[WIP]Data schema generation

[FanG-817]

This PR currently works. I continue improving the validation part.

Some good suggestions from @amad-person that might help to provide more error information for self-correction:

• If a user is creating the python object e.g. Column, a ValueError is raised with the specific error
• If a user is creating a JSON object, the same error is caught and passed up as a StructureError, which only shows where the JSON has an error, and not why
• The StructureError is what I think Claude is seeing. Making it see the traceback (which will have the ValueError) might be helpful for self-correction

[FanG-817]

I have updated extracting and parsing the structure error:

rockfish-mcp/src/rockfish_mcp/sdk_client.py

Lines 689 to 788 in 34c74ae

	def extract_structure_error_details(exc: Exception) -> Dict[str, Any]:
	"""
	Extract error message and location from StructureError exception chain.
	Simplified version that extracts only essential information:
	- error_message: The actual error text from the exception
	- location: The JSON path where the error occurred
	Args:
	exc: The exception to parse (typically StructureError)
	Returns:
	{
	"error_count": N,
	"summary": "Found N validation error(s)",
	"errors": [
	{
	"error_message": "spike_magnitude (5.0) must be in [0, 1]",
	"location": "$.entities[0].columns[1]"
	},
	...
	]
	}
	Example:
	>>> try:
	... rf.converter.structure(ts_dict, TimeseriesParams)
	... except Exception as e:
	... details = extract_structure_error_details(e)
	... for err in details['errors']:
	... print(f"{err['location']}: {err['error_message']}")
	"""
	errors = []
	def collect_errors(current_exc, location="$"):
	"""Recursively collect error messages and locations."""
	# Check for sub-exceptions (IterableValidationError)
	if hasattr(current_exc, "exceptions") and current_exc.exceptions:
	# Extract location info from wrapper message
	msg = str(current_exc)
	index_match = re.search(r"@ index (\d+)", msg)
	if index_match:
	index = index_match.group(1)
	if "list[Column]" in msg:
	new_location = f"{location}.columns[{index}]"
	elif "list[Entity]" in msg:
	new_location = f"{location}.entities[{index}]"
	else:
	new_location = f"{location}[{index}]"
	else:
	new_location = location
	for sub_exc in current_exc.exceptions:
	collect_errors(sub_exc, new_location)
	# Check for __cause__ chain
	elif hasattr(current_exc, "__cause__") and current_exc.__cause__:
	# Extract location from StructureError if present
	if type(current_exc).__name__ == "StructureError":
	msg = str(current_exc)
	loc_match = re.search(r"@ (\$\.[^\s]+)", msg)
	if loc_match:
	location = loc_match.group(1)
	collect_errors(current_exc.__cause__, location)
	# Leaf error - add to list
	else:
	# Skip wrapper exceptions
	if type(current_exc).__name__ not in (
	"StructureError",
	"ClassValidationError",
	"IterableValidationError",
	"ExceptionGroup",
	):
	errors.append({"error_message": str(current_exc), "location": location})
	# Extract top-level location from StructureError
	top_location = "$"
	if type(exc).__name__ == "StructureError":
	top_msg = str(exc)
	loc_match = re.search(r"@ (\$\.[^\s]+)", top_msg)
	if loc_match:
	top_location = loc_match.group(1)
	# Collect all errors
	collect_errors(exc, top_location)
	# If no errors found, use original exception
	if not errors:
	errors = [{"error_message": str(exc), "location": top_location}]
	# Return consistent format
	error_word = "error" if len(errors) == 1 else "errors"
	return {
	"error_count": len(errors),
	"summary": f"Found {len(errors)} validation {error_word}",
	"errors": errors,
	}

It helps a lot for self-correction by understanding the exact error.
See screenshot:

[FanG-817]

Next steps:

Right now, I have another script validators.py to validate additional rules beside using rf.converter.structure(). I need to check whether we already have those checks in cuttlefish & rockfish-sdk and whether we need to add these into rockfish modules directly.

Additional validation tickets are created in rockfish repo. We have been adding missing validations so validators.py has been removed in this PR.