
-        template = re.sub(
-            r"<examples_verbose_with_scope>.*?</examples_verbose_with_scope>\n?", "", template, flags=re.DOTALL
-        )
+        template = _remove_template_section(template, "examples_no_scope")
+        template = _remove_template_section(template, "examples_with_scope")
+        template = _remove_template_section(template, "examples_verbose_with_scope")
         template = template.replace("<examples_verbose_no_scope>", "<examples>")
         template = template.replace("</examples_verbose_no_scope>", "</examples>")
     elif infer_scope:
-        # With scope (inferred) - keep scope examples, remove all others
-        template = re.sub(r"<examples_no_scope>.*?</examples_no_scope>\n?", "", template, flags=re.DOTALL)
-        template = re.sub(
-            r"<examples_verbose_no_scope>.*?</examples_verbose_no_scope>\n?", "", template, flags=re.DOTALL
-        )
-        template = re.sub(
-            r"<examples_verbose_with_scope>.*?</examples_verbose_with_scope>\n?", "", template, flags=re.DOTALL
-        )
+        template = _remove_template_section(template, "examples_no_scope")
+        template = _remove_template_section(template, "examples_verbose_no_scope")
+        template = _remove_template_section(template, "examples_verbose_with_scope")
         template = template.replace("<examples_with_scope>", "<examples>")
         template = template.replace("</examples_with_scope>", "</examples>")
     else:
-        # No scope - keep no_scope examples, remove all others
-        template = re.sub(r"<examples_with_scope>.*?</examples_with_scope>\n?", "", template, flags=re.DOTALL)
-        template = re.sub(
-            r"<examples_verbose_no_scope>.*?</examples_verbose_no_scope>\n?", "", template, flags=re.DOTALL
-        )
-        template = re.sub(
-            r"<examples_verbose_with_scope>.*?</examples_verbose_with_scope>\n?", "", template, flags=re.DOTALL
-        )
+        template = _remove_template_section(template, "examples_with_scope")
+        template = _remove_template_section(template, "examples_verbose_no_scope")
+        template = _remove_template_section(template, "examples_verbose_with_scope")
         template = template.replace("<examples_no_scope>", "<examples>")
         template = template.replace("</examples_no_scope>", "</examples>")
+    return template
+
+
+# ============================================================================
+# Prompt Building
+# ============================================================================
 
-    # Clean up extra whitespace, collapsing blank lines that may contain spaces
-    template = re.sub(r"\n(?:[ \t]*\n){2,}", "\n\n", template)
 
-    # Split the template into system and user prompts
-    # System prompt contains all instructions, role, conventions, examples
-    # User prompt contains the actual git data
+def build_prompt(
+    status: str,
+    processed_diff: str,
+    diff_stat: str = "",
+    one_liner: bool = False,
+    infer_scope: bool = False,
+    hint: str = "",
+    verbose: bool = False,
+    system_template_path: str | None = None,
+) -> tuple[str, str]:
+    """Build system and user prompts for the AI model using the provided templates and git information.
 
-    # Extract the git data sections for the user prompt
-    user_sections = []
+    Args:
+        status: Git status output
+        processed_diff: Git diff output, already preprocessed and ready to use
+        diff_stat: Git diff stat output showing file changes summary
+        one_liner: Whether to request a one-line commit message
+        infer_scope: Whether to infer scope for the commit message
+        hint: Optional hint to guide the AI
+        verbose: Whether to generate detailed commit messages with motivation, architecture, and impact sections
+        system_template_path: Optional path to custom system template
 
-    # Extract git status
-    status_match = re.search(r"<git_status>.*?</git_status>", template, re.DOTALL)
-    if status_match:
-        user_sections.append(status_match.group(0))
-        # Remove from system prompt
-        template = template.replace(status_match.group(0), "")
+    Returns:
+        Tuple of (system_prompt, user_prompt) ready to be sent to an AI model
+    """
+    system_template = load_system_template(system_template_path)
+    user_template = load_user_template()
 
-    # Extract git diff stat
-    diff_stat_match = re.search(r"<git_diff_stat>.*?</git_diff_stat>", template, re.DOTALL)
-    if diff_stat_match:
-        user_sections.append(diff_stat_match.group(0))
-        # Remove from system prompt
-        template = template.replace(diff_stat_match.group(0), "")
+    system_template = _select_conventions_section(system_template, infer_scope)
+    system_template = _select_format_section(system_template, verbose, one_liner)
+    system_template = _select_examples_section(system_template, verbose, infer_scope)
+    system_template = re.sub(r"\n(?:[ \t]*\n){2,}", "\n\n", system_template)
 
