ara-cli 0.1.9.77__py3-none-any.whl → 0.1.10.8__py3-none-any.whl

ara_cli/prompt_handler.py

@@ -1,10 +1,4 @@
 import base64
-import litellm
-from ara_cli.classifier import Classifier
-from ara_cli.artefact_creator import ArtefactCreator
-from ara_cli.template_manager import TemplatePathManager
-from ara_cli.ara_config import ConfigManager, LLMConfigItem
-from ara_cli.file_lister import generate_markdown_listing
 from os.path import exists, join
 import os
 from os import makedirs
@@ -13,109 +7,350 @@ import re
 import shutil
 import glob
 import logging
+import warnings
+from io import StringIO
+from contextlib import redirect_stderr
+from langfuse import Langfuse
+from langfuse.api.resources.commons.errors import Error as LangfuseError, NotFoundError
+import litellm
+from ara_cli.classifier import Classifier
+from ara_cli.artefact_creator import ArtefactCreator
+from ara_cli.template_manager import TemplatePathManager
+from ara_cli.ara_config import ConfigManager
+from ara_cli.file_lister import generate_markdown_listing
 
 
 class LLMSingleton:
     _instance = None
-    _model = None
+    _default_model = None
+    _extraction_model = None
+    langfuse = None
 
-    def __init__(self, model_id):
+    def __init__(self, default_model_id, extraction_model_id):
         config = ConfigManager().get_config()
-        selected_config = config.llm_config.get(str(model_id))
+        default_config_data = config.llm_config.get(str(default_model_id))
 
-        if not selected_config:
-            raise ValueError(f"No configuration found for the model: {model_id}")
+        if not default_config_data:
+            raise ValueError(
+                f"No configuration found for the default model: {default_model_id}"
+            )
+        self.default_config_params = default_config_data.model_dump(exclude_none=True)
 
-        LLMSingleton._model = model_id
-        
-        # Typesafe for None values inside the config.
-        self.config_parameters = selected_config.model_dump(exclude_none=True)
+        extraction_config_data = config.llm_config.get(str(extraction_model_id))
+        if not extraction_config_data:
+            raise ValueError(
+                f"No configuration found for the extraction model: {extraction_model_id}"
+            )
+        self.extraction_config_params = extraction_config_data.model_dump(
+            exclude_none=True
+        )
+
+        langfuse_public_key = os.getenv("ARA_CLI_LANGFUSE_PUBLIC_KEY")
+        langfuse_secret_key = os.getenv("ARA_CLI_LANGFUSE_SECRET_KEY")
+        langfuse_host = os.getenv("LANGFUSE_HOST")
+
+        captured_stderr = StringIO()
+        with redirect_stderr(captured_stderr):
+            self.langfuse = Langfuse(
+                public_key=langfuse_public_key,
+                secret_key=langfuse_secret_key,
+                host=langfuse_host,
+            )
 
+        # Check if there was an authentication error
+        stderr_output = captured_stderr.getvalue()
+        if "Authentication error" in stderr_output:
+            warnings.warn(
+                "Invalid Langfuse credentials - prompt tracing disabled and using default prompts. "
+                "Set environment variables 'ARA_CLI_LANGFUSE_PUBLIC_KEY', 'ARA_CLI_LANGFUSE_SECRET_KEY', "
+                "'LANGFUSE_HOST' and restart application to use Langfuse capabilities",
+                UserWarning,
+            )
+
+        LLMSingleton._default_model = default_model_id
+        LLMSingleton._extraction_model = extraction_model_id
         LLMSingleton._instance = self
 
     @classmethod
     def get_instance(cls):
         if cls._instance is None:
             config = ConfigManager().get_config()
-            llm_config = config.llm_config
-            model_to_use = next(iter(llm_config))
-            default_llm = getattr(config, "default_llm", None)
-            if default_llm and default_llm in llm_config:
-                model_to_use = default_llm
-            cls(model_to_use)
+            default_model = config.default_llm
+            if not default_model:
+                if not config.llm_config:
+                    raise ValueError(
+                        "No LLM configurations are defined in the configuration file."
+                    )
+                default_model = next(iter(config.llm_config))
+
+            extraction_model = getattr(config, "extraction_llm", default_model)
+            if not extraction_model:
+                extraction_model = default_model
+
+            cls(default_model, extraction_model)
+        return cls._instance
+
+    @classmethod
+    def get_config_by_purpose(cls, purpose="default"):
+        """
+        purpose= 'default' or 'extraction'
+        """
+        instance = cls.get_instance()
+        if purpose == "extraction":
+            return instance.extraction_config_params.copy()
+        return instance.default_config_params.copy()
+
+    @classmethod
+    def set_default_model(cls, model_name):
+        """Sets the default language model for the current session."""
+        cls.get_instance()
+        if model_name == cls._default_model:
+            return cls._instance
+        cls(model_name, cls._extraction_model)
         return cls._instance
 
