solace-agent-mesh 0.2.4__py3-none-any.whl → 1.0.1__py3-none-any.whl

solace_agent_mesh/common/utils/embeds/modifiers.py

@@ -0,0 +1,816 @@
+"""
+Defines modifier implementation functions and their contracts.
+"""
+
+import re
+from typing import Any, Callable, Dict, Optional, Tuple, List
+
+from solace_ai_connector.common.log import log
+
+from .constants import EARLY_EMBED_TYPES, LATE_EMBED_TYPES
+
+try:
+    from jsonpath_ng.ext import parse as jsonpath_parse
+
+    JSONPATH_NG_AVAILABLE = True
+except ImportError:
+    JSONPATH_NG_AVAILABLE = False
+
+try:
+    import pystache
+
+    PYSTACHE_AVAILABLE = True
+except ImportError:
+    PYSTACHE_AVAILABLE = False
+
+from google.adk.artifacts import BaseArtifactService
+
+from .types import DataFormat
+
+
+def _apply_jsonpath(
+    current_data: Any, expression: str, mime_type: Optional[str], log_id: str
+) -> Tuple[Any, Optional[str], Optional[str]]:
+    """
+    Applies a JSONPath expression to parsed JSON data.
+
+    Args:
+        current_data: The input data (expected to be dict or list).
+        expression: The JSONPath expression string.
+        mime_type: The original mime type (passed through).
+        log_id: Identifier for logging.
+
+    Returns:
+        Tuple: (result_data, original_mime_type, error_string)
+               result_data is typically a list of matched values.
+    """
+    if not JSONPATH_NG_AVAILABLE:
+        return (
+            current_data,
+            mime_type,
+            "JSONPath modifier skipped: 'jsonpath-ng' not installed.",
+        )
+
+    if not isinstance(current_data, (dict, list)):
+        return (
+            current_data,
+            mime_type,
+            f"Input data for 'jsonpath' must be a JSON object or list, got {type(current_data).__name__}.",
+        )
+
+    try:
+        jsonpath_expr = jsonpath_parse(expression)
+        matches = [match.value for match in jsonpath_expr.find(current_data)]
+        return matches, mime_type, None
+    except Exception as e:
+        return (
+            current_data,
+            mime_type,
+            f"Error applying JSONPath expression '{expression}': {e}",
+        )
+
+
+def _apply_select_cols(
+    current_data: List[Dict], cols_str: str, mime_type: Optional[str], log_id: str
+) -> Tuple[Any, Optional[str], Optional[str]]:
+    """
+    Selects specific columns from data represented as a list of dictionaries.
+
+    Args:
+        current_data: The input data (expected List[Dict]).
+        cols_str: Comma-separated string of column names to keep.
+        mime_type: The original mime type (passed through).
+        log_id: Identifier for logging.
+
+    Returns:
+        Tuple: (result_data, original_mime_type, error_string)
+               result_data is List[Dict] containing only selected columns.
+    """
+    if not isinstance(current_data, list) or (
+        current_data and not isinstance(current_data[0], dict)
+    ):
+        return (
+            current_data,
+            mime_type,
+            f"Input data for 'select_cols' must be a list of dictionaries, got {type(current_data).__name__}.",
+        )
+
+    if not current_data:
+        return [], mime_type, None
+
+    try:
+        header = list(current_data[0].keys())
+        target_cols = [col.strip() for col in cols_str.split(",")]
+        output_list = []
+
+        for target_col in target_cols:
+            if target_col not in header:
+                return (
+                    current_data,
+                    mime_type,
+                    f"Column '{target_col}' not found in data keys: {header}",
+                )
+
+        for row_dict in current_data:
+            new_row = {col: row_dict.get(col) for col in target_cols}
+            output_list.append(new_row)
+
+        return output_list, mime_type, None
+
+    except Exception as e:
+        return current_data, mime_type, f"Error selecting columns '{cols_str}': {e}"
+
+
+def _apply_filter_rows_eq(
+    current_data: List[Dict], filter_spec: str, mime_type: Optional[str], log_id: str
+) -> Tuple[Any, Optional[str], Optional[str]]:
+    """
+    Filters a list of dictionaries based on a column's value equality.
+
+    Args:
+        current_data: The input data (expected List[Dict]).
+        filter_spec: String in the format 'column_name:value'.
+        mime_type: The original mime type (passed through).
+        log_id: Identifier for logging.
+
+    Returns:
+        Tuple: (result_data, original_mime_type, error_string)
+               result_data is List[Dict] containing only filtered rows.