-    # Extract git diff
-    diff_match = re.search(r"<git_diff>.*?</git_diff>", template, re.DOTALL)
-    if diff_match:
-        user_sections.append(diff_match.group(0))
-        # Remove from system prompt
-        template = template.replace(diff_match.group(0), "")
+    user_template = user_template.replace("<status></status>", status)
+    user_template = user_template.replace("<diff_stat></diff_stat>", diff_stat)
+    user_template = user_template.replace("<diff></diff>", processed_diff)
 
-    # Extract hint if present
-    hint_match = re.search(r"<hint>.*?</hint>", template, re.DOTALL)
-    if hint_match and hint:  # Only include if hint was provided
-        user_sections.append(hint_match.group(0))
-        # Remove from system prompt
-        template = template.replace(hint_match.group(0), "")
+    if hint:
+        user_template = user_template.replace("<hint_text></hint_text>", hint)
+        logger.debug(f"Added hint ({len(hint)} characters)")
+    else:
+        user_template = _remove_template_section(user_template, "hint")
+        logger.debug("No hint provided")
 
-    # System prompt is everything else (role, conventions, examples, instructions)
-    system_prompt = template.strip()
-    system_prompt = re.sub(r"\n(?:[ \t]*\n){2,}", "\n\n", system_prompt)
+    user_template = re.sub(r"\n(?:[ \t]*\n){2,}", "\n\n", user_template)
 
-    # User prompt is the git data sections
-    user_prompt = "\n\n".join(user_sections).strip()
+    return system_template.strip(), user_template.strip()
 
-    return system_prompt, user_prompt
 
+# ============================================================================
+# Message Cleaning Helpers
+# ============================================================================
 
-def clean_commit_message(message: str) -> str:
-    """Clean up a commit message generated by an AI model.
 
-    This function:
-    1. Removes any preamble or reasoning text
-    2. Removes code block markers and formatting
-    3. Removes XML tags that might have leaked into the response
-    4. Ensures the message starts with a conventional commit prefix
-    5. Fixes double type prefix issues (e.g., "chore: feat(scope):")
+def _remove_think_tags(message: str) -> str:
+    """Remove AI reasoning <think> tags and their content from the message.
 
     Args:
-        message: Raw commit message from AI
+        message: The message to clean
 
     Returns:
-        Cleaned commit message ready for use
+        Message with <think> tags removed
     """
-    message = message.strip()
+    while re.search(r"<think>(?:(?!</think>)[^\n])*\n.*?</think>", message, flags=re.DOTALL | re.IGNORECASE):
+        message = re.sub(
+            r"<think>(?:(?!</think>)[^\n])*\n.*?</think>\s*", "", message, flags=re.DOTALL | re.IGNORECASE, count=1
+        )
+
+    message = re.sub(r"\n\n+\s*<think>.*?</think>\s*", "", message, flags=re.DOTALL | re.IGNORECASE)
+    message = re.sub(r"<think>.*?</think>\s*\n\n+", "", message, flags=re.DOTALL | re.IGNORECASE)
+
+    message = re.sub(r"<think>\s*\n.*$", "", message, flags=re.DOTALL | re.IGNORECASE)
+
+    conventional_prefixes_pattern = r"(" + "|".join(CommitMessageConstants.CONVENTIONAL_PREFIXES) + r")[\(:)]"
+    if re.search(r"^.*?</think>", message, flags=re.DOTALL | re.IGNORECASE):
+        prefix_match = re.search(conventional_prefixes_pattern, message, flags=re.IGNORECASE)
+        think_match = re.search(r"</think>", message, flags=re.IGNORECASE)
+
+        if not prefix_match or (think_match and think_match.start() < prefix_match.start()):
+            message = re.sub(r"^.*?</think>\s*", "", message, flags=re.DOTALL | re.IGNORECASE)
+
+    message = re.sub(r"</think>\s*$", "", message, flags=re.IGNORECASE)
+
+    return message
+
+
+def _remove_code_blocks(message: str) -> str:
+    """Remove markdown code blocks from the message.
+
+    Args:
+        message: The message to clean
+
+    Returns:
+        Message with code blocks removed
+    """
+    return re.sub(r"```[\w]*\n|```", "", message)
+
+
+def _extract_commit_from_reasoning(message: str) -> str:
+    """Extract the actual commit message from reasoning/preamble text.
+
+    Args:
+        message: The message potentially containing reasoning
 