     @classmethod
-    def set_model(cls, model_name):
-        if model_name == cls._model:
+    def set_extraction_model(cls, model_name):
+        """Sets the extraction language model for the current session."""
+        cls.get_instance()
+        if model_name == cls._extraction_model:
             return cls._instance
-        cls(model_name)
-        print(f"Language model switched to '{model_name}'")
+        cls(cls._default_model, model_name)
         return cls._instance
 
     @classmethod
-    def get_model(cls):
+    def get_default_model(cls):
+        """Gets the default model name stored in the singleton instance."""
+        if cls._instance is None:
+            cls.get_instance()
+        return cls._default_model
+
+    @classmethod
+    def get_extraction_model(cls):
+        """Gets the extraction model name stored in the singleton instance."""
         if cls._instance is None:
             cls.get_instance()
-        return cls._model
+        return cls._extraction_model
 
 
 def write_string_to_file(filename, string, mode):
-    with open(filename, mode, encoding='utf-8') as file:
+    with open(filename, mode, encoding="utf-8") as file:
         file.write(f"\n{string}\n")
     return file
 
 
 def read_string_from_file(path):
-    with open(path, 'r', encoding='utf-8') as file:
+    with open(path, "r", encoding="utf-8") as file:
         text = file.read()
     return text
 
 
-def send_prompt(prompt):
-    chat = LLMSingleton.get_instance()
+def _is_valid_message(message: dict) -> bool:
+    """
+    Checks if a message in a prompt is valid (i.e., not empty).
+    It handles both string content and list content (for multimodal inputs).
+    """
+    content = message.get("content")
+
+    if isinstance(content, str):
+        return content.strip() != ""
+
+    if isinstance(content, list):
+        # For multimodal content, check if there's at least one non-empty text part.
+        return any(
+            item.get("type") == "text" and item.get("text", "").strip() != ""
+            for item in content
+        )
+
+    return False
+
 
-    config_parameters = chat.config_parameters.copy()
-    if "provider" in config_parameters:
-        del config_parameters["provider"]
+def _norm(p: str) -> str:
+    """Normalize slashes and collapse .. segments."""
+    return os.path.normpath(p) if p else p
 