+    """
+    if not isinstance(current_data, list) or (
+        current_data and not isinstance(current_data[0], dict)
+    ):
+        return (
+            current_data,
+            mime_type,
+            f"Input data for 'filter_rows_eq' must be a list of dictionaries, got {type(current_data).__name__}.",
+        )
+
+    if not current_data:
+        return [], mime_type, None
+
+    try:
+        parts = filter_spec.split(":", 1)
+        if len(parts) != 2:
+            return (
+                current_data,
+                mime_type,
+                f"Invalid filter format '{filter_spec}'. Expected 'column_name:value'.",
+            )
+        col_name, filter_value = parts[0].strip(), parts[1].strip()
+
+        header = list(current_data[0].keys())
+        if col_name not in header:
+            return (
+                current_data,
+                mime_type,
+                f"Filter column '{col_name}' not found in data keys: {header}",
+            )
+
+        output_list = [
+            row for row in current_data if str(row.get(col_name)) == filter_value
+        ]
+
+        return output_list, mime_type, None
+
+    except Exception as e:
+        return current_data, mime_type, f"Error filtering rows by '{filter_spec}': {e}"
+
+
+def _apply_slice_rows(
+    current_data: List[Dict], slice_spec: str, mime_type: Optional[str], log_id: str
+) -> Tuple[Any, Optional[str], Optional[str]]:
+    """
+    Selects a slice of rows from a list of dictionaries.
+
+    Args:
+        current_data: The input data (expected List[Dict]).
+        slice_spec: String in Python slice format 'start:end'.
+        mime_type: The original mime type (passed through).
+        log_id: Identifier for logging.
+
+    Returns:
+        Tuple: (result_data, original_mime_type, error_string)
+               result_data is List[Dict] containing the sliced rows.
+    """
+    if not isinstance(current_data, list):
+        return (
+            current_data,
+            mime_type,
+            f"Input data for 'slice_rows' must be a list, got {type(current_data).__name__}.",
+        )
+
+    try:
+        start_str, end_str = None, None
+        if ":" in slice_spec:
+            parts = slice_spec.split(":", 1)
+            start_str, end_str = parts[0].strip(), parts[1].strip()
+        else:
+            return (
+                current_data,
+                mime_type,
+                f"Invalid slice format '{slice_spec}'. Expected 'start:end'.",
+            )
+
+        start = int(start_str) if start_str else 0
+        end = int(end_str) if end_str else None
+
+        sliced_data = current_data[start:end]
+
+        return sliced_data, mime_type, None
+
+    except (ValueError, TypeError) as e:
+        return current_data, mime_type, f"Invalid slice indices in '{slice_spec}': {e}"
+    except Exception as e:
+        return current_data, mime_type, f"Error slicing rows '{slice_spec}': {e}"
+
+
+def _apply_slice_lines(
+    current_data: str, slice_spec: str, mime_type: Optional[str], log_id: str
+) -> Tuple[Any, Optional[str], Optional[str]]:
+    """
+    Selects a slice of lines from text data.
+
+    Args:
+        current_data: The input data (expected str).
+        slice_spec: String in Python slice format 'start:end'.
+        mime_type: The original mime type (passed through).
+        log_id: Identifier for logging.
+
+    Returns:
+        Tuple: (result_data, original_mime_type, error_string)
+               result_data is str containing the sliced lines.
+    """
+    if not isinstance(current_data, str):
+        return (
+            current_data,
+            mime_type,
+            f"Input data for 'slice_lines' must be a string, got {type(current_data).__name__}.",
+        )
+
+    try:
+        start_str, end_str = None, None
+        if ":" in slice_spec:
+            parts = slice_spec.split(":", 1)
+            start_str, end_str = parts[0].strip(), parts[1].strip()
+        else:
+            return (
+                current_data,
+                mime_type,
+                f"Invalid slice format '{slice_spec}'. Expected 'start:end'.",
+            )
+
+        start = int(start_str) if start_str else 0
+        end = int(end_str) if end_str else None
+
+        lines = current_data.splitlines(keepends=True)
+        sliced_lines = lines[start:end]
+
+        return "".join(sliced_lines), mime_type, None
+
+    except (ValueError, TypeError) as e:
+        return current_data, mime_type, f"Invalid slice indices in '{slice_spec}': {e}"
+    except Exception as e:
+        return current_data, mime_type, f"Error slicing text lines '{slice_spec}': {e}"
+
+
+def _apply_grep(
+    current_data: str, pattern: str, mime_type: Optional[str], log_id: str
+) -> Tuple[Any, Optional[str], Optional[str]]:
+    """
+    Filters lines matching a regex pattern from text data.
