feat: rework tool calling#753
Open
maksymbuleshnyi wants to merge 18 commits into
Open
Conversation
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
Coverage Report •
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
maksymbuleshnyi
added
the
run-integration-tests-with-creds
tyaroshko
added
run-integration-tests-with-creds
maksymbuleshnyi
added
run-integration-tests-with-creds
maksymbuleshnyi
removed
the
run-integration-tests-with-creds
maksymbuleshnyi
added
the
run-integration-tests-with-creds
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
There are 2 total unresolved issues (including 1 from previous review).
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit 7c7a0aa. Configure here.
maksymbuleshnyi
added
run-integration-tests-with-creds
olbychos
reviewed
Jun 1, 2026
Contributor
olbychos
left a comment
olbychos
left a comment
There was a problem hiding this comment.
Overall looks solid. Let’s consider better patterns and rely on tests plus Anthropic/Bedrock and other provider checks.
| self._strip_nulls_for_fields(tool.input_schema.model_fields, action_input) | ||
| return action_input | ||
|
|
||
| def _strip_nulls_for_fields(self, fields: Mapping[str, Any], data: Any) -> None: |
Contributor
There was a problem hiding this comment.
let clean/rework 'matreshka' code
|
|
||
| try: | ||
| native_parallel = self.parallel_tool_calls_enabled and self.inference_mode == InferenceMode.FUNCTION_CALLING | ||
| # In FUNCTION_CALLING mode with tools present, force a tool call so |
Contributor
Author
There was a problem hiding this comment.
because this is purpose of function calling inference mode for agent. Otherwise it may not call a tool, and there is no point of such iteration then.
|
|
||
| # JSON Schema keywords Anthropic's strict-mode compiler rejects. Strip them | ||
| # before sending; keep semantics in the description. | ||
| _ANTHROPIC_STRICT_UNSUPPORTED_KEYWORDS = frozenset( |
Contributor
There was a problem hiding this comment.
Is there official source ? looks bad
| default=None, | ||
| description="Configuration for fallback behavior including the fallback LLM.", | ||
| ) | ||
| strict_tools: bool | list[str] = Field( |
Contributor
There was a problem hiding this comment.
We have bias towards openai and anthropic/ then how about bedrock and anthropic models? gemini models? grok, etc?
| in_string = False | ||
| escape = False | ||
|
|
||
| for ch in raw: |
Contributor
There was a problem hiding this comment.
Same 'matreshka' processing; maybe reconsider utils
or reuse
from pydantic_core import from_json
def parse_partial_json(raw: str):
return from_json(raw, allow_partial=True)
or
from partial_json_parser import ensure_json
def repair_truncated_json(raw: str) -> str:
return ensure_json(raw)
| must recurse into the sub-model and parse it back -- otherwise the string survives | ||
| and the nested model's Pydantic validation rejects it. | ||
| """ | ||
| from typing import Any, ClassVar, Literal |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Note
High Risk
Changes core agent LLM invocation, tool schema shaping, and parsing across OpenAI/Anthropic/LiteLLM; regressions could break tool calls, streaming, or provider compatibility.
Overview
This PR reworks function-calling and strict tool schemas end-to-end: generation is simplified (no global
stricton agent schemas), while OpenAI and Anthropic LLM nodes optionally apply provider-specific strict transforms via newstrict_toolsonBaseLLM, invoked from completion param building.Agent parsing now normalizes model output before tool validation:
_coerce_json_fields(including nested models) turns JSON-string dict fields back into objects;_strip_protocol_nullsdropsnullfor non-nullable fields so OpenAI strict “use default” semantics match Pydantic. Truncated tool arguments userepair_truncated_json. In function-calling mode with tools,tool_choice: requiredis set when the caller did not override it.Anthropic adds strict schema cleaning, a 20-tool cap, and a LiteLLM monkey-patch to forward
strict. Streaming emit helpers return completion booleans; integration tests guard answer text vsaction_inputwrappers and compare streamed tool input with null keys pruned. XML ReAct instructions tighten single-turn behavior. Tests cover nested dict coercion and richer tool schemas.Reviewed by Cursor Bugbot for commit 02f9b48. Bugbot is set up for automated code reviews on this repo. Configure here.