-    filtered_prompt = list(filter(
-        lambda msg: (
-            msg.get('content') and
-            isinstance(msg['content'], list) and
-            any(
-                item.get('type') == 'text' and
-                item.get('text', '').strip() != ''
-                for item in msg['content']
+
+def resolve_existing_path(rel_or_abs_path: str, anchor_dir: str) -> str:
+    """
+    Resolve a potentially relative path to an existing absolute path.
+
+    Strategy:
+    - If already absolute and exists -> return it.
+    - Else, try from the anchor_dir.
+    - Else, walk up parent directories from anchor_dir and try joining at each level.
+    - If nothing is found, return the normalized original (will fail later with clear message).
+    """
+    if not rel_or_abs_path:
+        return rel_or_abs_path
+
+    candidate = _norm(rel_or_abs_path)
+
+    if os.path.isabs(candidate) and os.path.exists(candidate):
+        return candidate
+
+    anchor_dir = os.path.abspath(anchor_dir or os.getcwd())
+
+    # Try from anchor dir directly
+    direct = _norm(os.path.join(anchor_dir, candidate))
+    if os.path.exists(direct):
+        return direct
+
+    # Walk parents
+    cur = anchor_dir
+    prev = None
+    while cur and cur != prev:
+        test = _norm(os.path.join(cur, candidate))
+        if os.path.exists(test):
+            return test
+        prev = cur
+        cur = os.path.dirname(cur)
+
+    # Give back normalized candidate; open() will raise, but at least path is clean
+    return candidate
+
+
+def send_prompt(prompt, purpose="default"):
+    """Prepares and sends a prompt to the LLM, streaming the response."""
+    chat_instance = LLMSingleton.get_instance()
+    config_parameters = chat_instance.get_config_by_purpose(purpose)
+    model_info = config_parameters.get("model", "unknown_model")
+
+    with LLMSingleton.get_instance().langfuse.start_as_current_span(
+        name="send_prompt"
+    ) as span:
+        span.update_trace(
+            input={"prompt": prompt, "purpose": purpose, "model": model_info}
+        )
+
+        config_parameters.pop("provider", None)
+
+        filtered_prompt = [msg for msg in prompt if _is_valid_message(msg)]
+
+        completion = litellm.completion(
+            **config_parameters, messages=filtered_prompt, stream=True
+        )
+        response_text = ""
+        try:
+            for chunk in completion:
+                chunk_content = chunk.choices[0].delta.content
+                if chunk_content:
+                    response_text += chunk_content
+                yield chunk
+
+            # Update Langfuse span with success output
+            span.update(
+                output={
+                    "success": True,
+                    "response_length": len(response_text),
+                    "response": response_text,
+                }
             )
-        ) or (
-            isinstance(msg.get('content'), str) and
-            msg['content'].strip() != ''
-        ),
-        prompt
-    ))
-
-    completion = litellm.completion(
-        **config_parameters,
-        messages=filtered_prompt,
-        stream=True
-    )
-    for chunk in completion:
-        yield chunk
+
+        except Exception as e:
+            # Update Langfuse span with error details
+            span.update(output={"error": str(e)}, level="ERROR")
+            raise
+
+
+def describe_image(image_path: str) -> str:
+    """
+    Send an image to the LLM and get a text description.
+
+    Args:
+        image_path: Path to the image file
+
+    Returns:
+        Text description of the image
+    """
+    with LLMSingleton.get_instance().langfuse.start_as_current_span(
+        name="ara-cli/describe-image"
+    ) as span:
+        span.update_trace(input={"image_path": image_path})
+
+        try:
+            langfuse_prompt = LLMSingleton.get_instance().langfuse.get_prompt(
+                "ara-cli/describe-image"
+            )
+            describe_image_prompt = (
+                langfuse_prompt.prompt if langfuse_prompt.prompt else None
+            )
+        except (LangfuseError, NotFoundError, Exception) as e:
+            logging.info(f"Could not fetch Langfuse prompt: {e}")
+            describe_image_prompt = None
+
+        # Fallback to default prompt if Langfuse prompt is not available
+        if not describe_image_prompt:
+            logging.info("Using default describe-image prompt.")
+            describe_image_prompt = (
+                "Please describe this image in detail. If it contains text, transcribe it exactly. "
+                "If it's a diagram or chart, explain its structure and content. If it's a photo or illustration, "
+                "describe what you see."
+            )
+
+        # Resolve and read the image
+        resolved_image_path = resolve_existing_path(image_path, os.getcwd())
+        with open(resolved_image_path, "rb") as image_file:
+            base64_image = base64.b64encode(image_file.read()).decode("utf-8")
+
+        # Determine image type
+        image_extension = os.path.splitext(resolved_image_path)[1].lower()
+        mime_type = {
+            ".png": "image/png",
+            ".jpg": "image/jpeg",
+            ".jpeg": "image/jpeg",
+            ".gif": "image/gif",
+            ".bmp": "image/bmp",
+        }.get(image_extension, "image/png")
+
+        # Create message with image
+        message = {
+            "role": "user",
+            "content": [
+                {
+                    "type": "text",
+                    "text": describe_image_prompt,
+                },
+                {
+                    "type": "image_url",
+                    "image_url": {"url": f"data:{mime_type};base64,{base64_image}"},
+                },
+            ],
+        }
+
+        # Get response from LLM using the extraction model purpose
+        response_text = ""
+        for chunk in send_prompt([message], purpose="extraction"):
+            chunk_content = chunk.choices[0].delta.content
+            if chunk_content:
+                response_text += chunk_content
+
+        response_text = response_text.strip()
+
+        span.update(
+            output={
+                "success": True,
+                "description_length": len(response_text),
+                "response": response_text,
+            }
+        )
+
+        return response_text
 
 
 def append_headings(classifier, param, heading_name):
     sub_directory = Classifier.get_sub_directory(classifier)
 
-    artefact_data_path = f"ara/{sub_directory}/{param}.data/{classifier}.prompt_log.md"
+    artefact_data_path = _norm(
+        f"ara/{sub_directory}/{param}.data/{classifier}.prompt_log.md"
+    )
 