+
+    Args:
+        current_data: The input data (expected str).
+        pattern: The regex pattern string.
+        mime_type: The original mime type (passed through).
+        log_id: Identifier for logging.
+
+    Returns:
+        Tuple: (result_data, original_mime_type, error_string)
+               result_data is str containing only matching lines.
+    """
+    if not isinstance(current_data, str):
+        return (
+            current_data,
+            mime_type,
+            f"Input data for 'grep' must be a string, got {type(current_data).__name__}.",
+        )
+
+    try:
+        regex = re.compile(pattern)
+        lines = current_data.splitlines(keepends=True)
+        filtered_lines = [line for line in lines if regex.search(line)]
+        return "".join(filtered_lines), mime_type, None
+    except re.error as e:
+        return current_data, mime_type, f"Invalid regex pattern '{pattern}': {e}"
+    except Exception as e:
+        return current_data, mime_type, f"Error applying grep pattern '{pattern}': {e}"
+
+
+def _apply_head(
+    current_data: str, n_str: str, mime_type: Optional[str], log_id: str
+) -> Tuple[Any, Optional[str], Optional[str]]:
+    """
+    Returns the first N lines of text data.
+
+    Args:
+        current_data: The input data (expected str).
+        n_str: String representing the number of lines (N).
+        mime_type: The original mime type (passed through).
+        log_id: Identifier for logging.
+
+    Returns:
+        Tuple: (result_data, original_mime_type, error_string)
+               result_data is str containing the first N lines.
+    """
+    if not isinstance(current_data, str):
+        return (
+            current_data,
+            mime_type,
+            f"Input data for 'head' must be a string, got {type(current_data).__name__}.",
+        )
+
+    try:
+        n = int(n_str.strip())
+        if n < 0:
+            return current_data, mime_type, "Head count N cannot be negative."
+
+        lines = current_data.splitlines(keepends=True)
+        head_lines = lines[:n]
+        return "".join(head_lines), mime_type, None
+    except (ValueError, TypeError) as e:
+        return current_data, mime_type, f"Invalid head count N '{n_str}': {e}"
+    except Exception as e:
+        return current_data, mime_type, f"Error applying head '{n_str}': {e}"
+
+
+def _apply_tail(
+    current_data: str, n_str: str, mime_type: Optional[str], log_id: str
+) -> Tuple[Any, Optional[str], Optional[str]]:
+    """
+    Returns the last N lines of text data.
+
+    Args:
+        current_data: The input data (expected str).
+        n_str: String representing the number of lines (N).
+        mime_type: The original mime type (passed through).
+        log_id: Identifier for logging.
+
+    Returns:
+        Tuple: (result_data, original_mime_type, error_string)
+               result_data is str containing the last N lines.
+    """
+    if not isinstance(current_data, str):
+        return (
+            current_data,
+            mime_type,
+            f"Input data for 'tail' must be a string, got {type(current_data).__name__}.",
+        )
+
+    try:
+        n = int(n_str.strip())
+        if n < 0:
+            return current_data, mime_type, "Tail count N cannot be negative."
+        if n == 0:
+            return "", mime_type, None
+
+        lines = current_data.splitlines(keepends=True)
+        tail_lines = lines[-n:]
+        return "".join(tail_lines), mime_type, None
+    except (ValueError, TypeError) as e:
+        return current_data, mime_type, f"Invalid tail count N '{n_str}': {e}"
+    except Exception as e:
+        return current_data, mime_type, f"Error applying tail '{n_str}': {e}"
+
+
+def _apply_select_fields(
+    current_data: List[Dict], fields_str: str, mime_type: Optional[str], log_id: str
+) -> Tuple[Any, Optional[str], Optional[str]]:
+    """
+    Selects specific fields from a list of dictionaries.
+
+    Args:
+        current_data: The input data (expected List[Dict]).
+        fields_str: Comma-separated string of field names to keep.
+        mime_type: The original mime type (passed through).
+        log_id: Identifier for logging.
+
+    Returns:
+        Tuple: (result_data, original_mime_type, error_string)
+               result_data is List[Dict] containing only selected fields.