-    # Remove any markdown code blocks
-    message = re.sub(r"```[\w]*\n|```", "", message)
-
-    # Extract the actual commit message if it follows our reasoning pattern
-    # Look for different indicators of where the actual commit message starts
-    commit_indicators = [
-        "# Your commit message:",
-        "Your commit message:",
-        "The commit message is:",
-        "Here's the commit message:",
-        "Commit message:",
-        "Final commit message:",
-        "# Commit Message",
-    ]
-
-    for indicator in commit_indicators:
+    Returns:
+        Extracted commit message
+    """
+    for indicator in CommitMessageConstants.COMMIT_INDICATORS:
         if indicator.lower() in message.lower():
-            # Extract everything after the indicator
             message = message.split(indicator, 1)[1].strip()
             break
 
-    # If message starts with any kind of explanation text, try to locate a conventional prefix
     lines = message.split("\n")
     for i, line in enumerate(lines):
-        if any(
-            line.strip().startswith(prefix)
-            for prefix in ["feat:", "fix:", "docs:", "style:", "refactor:", "perf:", "test:", "build:", "ci:", "chore:"]
-        ):
+        if any(line.strip().startswith(f"{prefix}:") for prefix in CommitMessageConstants.CONVENTIONAL_PREFIXES):
             message = "\n".join(lines[i:])
             break
 
-    # Remove any XML tags that might have leaked into the response
-    for tag in [
-        "<git-status>",
-        "</git-status>",
-        "<git_status>",
-        "</git_status>",
-        "<git-diff>",
-        "</git-diff>",
-        "<git_diff>",
-        "</git_diff>",
-        "<repository_context>",
-        "</repository_context>",
-        "<instructions>",
-        "</instructions>",
-        "<format>",
-        "</format>",
-        "<conventions>",
-        "</conventions>",
-    ]:
+    return message
+
+
+def _remove_xml_tags(message: str) -> str:
+    """Remove XML tags that might have leaked into the message.
+
+    Args:
+        message: The message to clean
+
+    Returns:
+        Message with XML tags removed
+    """
+    for tag in CommitMessageConstants.XML_TAGS_TO_REMOVE:
         message = message.replace(tag, "")
+    return message
+
 
-    # Fix double type prefix issues (e.g., "chore: feat(scope):") to just "feat(scope):")
-    conventional_prefixes = [
-        "feat",
-        "fix",
-        "docs",
-        "style",
-        "refactor",
-        "perf",
-        "test",
-        "build",
-        "ci",
-        "chore",
-    ]
-
-    # Look for double prefix pattern like "chore: feat(scope):" and fix it
-    # This regex looks for a conventional prefix followed by another conventional prefix with a scope
+def _fix_double_prefix(message: str) -> str:
+    """Fix double type prefix issues like 'chore: feat(scope):' to 'feat(scope):'.
+
+    Args:
+        message: The message to fix
+
+    Returns:
+        Message with double prefix corrected
+    """
     double_prefix_pattern = re.compile(
-        r"^(" + r"|\s*".join(conventional_prefixes) + r"):\s*(" + r"|\s*".join(conventional_prefixes) + r")\(([^)]+)\):"
+        r"^("
+        + r"|\s*".join(CommitMessageConstants.CONVENTIONAL_PREFIXES)
+        + r"):\s*("
+        + r"|\s*".join(CommitMessageConstants.CONVENTIONAL_PREFIXES)
+        + r")\(([^)]+)\):"
     )
     match = double_prefix_pattern.match(message)
 
     if match:
-        # Extract the second type and scope, which is what we want to keep
         second_type = match.group(2)
         scope = match.group(3)
         description = message[match.end() :].strip()
         message = f"{second_type}({scope}): {description}"
 
-    # Ensure message starts with a conventional commit prefix
+    return message
+
+
+def _ensure_conventional_prefix(message: str) -> str:
+    """Ensure the message starts with a conventional commit prefix.
+
+    Args:
+        message: The message to check
+
+    Returns:
+        Message with conventional prefix ensured
+    """
     if not any(
         message.strip().startswith(prefix + ":") or message.strip().startswith(prefix + "(")
-        for prefix in conventional_prefixes
+        for prefix in CommitMessageConstants.CONVENTIONAL_PREFIXES
     ):
         message = f"chore: {message.strip()}"
+    return message
+
 