     # Check if the file exists, and if not, create an empty file
     if not os.path.exists(artefact_data_path):
-        with open(artefact_data_path, 'w', encoding='utf-8') as file:
+        with open(artefact_data_path, "w", encoding="utf-8") as file:
             file.write("")
 
     content = read_string_from_file(artefact_data_path)
-    pattern = r'## {}_(\d+)'.format(heading_name)
+    pattern = r"## {}_(\d+)".format(heading_name)
     matches = findall(pattern, content)
 
     max_number = 1
@@ -123,27 +358,27 @@ def append_headings(classifier, param, heading_name):
         max_number = max(map(int, matches)) + 1
     heading = f"## {heading_name}_{max_number}"
 
-    write_string_to_file(artefact_data_path, heading, 'a')
+    write_string_to_file(artefact_data_path, heading, "a")
 
 
 def write_prompt_result(classifier, param, text):
     sub_directory = Classifier.get_sub_directory(classifier)
-
-    # TODO change absolute path to relative path with directory navigator
-    artefact_data_path = f"ara/{sub_directory}/{param}.data/{classifier}.prompt_log.md"
-    write_string_to_file(artefact_data_path, text, 'a')
+    artefact_data_path = _norm(
+        f"ara/{sub_directory}/{param}.data/{classifier}.prompt_log.md"
+    )
+    write_string_to_file(artefact_data_path, text, "a")
 
 
 def prompt_data_directory_creation(classifier, parameter):
     sub_directory = Classifier.get_sub_directory(classifier)
-    prompt_data_path = f"ara/{sub_directory}/{parameter}.data/prompt.data"
+    prompt_data_path = _norm(f"ara/{sub_directory}/{parameter}.data/prompt.data")
     if not exists(prompt_data_path):
         makedirs(prompt_data_path)
     return prompt_data_path
 
 
 def get_file_content(path):
-    with open(path, 'r', encoding='utf-8') as file:
+    with open(path, "r", encoding="utf-8") as file:
         return file.read()
 
 
@@ -151,14 +386,25 @@ def initialize_prompt_templates(classifier, parameter):
     prompt_data_path = prompt_data_directory_creation(classifier, parameter)
     prompt_log_path = os.path.dirname(prompt_data_path)
 
-    template_path = os.path.join(os.path.dirname(__file__), 'templates')
+    template_path = os.path.join(os.path.dirname(__file__), "templates")
     artefact_creator = ArtefactCreator()
-    artefact_creator.create_artefact_prompt_files(prompt_log_path, template_path, classifier)
+    artefact_creator.create_artefact_prompt_files(
+        prompt_log_path, template_path, classifier
+    )
 
     generate_config_prompt_template_file(prompt_data_path, "config.prompt_templates.md")
 
     # Mark the relevant artefact in the givens list
-    generate_config_prompt_givens_file(prompt_data_path, "config.prompt_givens.md", artefact_to_mark=f"{parameter}.{classifier}")
+    generate_config_prompt_givens_file(
+        prompt_data_path,
+        "config.prompt_givens.md",
+        artefact_to_mark=f"{parameter}.{classifier}",
+    )
+
+    # Only once (was duplicated before)
+    generate_config_prompt_global_givens_file(
+        prompt_data_path, "config.prompt_global_givens.md"
+    )
 
 
 def write_template_files_to_config(template_type, config_file, base_template_path):
@@ -170,14 +416,14 @@ def write_template_files_to_config(template_type, config_file, base_template_pat
 
 def load_selected_prompt_templates(classifier, parameter):
     sub_directory = Classifier.get_sub_directory(classifier)
-    prompt_data_path = f"ara/{sub_directory}/{parameter}.data/prompt.data"
+    prompt_data_path = _norm(f"ara/{sub_directory}/{parameter}.data/prompt.data")
     config_file_path = os.path.join(prompt_data_path, "config.prompt_templates.md")
 
     if not os.path.exists(config_file_path):
         print("WARNING: config.prompt_templates.md does not exist.")
         return
 
-    with open(config_file_path, 'r', encoding='utf-8') as config_file:
+    with open(config_file_path, "r", encoding="utf-8") as config_file:
         content = config_file.read()
 
     global_base_template_path = TemplatePathManager.get_template_base_path()
@@ -213,7 +459,9 @@ def find_files_with_endings(directory, endings):
     # Create an empty dictionary to store files according to their endings
     files_by_ending = {ending: [] for ending in endings}
 