+    """
+    if not isinstance(current_data, list) or (
+        current_data and not isinstance(current_data[0], dict)
+    ):
+        return (
+            current_data,
+            mime_type,
+            f"Input data for 'select_fields' must be a list of dictionaries, got {type(current_data).__name__}.",
+        )
+
+    target_fields = [field.strip() for field in fields_str.split(",")]
+    if not target_fields:
+        return current_data, mime_type, "No fields specified for 'select_fields'."
+
+    output_list = []
+    try:
+        for item in current_data:
+            if isinstance(item, dict):
+                new_item = {
+                    field: item.get(field) for field in target_fields if field in item
+                }
+                output_list.append(new_item)
+            else:
+                log.warning(
+                    "%s Skipping non-dictionary item in list during select_fields.",
+                    log_id,
+                )
+                continue
+        return output_list, mime_type, None
+    except Exception as e:
+        return current_data, mime_type, f"Error selecting fields '{fields_str}': {e}"
+
+
+async def _apply_template(
+    current_data: Any,
+    template_spec: str,
+    mime_type: Optional[str],
+    log_id: str,
+    context: Any,
+) -> Tuple[Any, Optional[str], Optional[str]]:
+    """
+    Applies a Mustache template loaded from an artifact to the input data.
+    This version first renders the template, then resolves embeds on the result.
+
+    Args:
+        current_data: The input data (expected dict, list, or str).
+        template_spec: String 'template_filename[:version]'.
+        mime_type: The original mime type (passed through).
+        log_id: Identifier for logging.
+        context: The Gateway context dictionary containing artifact_service and session_context.
+
+    Returns:
+        Tuple: (result_data, original_mime_type, error_string)
+               result_data is the rendered and resolved string.
+    """
+    from .resolver import resolve_embeds_recursively_in_string, evaluate_embed
+
+    if not PYSTACHE_AVAILABLE:
+        return (
+            current_data,
+            mime_type,
+            "Template modifier skipped: 'pystache' not installed.",
+        )
+
+    if not isinstance(current_data, (dict, list, str)):
+        return (
+            current_data,
+            mime_type,
+            f"Input data for 'apply_to_template' must be dict, list, or string, got {type(current_data).__name__}.",
+        )
+
+    parts = template_spec.strip().split(":", 1)
+    template_filename = parts[0]
+    template_version_str = parts[1] if len(parts) > 1 else None
+    template_version = None
+
+    if not template_filename:
+        return current_data, mime_type, "Template filename cannot be empty."
+
+    if not isinstance(context, dict):
+        return current_data, mime_type, "Invalid context for template loading."
+    artifact_service: Optional[BaseArtifactService] = context.get("artifact_service")
+    session_context = context.get("session_context")
+    if not artifact_service or not session_context:
+        return (
+            current_data,
+            mime_type,
+            "ArtifactService or session context not available for template loading.",
+        )
+
+    app_name = session_context.get("app_name")
+    user_id = session_context.get("user_id")
+    session_id = session_context.get("session_id")
+    if not all([app_name, user_id, session_id]):
+        return (
+            current_data,
+            mime_type,
+            "Missing required session identifiers in context for template loading.",
+        )
+
+    try:
+        if template_version_str:
+            template_version = int(template_version_str)
+        else:
+            versions = await artifact_service.list_versions(
+                app_name=app_name,
+                user_id=user_id,
+                session_id=session_id,
+                filename=template_filename,
+            )
+            if not versions:
+                return (
+                    current_data,
+                    mime_type,
+                    f"Template artifact '{template_filename}' (latest) not found.",
+                )
+            template_version = max(versions)
+
+        template_part = await artifact_service.load_artifact(
+            app_name=app_name,
+            user_id=user_id,
+            session_id=session_id,
+            filename=template_filename,
+            version=template_version,
+        )
+
+        if not template_part or not template_part.inline_data:
+            return (
+                current_data,
+                mime_type,
+                f"Template artifact '{template_filename}' v{template_version} not found or empty.",
+            )
+
+        template_bytes = template_part.inline_data.data
+        try:
+            raw_template_string = template_bytes.decode("utf-8")
+        except UnicodeDecodeError:
+            return (