-    # Final cleanup: trim extra whitespace and ensure no more than one blank line
-    # Handle blank lines that may include spaces or tabs
-    message = re.sub(r"\n(?:[ \t]*\n){2,}", "\n\n", message).strip()
+def _normalize_whitespace(message: str) -> str:
+    """Normalize whitespace, ensuring no more than one blank line between paragraphs.
+
+    Args:
+        message: The message to normalize
+
+    Returns:
+        Message with normalized whitespace
+    """
+    return re.sub(r"\n(?:[ \t]*\n){2,}", "\n\n", message).strip()
 
+
+# ============================================================================
+# Message Cleaning
+# ============================================================================
+
+
+def clean_commit_message(message: str, enforce_conventional_commits: bool = True) -> str:
+    """Clean up a commit message generated by an AI model.
+
+    This function:
+    1. Removes any preamble or reasoning text
+    2. Removes code block markers and formatting
+    3. Removes XML tags that might have leaked into the response
+    4. Ensures the message starts with a conventional commit prefix (if enforce_conventional_commits is True)
+    5. Fixes double type prefix issues (e.g., "chore: feat(scope):")
+
+    Args:
+        message: Raw commit message from AI
+        enforce_conventional_commits: If True, ensures message has conventional commit prefix.
+                                      Set to False when using custom system prompts.
+
+    Returns:
+        Cleaned commit message ready for use
+    """
+    message = message.strip()
+    message = _remove_think_tags(message)
+    message = _remove_code_blocks(message)
+    message = _extract_commit_from_reasoning(message)
+    message = _remove_xml_tags(message)
+    message = _fix_double_prefix(message)
+    if enforce_conventional_commits:
+        message = _ensure_conventional_prefix(message)
+    message = _normalize_whitespace(message)
     return message

{gac-1.13.0.dist-info → gac-1.14.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gac
-Version: 1.13.0
+Version: 1.14.0
 Summary: LLM-powered Git commit message generator with multi-provider support
 Project-URL: Homepage, https://github.com/cellwebb/gac
 Project-URL: Documentation, https://github.com/cellwebb/gac#readme
@@ -94,7 +94,7 @@ gac
 - **Anthropic** • **Cerebras** • **Chutes.ai** • **DeepSeek** • **Fireworks**
 - **Gemini** • **Groq** • **LM Studio** • **MiniMax** • **Ollama** • **OpenAI**
 - **OpenRouter** • **Streamlake** • **Synthetic.new** • **Together AI**
-- **Z.AI** • **Z.AI Coding** • **Custom Endpoint (Anthropic/OpenAI)**
+- **Z.AI** • **Z.AI Coding** • **Custom Endpoints (Anthropic/OpenAI)**
 
 ### 🧠 **Smart LLM Analysis**
 
@@ -197,11 +197,14 @@ ANTHROPIC_API_KEY=your_key_here
 
 See `.gac.env.example` for all available options.
 
+**Want to customize commit message style?** See [docs/CUSTOM_SYSTEM_PROMPTS.md](docs/CUSTOM_SYSTEM_PROMPTS.md) for guidance on writing custom system prompts.
+
 ---
 
 ## Getting Help
 
 - **Full documentation**: [USAGE.md](USAGE.md) - Complete CLI reference
+- **Custom prompts**: [CUSTOM_SYSTEM_PROMPTS.md](docs/CUSTOM_SYSTEM_PROMPTS.md) - Customize commit message style
 - **Troubleshooting**: [TROUBLESHOOTING.md](docs/TROUBLESHOOTING.md) - Common issues and solutions
 - **Contributing**: [CONTRIBUTING.md](docs/CONTRIBUTING.md) - Development setup and guidelines
 