-    files = [f for f in os.listdir(directory) if os.path.isfile(os.path.join(directory, f))]
+    files = [
+        f for f in os.listdir(directory) if os.path.isfile(os.path.join(directory, f))
+    ]
     # Walk through the files list
     for file in files:
         # Check each file to see if it ends with one of the specified endings
@@ -248,75 +496,114 @@ def move_and_copy_files(source_path, prompt_data_path, prompt_archive_path):
 
                     # Move all existing files with the same ending to the prompt_archive_path
                     for existing_file in glob.glob(glob_pattern):
-                        archived_file_path = os.path.join(prompt_archive_path, os.path.basename(existing_file))
+                        archived_file_path = os.path.join(
+                            prompt_archive_path, os.path.basename(existing_file)
+                        )
                         shutil.move(existing_file, archived_file_path)
-                        print(f"Moved existing prompt-module: {os.path.basename(existing_file)} to prompt.archive")
-                    
+                        print(
+                            f"Moved existing prompt-module: {os.path.basename(existing_file)} to prompt.archive"
+                        )
+
                     # Copy the source_path file to the prompt_data_path directory
                     target_path = os.path.join(prompt_data_path, file_name)
                     shutil.copy(source_path, target_path)
                     print(f"Loaded new prompt-module: {os.path.basename(target_path)}")
 
         else:
-            print(f"File name {file_name} does not end with one of the specified patterns, skipping move and copy.")
+            print(
+                f"File name {file_name} does not end with one of the specified patterns, skipping move and copy."
+            )
     else:
         print(f"WARNING: template {source_path} does not exist.")
 
 
 def extract_and_load_markdown_files(md_prompt_file_path):
     """
-    Extracts markdown files paths based on checked items and constructs proper paths respecting markdown header hierarchy.
+    Extracts markdown files paths based on checked items and constructs proper paths
+    respecting markdown header hierarchy. **Returns normalized relative paths**
+    (not resolved), and resolution happens later relative to the config file dir.
     """
     header_stack = []
     path_accumulator = []
-    with open(md_prompt_file_path, 'r', encoding='utf-8') as file:
+    with open(md_prompt_file_path, "r", encoding="utf-8") as file:
         for line in file:
-            if line.strip().startswith('#'):
-                level = line.count('#')
-                header = line.strip().strip('#').strip()
+            if line.strip().startswith("#"):
+                level = line.count("#")
+                header = line.strip().strip("#").strip()
                 # Adjust the stack based on the current header level
                 current_depth = len(header_stack)
                 if level <= current_depth:
-                    header_stack = header_stack[:level-1]
+                    header_stack = header_stack[: level - 1]
                 header_stack.append(header)
-            elif '[x]' in line:
-                relative_path = line.split(']')[-1].strip()
-                full_path = os.path.join('/'.join(header_stack), relative_path)
-                path_accumulator.append(full_path)
+            elif "[x]" in line:
+                relative_path = line.split("]")[-1].strip()
+                # Use os.path.join for OS-safe joining, then normalize
+                full_rel_path = os.path.join(*header_stack, relative_path) if header_stack else relative_path
+                path_accumulator.append(_norm(full_rel_path))
     return path_accumulator
 
 
 def load_givens(file_path):
-    content = "### GIVENS\n\n"
+    """
+    Reads marked givens from a config markdown and returns:
+      - combined markdown content (including code fences / images)
+      - a list of image data dicts for the multimodal message
+    Paths inside the markdown are resolved robustly relative to the config file directory (and its parents).
+    """
+    content = ""
     image_data_list = []
     markdown_items = extract_and_load_markdown_files(file_path)
 