+                current_data,
+                mime_type,
+                f"Cannot render non-UTF-8 decodable binary template '{template_filename}' v{template_version}.",
+            )
+
+    except FileNotFoundError:
+        return (
+            current_data,
+            mime_type,
+            f"Template artifact '{template_filename}' v{template_version_str or 'latest'} not found.",
+        )
+    except ValueError as e:
+        return (
+            current_data,
+            mime_type,
+            f"Invalid version specified for template: '{template_version_str}' or other value error: {e}",
+        )
+    except Exception as e:
+        return (
+            current_data,
+            mime_type,
+            f"Error loading template artifact '{template_filename}' v{template_version_str or 'latest'}: {e}",
+        )
+
+    try:
+        log.info(
+            "%s [apply_to_template] Preparing render context. Input data type: %s, Original MIME: %s",
+            log_id,
+            type(current_data).__name__,
+            mime_type,
+        )
+        render_context: Dict[str, Any]
+
+        if isinstance(current_data, list):
+            if mime_type and "csv" in mime_type.lower():
+                log.info(
+                    "%s [apply_to_template] Input is a list and original MIME is CSV. Structuring context with 'headers' and 'data_rows'.",
+                    log_id,
+                )
+                if not current_data:
+                    render_context = {"headers": [], "data_rows": []}
+                else:
+                    if all(isinstance(item, dict) for item in current_data):
+                        headers = list(current_data[0].keys()) if current_data else []
+                        data_rows = [list(row.values()) for row in current_data]
+                        render_context = {"headers": headers, "data_rows": data_rows}
+                    else:
+                        log.warning(
+                            "%s [apply_to_template] Input is list from CSV, but items are not all dictionaries. Falling back to 'items' context.",
+                            log_id,
+                        )
+                        render_context = {"items": current_data}
+            else:
+                log.info(
+                    "%s [apply_to_template] Input is a list (from JSON/YAML). Data available under 'items' key.",
+                    log_id,
+                )
+                render_context = {"items": current_data}
+        elif isinstance(current_data, dict):
+            render_context = current_data
+            log.info(
+                "%s [apply_to_template] Input is dict. Keys directly available in template.",
+                log_id,
+            )
+        elif isinstance(current_data, str):
+            render_context = {"text": current_data}
+            log.info(
+                "%s [apply_to_template] Input is string. Data available under 'text' key.",
+                log_id,
+            )
+        else:
+            log.warning(
+                "%s [apply_to_template] Input is unexpected type %s. Converting to string and placing under 'value' key.",
+                log_id,
+                type(current_data).__name__,
+            )
+            render_context = {"value": str(current_data)}
+
+        log.info(
+            "%s [apply_to_template] Render context keys: %s",
+            log_id,
+            list(render_context.keys()),
+        )
+        if "items" in render_context and isinstance(render_context["items"], list):
+            log.info(
+                "%s [apply_to_template] Render context 'items' length: %d",
+                log_id,
+                len(render_context["items"]),
+            )
+
+        intermediate_rendered_string = pystache.render(
+            raw_template_string, render_context
+        )
+        log.debug(
+            "%s [apply_to_template] Intermediate rendered string: %s",
+            log_id,
+            intermediate_rendered_string[:200] + "...",
+        )
+
+    except Exception as e:
+        return (
+            current_data,
+            mime_type,
+            f"Error preparing context or rendering template '{template_filename}' v{template_version}: {e}",
+        )
+
+    try:
+        log.debug(
+            "%s [apply_to_template] Resolving embeds on rendered template output.",
+            log_id,
+        )
+        resolver_config = context.get("config", {})
+        if not resolver_config:
+            log.warning(
+                "%s 'config' not found in context for template embed resolution. Using defaults.",
+                log_id,
+            )
+
+        final_rendered_string = await resolve_embeds_recursively_in_string(
+            text=intermediate_rendered_string,
+            context=context,
+            resolver_func=evaluate_embed,
+            types_to_resolve=EARLY_EMBED_TYPES.union(LATE_EMBED_TYPES),
+            log_identifier=f"{log_id}[TemplateEmbeds]",