{gac-1.13.0.dist-info → gac-1.14.0.dist-info}/RECORD

@@ -1,18 +1,18 @@
 gac/__init__.py,sha256=z9yGInqtycFIT3g1ca24r-A3699hKVaRqGUI79wsmMc,415
-gac/__version__.py,sha256=oFvFxf8Y4B3yJ6IccvXPcWK4lxUIK8Dr3Vc_FOg4a8E,67
+gac/__version__.py,sha256=t4OpwkBuRKYfC3t2AQWw9JbjqP0r1BQFAjOGMvC2Swc,67
 gac/ai.py,sha256=fg642la4yMecOwfZHQ7Ixl6z-5_qj9Q1SxwVMnPDCcY,4244
 gac/ai_utils.py,sha256=EDkw0nnwnV5Ba2CLEo2HC15-L5BZtGJATin5Az0ZHkg,7426
 gac/cli.py,sha256=crUUI6osYtE3QAZ7r6DRlVk9gR3X2PctzS1sssVQ9_g,5070
-gac/config.py,sha256=n3TkQYBqSKkH68QUM6M7kwSK83ghmItoh0p5ZDFnhHA,1746
+gac/config.py,sha256=N0Shy7O0LSEWRtVUbwFJ9HeF93K3Mvc5P6p2HnSB9A4,1813
 gac/config_cli.py,sha256=v9nFHZO1RvK9fzHyuUS6SG-BCLHMsdOMDwWamBhVVh4,1608
-gac/constants.py,sha256=8GHB7yeK2CYT0t80-k9N6LvgZPe-StNH3dK3NsUO46c,4977
+gac/constants.py,sha256=LfaWw6vQ9JPSwZzwZ2HQ-TmTrSxTsBRjvjALRvsVcKE,6189
 gac/diff_cli.py,sha256=wnVQ9OFGnM0d2Pj9WVjWbo0jxqIuRHVAwmb8wU9Pa3E,5676
 gac/errors.py,sha256=ysDIVRCd0YQVTOW3Q6YzdolxCdtkoQCAFf3_jrqbjUY,7916
 gac/git.py,sha256=g6tvph50zV-wrTWrxARYXEpl0NeI8-ffFwHoqhp3fSE,8033
 gac/init_cli.py,sha256=JsHMZBFt_2aFMATlbL_ugSZGQGJf8VRosFjNIRGNM8U,6573
-gac/main.py,sha256=dJrBSN5rJlbWspLGDx3eUJU4uZFVhvuv7qtgIvF7RH4,14723
-gac/preprocess.py,sha256=aMxsjGxy9YP752NWjgf0KP5Sn6p8keIJAGlMYr8jDgQ,15373
-gac/prompt.py,sha256=d_kBXmhf3bDVLyDj8J7AS7GBAxF2jlc8lXoHX3Dzi5k,24255
+gac/main.py,sha256=0NTFPX8Jg2Nd7DoX8VKwEONre1iwVf1i7rMM-ilZY-Y,15153
+gac/preprocess.py,sha256=hk2p2X4-xVDvuy-T1VMzMa9k5fTUbhlWDyw89DCf81Q,15379
+gac/prompt.py,sha256=k3Fy6RxVze4ZCQKa6tS9aY9jUQOWDQW-UEnSUYFHYZs,27111
 gac/security.py,sha256=15Yp6YR8QC4eECJi1BUCkMteh_veZXUbLL6W8qGcDm4,9920
 gac/utils.py,sha256=nV42-brIHW_fBg7x855GM8nRrqEBbRzTSweg-GTyGE8,3971
 gac/providers/__init__.py,sha256=3WTzh3ngAdvR40eezpMMFD7Zibb-LxexDYUcSm4axQI,1305
@@ -34,8 +34,8 @@ gac/providers/streamlake.py,sha256=KAA2ZnpuEI5imzvdWVWUhEBHSP0BMnprKXte6CbwBWY,2
 gac/providers/synthetic.py,sha256=sRMIJTS9LpcXd9A7qp_ZjZxdqtTKRn9fl1W4YwJZP4c,1855
 gac/providers/together.py,sha256=1bUIVHfYzcEDw4hQPE8qV6hjc2JNHPv_khVgpk2IJxI,1667
 gac/providers/zai.py,sha256=kywhhrCfPBu0rElZyb-iENxQxxpVGykvePuL4xrXlaU,2739
-gac-1.13.0.dist-info/METADATA,sha256=pu638LGgkEXsDWzoHnchIkOOne5MjzQ25ILkDtXVOfs,7878
-gac-1.13.0.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-gac-1.13.0.dist-info/entry_points.txt,sha256=tdjN-XMmcWfL92swuRAjT62bFLOAwk9bTMRLGP5Z4aI,36
-gac-1.13.0.dist-info/licenses/LICENSE,sha256=vOab37NouL1PNs5BswnPayrMCqaN2sqLfMQfqPDrpZg,1103
-gac-1.13.0.dist-info/RECORD,,
+gac-1.14.0.dist-info/METADATA,sha256=AJVGolXlFNslvzHQO-7PoVdJeBO2VvxT0PQWfUdOR1w,8151
+gac-1.14.0.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+gac-1.14.0.dist-info/entry_points.txt,sha256=tdjN-XMmcWfL92swuRAjT62bFLOAwk9bTMRLGP5Z4aI,36
+gac-1.14.0.dist-info/licenses/LICENSE,sha256=vOab37NouL1PNs5BswnPayrMCqaN2sqLfMQfqPDrpZg,1103
+gac-1.14.0.dist-info/RECORD,,