+    if not markdown_items:
+        return "", []
+
+    content = "### GIVENS\n\n"
+
+    anchor_dir = os.path.dirname(os.path.abspath(file_path))
+
     for item in markdown_items:
-        if item.lower().endswith(('.png', '.jpeg', '.jpg')):
-            with open(item, "rb") as image_file:
+        resolved = resolve_existing_path(item, anchor_dir)
+        # Keep the listing line readable, show the original relative item
+        content += item + "\n"
+
+        ext = os.path.splitext(resolved)[1].lower()
+
+        # Image branch
+        if ext in (".png", ".jpeg", ".jpg", ".gif", ".bmp"):
+            with open(resolved, "rb") as image_file:
                 base64_image = base64.b64encode(image_file.read()).decode("utf-8")
-            image_data_list.append({"type": "image_url", "image_url": {"url": f"data:image/png;base64,{base64_image}"}})
-            content += item + "\n"
-            content += f'![{item}](data:image/png;base64,{base64_image})' + "\n"
-        else:
-            # Check if the item specifies line ranges
-            # TODO item has currently no trailing [] see extraction and handover method in extract and load 
-            # item = f"[10:29] {item}"
-            # print(f"found {item}, check for subsection")
-            # TODO re.match can not split the item with [] correctly and extract the line numbers
-            # TODO logic of subsections is not supported by the update algorithm of the config prompt givens updater
-            # TODO extract in lines of *.md files potential images and add them to the image list
 
+            mime_type = {
+                ".png": "image/png",
+                ".jpg": "image/jpeg",
+                ".jpeg": "image/jpeg",
+                ".gif": "image/gif",
+                ".bmp": "image/bmp",
+            }.get(ext, "image/png")
+
+            image_data_list.append(
+                {
+                    "type": "image_url",
+                    "image_url": {"url": f"data:{mime_type};base64,{base64_image}"},
+                }
+            )
+            # Also embed inline for the prompt markdown (use png as a neutral default for data URI)
+            content += f"![{item}](data:{mime_type};base64,{base64_image})\n"
+
+        else:
+            # Check if the item specifies line ranges: e.g. "[10:20,25:30] filePath"
             match = re.match(r".*?\[(\d+:\d+(?:,\s*\d+:\d+)*)\]\s+(.+)", item)
             if match:
                 line_ranges, file_name = match.groups()
-                content += file_name + "\n" + "```\n"
-                content += get_partial_file_content(file_name, line_ranges) + "\n"
+                resolved_sub = resolve_existing_path(file_name, anchor_dir)
+                content += "```\n"
+                content += get_partial_file_content(resolved_sub, line_ranges) + "\n"
                 content += "```\n\n"
             else:
-                content += item + "\n" + "```\n"
-                content += get_file_content(item) + "\n"
+                content += "```\n"
+                content += get_file_content(resolved) + "\n"
                 content += "```\n\n"
+
     return content, image_data_list
 
 
@@ -325,25 +612,25 @@ def get_partial_file_content(file_name, line_ranges):
     Reads specific lines from a file based on the line ranges provided.
 
     Args:
-    file_name (str): The path to the file.
+    file_name (str): The path to the file (absolute or relative, already resolved by caller).
     line_ranges (str): A string representing the line ranges to read, e.g., '10:20,25:30'.
 
     Returns:
     str: The content of the specified lines.
     """
-    line_ranges = line_ranges.strip('[]').split(',')
+    line_ranges = line_ranges.strip("[]").split(",")
     lines_to_read = []
     for line_range in line_ranges:
-        start, end = map(int, line_range.split(':'))
+        start, end = map(int, line_range.split(":"))
         lines_to_read.extend(range(start, end + 1))
 
     partial_content = []
-    with open(file_name, 'r', encoding='utf-8') as file:
+    with open(file_name, "r", encoding="utf-8") as file:
         for i, line in enumerate(file, 1):
             if i in lines_to_read:
                 partial_content.append(line)
 
-    return ''.join(partial_content)
+    return "".join(partial_content)
 
 
 def collect_file_content_by_extension(prompt_data_path, extensions):
@@ -353,7 +640,7 @@ def collect_file_content_by_extension(prompt_data_path, extensions):
         files = find_files_with_endings(prompt_data_path, [ext])
         for file_name in files:
             file_path = join(prompt_data_path, file_name)
-            if ext == ".prompt_givens.md":
+            if ext in [".prompt_givens.md", ".prompt_global_givens.md"]:
                 givens, image_data = load_givens(file_path)
                 combined_content += givens
                 image_data_list.extend(image_data)
@@ -363,27 +650,53 @@ def collect_file_content_by_extension(prompt_data_path, extensions):
 