+            config=resolver_config,
+            max_depth=resolver_config.get("gateway_recursive_embed_depth", 12),
+            current_depth=0,
+            visited_artifacts=set(),
+            accumulated_size=0,
+            max_total_size=resolver_config.get(
+                "gateway_max_artifact_resolve_size_bytes", -1
+            ),
+        )
+        log.debug(
+            "%s [apply_to_template] Final rendered string after embed resolution: %s",
+            log_id,
+            final_rendered_string[:200] + "...",
+        )
+    except Exception as recurse_err:
+        log.exception(
+            "%s Error during recursive resolution of rendered template: %s",
+            log_id,
+            recurse_err,
+        )
+        return (
+            current_data,
+            mime_type,
+            f"Error resolving embeds within rendered template: {recurse_err}",
+        )
+
+    return final_rendered_string, mime_type, None
+
+
+MODIFIER_IMPLEMENTATIONS: Dict[
+    str, Callable[..., Tuple[Any, Optional[str], Optional[str]]]
+] = {
+    "jsonpath": _apply_jsonpath,
+    "select_cols": _apply_select_cols,
+    "filter_rows_eq": _apply_filter_rows_eq,
+    "slice_rows": _apply_slice_rows,
+    "slice_lines": _apply_slice_lines,
+    "grep": _apply_grep,
+    "head": _apply_head,
+    "tail": _apply_tail,
+    "select_fields": _apply_select_fields,
+    "apply_to_template": _apply_template,
+}
+
+MODIFIER_DEFINITIONS: Dict[str, Dict[str, Any]] = {
+    "jsonpath": {
+        "function": _apply_jsonpath,
+        "accepts": [DataFormat.JSON_OBJECT],
+        "produces": DataFormat.JSON_OBJECT,
+    },
+    "select_cols": {
+        "function": _apply_select_cols,
+        "accepts": [DataFormat.LIST_OF_DICTS],
+        "produces": DataFormat.LIST_OF_DICTS,
+    },
+    "filter_rows_eq": {
+        "function": _apply_filter_rows_eq,
+        "accepts": [DataFormat.LIST_OF_DICTS],
+        "produces": DataFormat.LIST_OF_DICTS,
+    },
+    "slice_rows": {
+        "function": _apply_slice_rows,
+        "accepts": [DataFormat.LIST_OF_DICTS],
+        "produces": DataFormat.LIST_OF_DICTS,
+    },
+    "slice_lines": {
+        "function": _apply_slice_lines,
+        "accepts": [DataFormat.STRING],
+        "produces": DataFormat.STRING,
+    },
+    "grep": {
+        "function": _apply_grep,
+        "accepts": [DataFormat.STRING],
+        "produces": DataFormat.STRING,
+    },
+    "head": {
+        "function": _apply_head,
+        "accepts": [DataFormat.STRING],
+        "produces": DataFormat.STRING,
+    },
+    "tail": {
+        "function": _apply_tail,
+        "accepts": [DataFormat.STRING],
+        "produces": DataFormat.STRING,
+    },
+    "select_fields": {
+        "function": _apply_select_fields,
+        "accepts": [DataFormat.LIST_OF_DICTS],
+        "produces": DataFormat.LIST_OF_DICTS,
+    },
+    "apply_to_template": {
+        "function": _apply_template,
+        "accepts": [
+            DataFormat.JSON_OBJECT,
+            DataFormat.LIST_OF_DICTS,
+            DataFormat.STRING,
+        ],
+        "produces": DataFormat.STRING,
+    },
+}
+
+
+def _parse_modifier_chain(
+    expression: str,
+) -> Tuple[str, List[Tuple[str, str]], Optional[str]]:
+    """
+    Parses the expression part of an artifact_content embed.
+
+    Separates the artifact specifier, modifier chain, and final format specifier.
+
+    Args:
+        expression: The full expression string after 'artifact_content:'.
+
+    Returns:
+        A tuple containing:
+        - artifact_spec (str): The filename and optional version (e.g., "data.csv:1").
+        - modifiers (List[Tuple[str, str]]): A list of (prefix, value) tuples for modifiers.
+        - output_format (Optional[str]): The final output format string (e.g., "text", "json").
+                                          Returns None if the format step is missing or invalid.
+    """
+    from .constants import EMBED_CHAIN_DELIMITER
+
+    parts = expression.split(EMBED_CHAIN_DELIMITER)
+    if not parts:
+        return expression, [], None
+
+    artifact_spec = parts[0].strip()
+    modifiers = []
+    output_format = None
+
+    for i in range(1, len(parts)):
+        part = parts[i].strip()
+        if not part:
+            continue
+
+        if i == len(parts) - 1:
+            format_match = re.match(r"format:(.*)", part, re.DOTALL)
+            if format_match:
+                output_format = format_match.group(1).strip()
+                continue
+
+        modifier_parts = part.split(":", 1)
+        if len(modifier_parts) == 2:
+            prefix = modifier_parts[0].strip()
+            value = modifier_parts[1].strip()
+            if prefix and value:
+                modifiers.append((prefix, value))
+            else:
+                log.warning("Ignoring invalid modifier step format: '%s'", part)
+        else:
+            log.warning("Ignoring invalid modifier step format: '%s'", part)
+
+    return artifact_spec, modifiers, output_format