fix(python): fallback to 'GeneratedModel' when schema has no title

[matingathani]

Summary

Fixes a crash in json_schema_to_model when a JSON schema has no title field.

Root cause: json_schema_to_model calls create_model(model_name, ...) where model_name = json_schema.get("title"). When title is absent, model_name is None, and Pydantic raises:

TypeError: type.__new__() argument 1 must be str, not None

This hits in practice when LangGraph/LangChain/CrewAI providers wrap tools whose parameter schemas contain anonymous nested objects — for example, array-item schemas that have no explicit title key.

Fix: Fall back to "GeneratedModel" when title is absent, matching the convention already used in schema_converter.py for the same purpose.

Stack trace from the issue:

File ".../composio/utils/shared.py", line 194
    return create_model(model_name, **field_definitions)
                        ^
               model_name is None (should be a string)

Test plan

• New test: json_schema_to_model with no title creates a valid model (no crash)
• New test: required fields in a title-less schema remain required
• Full schema parser test suite passes (pytest tests/test_schema_parser.py — 67 tests)
• Full unit test suite passes (pytest tests/ — 592 tests)
• Linter passes (ruff check)

Closes #2435

[vercel]

@matingathani is attempting to deploy a commit to the Composio Team on Vercel.

A member of the Team first needs to authorize it.

[Copilot]

Pull request overview

Fixes a crash in the Python schema-to-Pydantic conversion path when a JSON schema object lacks a title, which previously caused pydantic.create_model(None, ...) to raise a TypeError. This improves compatibility with providers (LangGraph/LangChain/CrewAI) that encounter anonymous nested object schemas.

Changes:

• Add a fallback model name ("GeneratedModel") in json_schema_to_model when title is missing.
• Add tests covering title-less schemas (no-crash + required-field behavior).

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
python/composio/utils/shared.py	Prevents create_model from receiving None by adding a fallback model name.
python/tests/test_schema_parser.py	Adds regression tests for title-less object schemas and required-field enforcement.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[jkomyno]

Hi @matingathani — really good catch on this one. I built and locally verified a refinement that I think is worth folding in before merge.

Empirical finding: pydantic.create_model crashes only on None, not on "". I confirmed this by running:

from pydantic import create_model
create_model(None, foo=(str, ...))   # TypeError (the bug)
create_model("",   foo=(str, ...))   # OK — produces a model named ''

So the original Copilot comment about preserving "" is technically right, but the current key-presence form leaves a real crash uncovered: {"title": None, ...} still hits TypeError: type.__new__() argument 1 must be str, not None. CrewAI / LangChain emit this shape in the wild when older Pydantic v1 paths serialize anonymous models with an explicit-None title.

Two suggestions below: one extends the prod fix to also cover None (preserving "" as Copilot requested); the other parametrizes the test to lock in both crash modes. Both verified locally against the full schema-parser suite (68 tests pass, ruff clean).

[jkomyno]

The current form correctly handles a missing title key, but {"title": None, ...} still slips through and crashes create_model. Two-line fix that covers both shapes while preserving empty-string titles unchanged (matching Copilot's earlier feedback):

Suggested change

	model_name = json_schema["title"] if "title" in json_schema else "GeneratedModel"
	model_name = json_schema.get("title")
	if model_name is None:
	model_name = "GeneratedModel"

[jkomyno]

If you take the shared.py suggestion above, this test can collapse into one parametrized case that pins both crash modes (missing key + explicit None). Sentinel ... distinguishes "key absent" from "key present with value None":

Suggested change

	def test_no_title_fallback(self):
	"""Issue #2435: anonymous schemas (no 'title') must not crash with TypeError.
	LangGraph/LangChain/CrewAI providers call json_schema_to_model for
	array-item schemas that have no 'title' key. Without a fallback,
	create_model(None, ...) raises TypeError: type.__new__() argument 1
	must be str, not None.
	"""
	json_schema = {
	"type": "object",
	"properties": {
	"name": {"type": "string"},
	"value": {"type": "integer"},
	},
	"required": ["name"],
	}
	model_class = json_schema_to_model(json_schema)
	assert model_class.__name__ == "GeneratedModel"
	instance = model_class(name="test", value=42)
	assert instance.name == "test"
	assert instance.value == 42
	@pytest.mark.parametrize(
	"title_value",
	[
	pytest.param(..., id="missing"),
	pytest.param(None, id="none"),
	],
	)
	def test_none_title_falls_back_to_generated_model(self, title_value):
	"""Issue #2435: schemas with missing or None title must not crash.
	LangGraph/LangChain/CrewAI providers emit array-item schemas without a
	usable title — the key is absent, or present with an explicit None
	value. Both forms used to crash create_model with
	TypeError: type.__new__() argument 1 must be str, not None.
	Empty-string titles are preserved (they don't crash create_model).
	"""
	json_schema = {
	"type": "object",
	"properties": {
	"name": {"type": "string"},
	"value": {"type": "integer"},
	},
	"required": ["name"],
	}
	if title_value is not ...:
	json_schema["title"] = title_value
	model_class = json_schema_to_model(json_schema)
	assert model_class.__name__ == "GeneratedModel"
	instance = model_class(name="test", value=42)
	assert instance.name == "test"
	assert instance.value == 42