 
 def prepend_system_prompt(message_list):
-    system_prompt = {
-        "role": "system",
-        "content": "You are a helpful assistant that can process both text and images."
-    }
-    message_list.insert(0, system_prompt)
+    try:
+        langfuse_prompt = LLMSingleton.get_instance().langfuse.get_prompt(
+            "ara-cli/system-prompt"
+        )
+        system_prompt = langfuse_prompt.prompt if langfuse_prompt.prompt else None
+    except (LangfuseError, NotFoundError, Exception) as e:
+        logging.info(f"Could not fetch Langfuse system prompt: {e}")
+        system_prompt = None
+
+    # Fallback to default prompt if Langfuse prompt is not available
+    if not system_prompt:
+        logging.info("Using default system prompt.")
+        system_prompt = "You are a helpful assistant that can process both text and images."
+
+    # Prepend the system prompt
+    system_prompt_message = {"role": "system", "content": system_prompt}
+
+    message_list.insert(0, system_prompt_message)
     return message_list
 
 
 def append_images_to_message(message, image_data_list):
+    """
+    Appends image data list to a single message dict (NOT to a list).
+    """
     logger = logging.getLogger(__name__)
 
-    logger.debug(f"append_images_to_message called with image_data_list length: {len(image_data_list) if image_data_list else 0}")
+    logger.debug(
+        f"append_images_to_message called with image_data_list length: {len(image_data_list) if image_data_list else 0}"
+    )
 
     if not image_data_list:
         logger.debug("No images to append, returning original message")
         return message
 
-    message_content = message["content"]
+    message_content = message.get("content")
     logger.debug(f"Original message content: {message_content}")
 
-    message["content"] = message_content + image_data_list
+    if isinstance(message_content, str):
+        message["content"] = [{"type": "text", "text": message_content}]
+
+    if isinstance(message["content"], list):
+        message["content"].extend(image_data_list)
+    else:
+        # If somehow content is not list or str, coerce to list
+        message["content"] = [{"type": "text", "text": str(message_content)}] + image_data_list
+
     logger.debug(f"Updated message content with {len(image_data_list)} images")
 
     return message
@@ -391,74 +704,123 @@ def append_images_to_message(message, image_data_list):
 
 def create_and_send_custom_prompt(classifier, parameter):
     sub_directory = Classifier.get_sub_directory(classifier)
-    prompt_data_path = f"ara/{sub_directory}/{parameter}.data/prompt.data"
+    prompt_data_path = _norm(f"ara/{sub_directory}/{parameter}.data/prompt.data")
     prompt_file_path_markdown = join(prompt_data_path, f"{classifier}.prompt.md")
 
-    extensions = [".blueprint.md", ".rules.md", ".prompt_givens.md", ".intention.md", ".commands.md"]
-    combined_content_markdown, image_data_list = collect_file_content_by_extension(prompt_data_path, extensions)
+    extensions = [
+        ".blueprint.md",
+        ".rules.md",
+        ".prompt_givens.md",
+        ".prompt_global_givens.md",
+        ".intention.md",
+        ".commands.md",
+    ]
+    combined_content_markdown, image_data_list = collect_file_content_by_extension(
+        prompt_data_path, extensions
+    )
 
-    with open(prompt_file_path_markdown, 'w', encoding='utf-8') as file:
+    with open(prompt_file_path_markdown, "w", encoding="utf-8") as file:
         file.write(combined_content_markdown)
 
     prompt = read_string_from_file(prompt_file_path_markdown)
     append_headings(classifier, parameter, "prompt")
     write_prompt_result(classifier, parameter, prompt)
 
-    message = {
-        "role": "user",
-        "content": combined_content_markdown
-    }
-
+    # Build message and append images correctly (fixed)
+    message = {"role": "user", "content": combined_content_markdown}
+    message = append_images_to_message(message, image_data_list)
     message_list = [message]
 
-    message_list = append_images_to_message(message_list, image_data_list)
     append_headings(classifier, parameter, "result")
 
-    artefact_data_path = f"ara/{sub_directory}/{parameter}.data/{classifier}.prompt_log.md"
-    with open(artefact_data_path, 'a', encoding='utf-8') as file:
+    artefact_data_path = _norm(
+        f"ara/{sub_directory}/{parameter}.data/{classifier}.prompt_log.md"
+    )
+    with open(artefact_data_path, "a", encoding="utf-8") as file:
         for chunk in send_prompt(message_list):
             chunk_content = chunk.choices[0].delta.content
             if not chunk_content:
                 continue
             file.write(chunk_content)
             file.flush()
-    # write_prompt_result(classifier, parameter, response)
 
 
-def generate_config_prompt_template_file(prompt_data_path, config_prompt_templates_name):
-    config_prompt_templates_path = os.path.join(prompt_data_path, config_prompt_templates_name)
-    config = ConfigManager.get_config()
+def generate_config_prompt_template_file(
+    prompt_data_path, config_prompt_templates_name
+):
+    config_prompt_templates_path = os.path.join(
+        prompt_data_path, config_prompt_templates_name
+    )
+    # Use instance method consistently
+    config = ConfigManager().get_config()
     global_prompt_template_path = TemplatePathManager.get_template_base_path()
-    dir_list = ["ara/.araconfig/custom-prompt-modules"] + [f"{os.path.join(global_prompt_template_path,'prompt-modules')}"]
-    file_list = ['*.blueprint.md','*.rules.md','*.intention.md', '*.commands.md']
+    dir_list = ["ara/.araconfig/custom-prompt-modules"] + [
+        f"{os.path.join(global_prompt_template_path,'prompt-modules')}"
+    ]
+    file_list = ["*.blueprint.md", "*.rules.md", "*.intention.md", "*.commands.md"]
 
     print(f"used {dir_list} for prompt templates file listing")
     generate_markdown_listing(dir_list, file_list, config_prompt_templates_path)
 
 
-def generate_config_prompt_givens_file(prompt_data_path, config_prompt_givens_name, artefact_to_mark=None):
-    config_prompt_givens_path = os.path.join(prompt_data_path, config_prompt_givens_name)
-    config = ConfigManager.get_config()
-    dir_list = ["ara"] + [ext.source_dir for ext in config.ext_code_dirs] + [config.doc_dir] + [config.glossary_dir]
+def generate_config_prompt_givens_file(
+    prompt_data_path, config_prompt_givens_name, artefact_to_mark=None
+):
+    config_prompt_givens_path = os.path.join(
+        prompt_data_path, config_prompt_givens_name
+    )
+    config = ConfigManager().get_config()
+    dir_list = (
+        ["ara"]
+        + [path for d in config.ext_code_dirs for path in d.values()]
+        + [config.doc_dir]
+        + [config.glossary_dir]
+    )
 
     print(f"used {dir_list} for prompt givens file listing")
-    generate_markdown_listing(dir_list, config.ara_prompt_given_list_includes, config_prompt_givens_path)
+    generate_markdown_listing(
+        dir_list, config.ara_prompt_given_list_includes, config_prompt_givens_path
+    )
 
     # If an artefact is specified, mark it with [x]
     if artefact_to_mark:
-        print(f"artefact {artefact_to_mark} marked in related config.prompt_givens.md per default")
+        print(
+            f"artefact {artefact_to_mark} marked in related config.prompt_givens.md per default"
+        )
 
         # Read the generated file content
-        with open(config_prompt_givens_path, 'r', encoding='utf-8') as file:
+        with open(config_prompt_givens_path, "r", encoding="utf-8") as file:
             markdown_listing = file.readlines()
 
         updated_listing = []
         for line in markdown_listing:
             # Use a regular expression to match the exact string
-            if re.search(r'\b' + re.escape(artefact_to_mark) + r'\b', line):
+            if re.search(r"\b" + re.escape(artefact_to_mark) + r"\b", line):
                 line = line.replace("[]", "[x]")
             updated_listing.append(line)
 
         # Write the updated listing back to the file
-        with open(config_prompt_givens_path, 'w', encoding='utf-8') as file:
-            file.write("".join(updated_listing))
+        with open(config_prompt_givens_path, "w", encoding="utf-8") as file:
+            file.write("".join(updated_listing))
+
+
+def generate_config_prompt_global_givens_file(
+    prompt_data_path, config_prompt_givens_name, artefact_to_mark=None
+):
+    from ara_cli.global_file_lister import generate_global_markdown_listing
+
+    config_prompt_givens_path = os.path.join(
+        prompt_data_path, config_prompt_givens_name
+    )
+    config = ConfigManager().get_config()
+
+    if not hasattr(config, "global_dirs") or not config.global_dirs:
+        return
+
+    dir_list = [path for d in config.global_dirs for path in d.values()]
+    print(
+        f"used {dir_list} for global prompt givens file listing with absolute paths"
+    )
+    generate_global_markdown_listing(
+        dir_list, config.ara_prompt_given_list_includes, config_prompt_givens_path
+    )